DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Receipt of Applicant’s Amendment filed October 15, 2025, is acknowledged.
Response to Amendment
Claims 1, 2, 4, 6, 10-12, 14, 16-18, and 20-25 have been amended. Claims 3, 5, 7, 13, and 19 have been canceled. Claims 26-30 are new. Claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 are pending and are provided to be examined upon their merits.
Response to Arguments
Applicant’s arguments with respect to claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. A response is provided below in bold where appropriate.
37 CFR 1.105 Information Request (7/15/2025 in Non-Final Rejection)
Applicant is requested to respond to the 1.105 information request and the Examiner thanks Applicant in advance.
Applicant argues 35 USC §101 Rejection, starting pg. 12 of Remarks:
III. REJECTIONS UNDER 35 U.S.C. § 101
Claims 1-25 were rejected under 35 U.S.C. § 101 because the claimed invention is allegedly directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
IV. RESPONSE TO REJECTIONS UNDER 35 U.S.C. § 101
Applicant respectfully traverses the rejection and asserts that the invention provides an improvement in computer-related technology as well as an improvement to at least the field of disease progression modeling by modeling disease progression for different patient groups separately with different inherent genetic compositions including a personalized framework for unsupervised disease progression models, in independent claims 1, 10, and 16, and supervised disease progression models, in independent claims 21 and 24. As stated in paragraph [0024] of the specification, the invention improves upon current approaches to disease progression pathways which are often not personalized by modeling "the disease progression for unique patient groups separately according to different inherent characteristics defined by genetic compositions which require a personalized framework for both supervised and unsupervised disease progression models." The specification, specifically paragraphs [0024] - [0034] detail numerous improvements to the field of disease progression modeling reflected in the claim language.
Modeling disease progression as claimed is abstract. A person with pen and paper can build a disease progression framework using patient data, determine a disease progression pattern for patient groupings and compare actual disease progression to a personalized disease progression trajectory. This is also abstract as managing relationships between people by teaching (compare actual disease progression pattern to a personalized disease progression trajectory).
The recent August 4th Memo regarding Subject Matter Eligibility Guidance is particularly instructive when reviewing Applicant's claims under the Improvements Consideration at MPEP 2106.05(a). In the Memo it states that the "Examiner is reminded to consult the specification to determine whether the disclosed invention improves technology or a technical field, and evaluate the claim to ensure it reflects the disclosed improvement." Additionally, the Memo goes on to state that the "claim itself does not need to explicitly recite the improvement described in the specification." Accordingly, while Applicant believes the previous claim set reflected the disclosed improvements, in the interest of expediting prosecution, Applicant has amended independent claims 1, 10, 16, 21, and 24 to include language further detailing comparing of an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received. Applicant has further amended dependent claims 2, 4, 6, 11, 12, 14,17, 18, 20, 22, 23, and 25 and added new claims 26-30 to ensure the claims reflect the disclosed improvements described in the specification, specifically between paragraphs [0024] - [0034]. More specifically, Applicant has amended independent claim 1 to include language reading:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; [[and]]
determining a disease progression pattern for each of the one or more patient groupings using by a personalized disease progression model using a neural network based state-space model, wherein the personalized disease progression model utilizes unsupervised computational approaches in leveraging the personalized disease progression framework; and [[.]]
comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
Additionally, Applicant has amended independent claim 21 to include language reading:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; [[and]]
determining a personalized disease progression trajectory for each of the one or more patient groupings using a personalized disease progression model; and [[.]]
comparing an actual disease progression pattern to the personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
The above cited memo was about Office Example 39, which did not recite a judicial exception. Applicant’s claims recite steps that are abstract.
Accordingly, the claim language now directly reflects the improvements to personalized disease progression modeling under MPEP 2106.05(a), and the claims are eligible in Step 2A Prong Two because the claims reflect an improvement to the functioning of a computer or to another technology or technical field, integrating a recited judicial exception into a practical application of the exception. Furthermore, the 2019 Revised Patent Subject Matter Eligibility Guidance provides that an invention is not abstract if it is determined that additional elements are recited in the claim beyond an alleged judicial exception which integrate the exception into a practical application of the exception. This requires an element in the claim to apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception.
Based on this recent guidance, Applicant's claims, which are implemented on a computer, are necessarily not abstract. Specifically, Applicant's invention recites a computer-implemented method, system, and program product for disease progression modeling. Applicant asserts that these claims do not seek to cover a judicial exception to patent eligibility.
Respectfully, claim 21 does not recite a computer. In any event, using a computer and neural network is not improving such technology but using existing technology.
Therefore, for at least the above reasons, Applicant respectfully requests that the rejection under 35 U.S.C. § 101 be reconsidered and withdrawn.
The rejection is respectfully maintained but modified for the claim amendments.
Applicant argues 35 USC §103 Rejection, starting pg. 14 of Remarks:
V. REJECTIONS UNDER 35 U.S.C. § 103
Claims 1-4, 7, 10-12, 16-18, and 21-25 were rejected under 35 U.S.C. § 103 as being allegedly unpatentable over "Impact of Clinical and Genomic Factors on COVID-19 Disease Severity" (hereinafter, "Dey") in view of U.S. Pub./Pat. No. 2018/0233233 (hereinafter, "Sharma").
Claims 5, 6, 8, 9, 13-15, 19, and 20 were rejected under 35 U.S.C. § 103 as being allegedly unpatentable over Dey, Sharma, and further in view of U.S. Pub./Pat. No. 2021/0057107 (hereinafter, "Solomon").
VI. RESPONSE TO REJECTIONS UNDER 35 U.S.C. § 103
In view of the interview between Applicant and the Examiner, Applicant has accordingly amended independent claims 1, 10, 16, 21, and 24 to include language detailing comparing of an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received. More specifically, Applicant has amended independent claim 1 to include language reading:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; [[and]]
determining a disease progression pattern for each of the one or more patient groupings using by a personalized disease progression model using a neural network based state-space model, wherein the personalized disease progression model utilizes unsupervised computational approaches in leveraging the personalized disease progression framework; and [[.]]
comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
Additionally, Applicant has amended independent claim 21 to include language reading:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; [[and]]
determining a personalized disease progression trajectory for each of the one or more patient groupings using a personalized disease progression model; and [[.]]
comparing an actual disease progression pattern to the personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
On page 13 of the Office Action, it states that Dey discloses "determining a disease progression pattern for each of the one or more patient groupings using personalized disease progression model, wherein the personalized disease progression model utilizes unsupervised computational approaches in leveraging the personalized disease progression framework." However, Dey fails to disclose the utilization of a neural network based state-space model by the personalized disease progression model. Furthermore, the Dey reference fails to disclose, suggest, or render predictable "comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received." While the Dey reference characterizes the factors that impact a severity of an illness, Dey is only concerned with COVID-19 as opposed to Applicant's invention which describes the utilization of different computational approaches (supervised or unsupervised) based on whether disease onset may be measured, see [0074] of Applicant's specification for onset agnostic diseases and see [0081] for diseases which progress through multiple stages. Accordingly, the Dey reference does not disclose the generic framework utilized by the personalized disease progression models for any chronic disease claimed by Applicant but instead Dey discloses a specific means for characterizing the factors and their impact on a specific illness.
Applicant has amended their claims requiring further prior art. As indicated in the primary cited art of Dey et al., machine learning is used. Machine learning encompasses neural networks, therefore it would be obvious to consider neural networks.
From Dey…
“Besides finding either clinical or genomic risk factors of severe COVID-19, several machine learning (ML) and artificial intelligence (AI) have also been applied successfully to solve a wide range of needs, ranging from fast ML based COVID-19 infection prediction 20 to routine blood-tests21 as fast alternatives to costly and time-consuming PCR tests22 to predicting disease state of individual cells given their transcriptomes23. However, to the best of our knowledge, we did not find any machine learning model built on combined clinico-genomic dataset to assess the COVID-19 severity.” (pg. 379, para. 5)
Therefore, Dey alone teaches using machine learning model.
Regarding "comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received” that is the result of what Dey is doing. The point of their framework/model is to use genetic and clinical data for predictive learning.
From Dey…
Clinical-genomic model (personalized disease progression trajectory) associated with (comparing) COVID-19 severity (actual disease progression pattern)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
Where clinic-genomic framework predicts (personalized disease progression trajectory) severe COVID-19 cases…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, last para.)
Thus, the Dey reference fails to disclose, suggest, or render predictable amended independent claim 1 and independent claim 21. The Sharma reference fails to cure the deficiencies of Dey. Therefore, for at least the above reasons, independent claim 1 and independent claim 21 are patentable over Dey and Sharma. For similar reasons, independent claims 10, 16, and 24 are also patentable over Dey and Sharma. Reconsideration and withdrawal of the rejection is, therefore, respectfully requested.
With all due respect, the Dey reference appears to be associated with Applicant’s inventors and is directly related to the filed claims. The Examiner respectfully requests Applicant comply with the 37 CFR 1.105 information request for Applicant to provide to the Office all relevant data related to this application.
VII. DEPENDENT CLAIMS
The other claims are dependent from one of the independent claims (i.e., claims 1, 10, 16, 20, and 24) discussed above, and are therefore believed patentable for at least the same reasons. Since each dependent claim is also deemed to define an additional aspect of the invention, however, the individual reconsideration of the patentability of each on its own merits is respectfully requested.
Similarly, because Applicant maintains that all claims are allowable for at least the reasons presented hereinabove, in the interests of brevity, this response does not comment on each and every comment made by the Examiner in the Office Action. This should not be taken as acquiescence of the substance of those comments, and Applicant reserves the right to address such comments.
The rejection has been modified for the claim amendments but respectfully maintained.
Claim Objections
Claim 30 is objected to because of the following informalities:
“… patient groupings are continuously updates as the additional…” where the word “updates” should probably be “updated.”
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 are directed to a method, system or product, which are statutory categories of invention. (Step 1: YES).
The Examiner has identified method claims 1 and 21 as the claims that represent the claimed invention for analysis and are similar to system claims 10 and 24 and product claim 16.
Claim 1 recites the limitations of:
A method for unsupervised disease progression modeling, the method comprising:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; and
determining a disease progression pattern for each of the one or more patient groupings by a personalized disease progression model using a neural network based state-space model, wherein the personalized disease progression model utilizes unsupervised computational approaches in leveraging the personalized disease progression framework; and
comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
These above limitations, under their broadest reasonable interpretation, cover performance of the limitation as mental processes. The claim recites elements, highlighted in bold above, which covers performance of the limitation that can be concepts performed in the mind of a person or with pen and paper. A person can receive patient data (write down pen and paper), build a personalized disease progression framework in their mind or with pen and paper derived from the patient data, determine disease progression pattern, and compare in their mind or with pen and paper actual disease progression pattern to a personalized disease progression trajectory. Building a personalized disease framework may be various things such as learning patient groupings using genetic data and may use state-space technique for learning personalized disease progression trajectory or pathway (para. [0058] of the specification). Further, the framework may be a generic framework for finding disease progression models for chronic disease (para. [0068]), therefore, a person with pen and paper can come up with various frameworks. Determining a disease progression pattern or trajectory could be performed with pen and paper or using a generic computer. A person can identify a current stage of a disease and a progression pattern for patient groupings (para. [0008] of the specification). See also MPEP 2106.04(a)(2) III C and using a generic computer for a mental process. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation as a mental process, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. Claims 10 and 16 are also abstract for similar reasons. (Step 2A-Prong 1: YES. The claims are abstract)
Also, building a disease progression framework using patient groupings derived from patient data and determining a disease progression pattern or trajectory for the patient groupings is abstract under certain methods of organizing human activity as managing personal behavior and relationships between people. Giving the claim it’s broadest reasonable interpretation, determining and comparing a disease progression pattern is diagnosing (instructing or teaching) health status for patient groupings.
This judicial exception is not integrated into a practical application. In particular, the claims only recite: computer, processors, computer-readable memories, computer-readable tangible storage medium (Claim 10); non-transitory computer readable storage media (Claim 16). The computer hardware is recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer component. See Applicant’s specification para’s. [0035] and [0036] about implantation using general purpose computing devices and MPEP 2106.05(f) where applying a computer as a tool is not indicative of significantly more. The ”personalized disease progression framework,” “personalized disease progression model,” and “utilizes unsupervised computational approaches” are all recited and applied at a high level of generality.
The claim recites “using a neural network based state-space model, wherein the personalized disease progression model utilizes unsupervised computational approaches…” where using a neural network model is recited at a high level of generality. Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Therefore claims 1, 10, and 16 are directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered separately and as an ordered combination, they do not add significantly more (also known as an “inventive concept”) to the exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using a computer hardware amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Steps such as receiving are steps that are considered insignificant extra solution activity and mere instructions to apply the exception using general computer components (see MPEP 2106.05(d), II). Thus claims 1, 10, and 16 are not patent eligible. (Step 2B: NO. The claims do not provide significantly more)
Dependent claims 2, 4, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, and 26-28 further define the abstract idea that is present in their respective independent claims 1, 10 and, 16 thus correspond to Mental Processes and Certain Methods of Organizing Human Activity and hence are abstract for the reasons presented above. The dependent claims do not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. The dependent claims further limit abstract steps or are themselves abstract. The variational auto-encoder in claims 2, 11, 17, and 27 are recited at a high level of generality. The retraining in claims 6, 14, and 20 is recited at a high level of generality. Claims 4 and 12 recite a user interface, which is a generic interface recited at a high level of generality. Claims 6, 14, and 20 recite a disease progression model applied at a high level of generality. Therefore, the claims 2, 4, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, and 26-28 are directed to an abstract idea. Thus, the claims 1, 2, 4, 6, 8-12, 14-18, 20, and 26-28 are not patent-eligible.
Claim 21 recites the limitations of:
A method for supervised disease progression modeling, the method comprising:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data;
determining a personalized disease progression trajectory for each of the one or more patient groupings using a personalized disease progression model; and
comparing an actual disease progression pattern to the personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
These above limitations, under their broadest reasonable interpretation, cover performance of the limitation as mental processes. The claim recites elements, highlighted in bold above, which covers performance of the limitation that can be concepts performed in the mind of a person or with pen and paper. A person can receive patient data (write down pen and paper), build a personalized disease progression framework in their mind or with pen and paper derived from the patient data, determine in their mind with pen and paper disease progression trajectory using a personalized disease model, and compare in their mind or with pen and paper actual disease progression pattern to a personalized disease progression trajectory. Building a personalized disease framework may be various things such as learning patient groupings using genetic data and may use state-space technique for learning personalized disease progression trajectory or pathway (para. [0058] of the specification). Further, the framework may be a generic framework for finding disease progression models for chronic disease (para. [0068]), therefore, a person with pen and paper can come up with various frameworks. Determining a disease progression pattern or trajectory could be performed with pen and paper or using a generic computer. A person can identify a current stage of a disease and a progression pattern for patient groupings (para. [0008] of the specification). See also MPEP 2106.04(a)(2) III C and using a generic computer for a mental process. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation as a mental process, then it falls within the “Mental Processes” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. Claim 24 is are also abstract for similar reasons. (Step 2A-Prong 1: YES. The claims are abstract)
Also, building a disease progression framework using patient groupings derived from patient data and determining a disease progression pattern or trajectory for the patient groupings is abstract under certain methods of organizing human activity by managing personal behavior and relationships between people. Giving the claim it’s broadest reasonable interpretation, determining and comparing a disease progression pattern is diagnosing (instructing or teaching) health status for patient groupings.
This judicial exception is not integrated into a practical application. In particular, the claims only recite: computer, processors, computer-readable memories, computer-readable tangible storage medium (Claim 24). The computer hardware is recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer component. See Applicant’s specification para’s. [0035] and [0036] about implantation using general purpose computing devices and MPEP 2106.05(f) where applying a computer as a tool is not indicative of significantly more. The ”personalized disease progression framework,” “personalized disease progression model,” and “utilizes unsupervised computational approaches” are all recited and applied at a high level of generality. Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Therefore claims 21 and 24 are directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered separately and as an ordered combination, they do not add significantly more (also known as an “inventive concept”) to the exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional element of using a computer hardware amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Steps such as receiving are steps that are considered insignificant extra solution activity and mere instructions to apply the exception using general computer components (see MPEP 2106.05(d), II). Thus claims 21 and 24 are not patent eligible. (Step 2B: NO. The claims do not provide significantly more)
Dependent claims 22, 23, 25, 29, and 30 further define the abstract idea that is present in their respective independent claims 21 and 24 and thus correspond to Mental Processes and Certain Methods of Organizing Human Activity and hence are abstract for the reasons presented above. The dependent claims do not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. The dependent claims further limit abstract steps or are themselves abstract. Claims 22, 23, 25, 29, and 30 recite a disease progression model applied at a high level of generality. Claim 22 recites recurrent neural network model at a high level of generality. The training in claim 29 and retraining in claim 30 is recited at a high level of generality. Therefore, the claims 22, 23, 25, 29, and 30 are directed to an abstract idea. Thus, the claims 21-25, 29, and 30 are not patent-eligible.
Law of Nature
Claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a law of nature without significantly more.
Claims 1, 2, 4, 6, 8-12, 14-18, and 20-30 are directed to a method, system or product, which are statutory categories of invention. (Step 1: YES).
Claims 1 and 21 are directed to building a personalized disease progression framework derived from patient data, where the patient data includes genetic data. Disease progression from genetic data recites a law of nature as it describes a naturally occurring relationship between genetic data and disease progression (see para. [0002] of the specification, where disease progression depends on genetics). Independent claims 10, 16, and 24 also recite a law of nature for the reasons given above. (Step 2A-Prong 1: YES. The claims recite a law of nature)
For the reasons provided above (see the abstract analysis), there are no practical applications or significantly more claimed. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application) (Step 2B: NO. The claims do not provide significantly more)
Dependent claims 2, 4, 6, 8, 9, 11, 12, 14, 15, 17, 18, 20, 22, 23, 25, and 25-30 further define the law of nature that is present in their respective independent claims 1, 10, 16, 21, and 24 and hence are related to a law of nature for the reasons presented above. The dependent claims do not include any additional elements that integrate the law of nature into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination.
Claim Interpretation
Claim 16 recites “computer readable storage product” and the specification teaches “A computer readable storage medium may, as that term is used in the present disclosure, is not to be construed as storage in the form of one or more transitory signals per se…” Therefore, claim 16 storage product and storage media are interpreted as non-transitory computer readable media, as are it’s dependent claims 17, 18, and 20.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 6, 14, 20, and 30 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claim 6 recites “retraining the personalized disease progression model based on the comparing of the actual disease progression pattern to the personalized disease progression trajectory for the corresponding patient grouping…” where no teaching of retraining based on comparing actual disease progression to personalized disease progression can be found in the specification.
The specification teaches:
“The personalized progression module 150 may also monitor the accuracy of the output received from the personalized disease progression model as well as the effectiveness of the one or more recommendations derived from that output. For example, the personalized progression module 150 may continuously receive additional patient data, such as clinical observations recorded in additional practitioner visits. The additional patient data may be based on specific form documents generated by the personalized progression module 150 which include data entry options for specific continuous data which may further improve the personalized disease progression model. In this embodiment, the personalized progression module 150 may utilize the actual disease progression derived from the additional patient data in retraining and/or improving the personalized disease progression model to provide improved recommendations and more accurate output as additional patient data is received. The personalized progression module 150 may retrain and/or improve the personalized disease progression model by utilizing the actual disease progression derived in updating the personalized disease progression framework described at step 204. For example, the personalized disease progression module 150 may identify one or more sub groupings within the one or more patient groupings based on the additional patient data received, this may be utilized as feedback for the personalized disease progression framework such that patient groupings and their corresponding disease progression trajectories and/or pathways may be continuously learned and refined over time, enabling the personalized disease progression model to provide more detailed and accurate output according to more specific patient groupings which in turn may be utilized for more personalized early intervention protocols. The personalize progression module 150 may compare the actual disease progression derived from the additional patient data received to the personalized disease progression trajectory generated for the corresponding patient grouping. Furthermore, the personalized progression module 150 may revise the early intervention protocols for future recommendations.” [0086]
The above teaches retraining using actual disease progression from additional patient data, not retraining based on comparing the actual disease progression pattern to the personalized disease progression trajectory. For examination purposes, this is interpreted as retraining a model based on additional patient data. Claims 14, 20, and 30 are rejected for a similar problem as they also recite retraining based on comparing.
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 6, 14, 20, and 30 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 6 recites “retraining the personalized disease progression model based on the comparing of the actual disease progression pattern to the personalized disease progression trajectory for the corresponding patient grouping…” where it is indefinite as to retraining a model based on comparing data. For examination purposes, this is interpreted as retraining a model based on additional patient data. Claims 14, 20, and 30 are rejected for a similar problem.
Examiner Request
The Applicant is requested to indicate where in the specification there is support for amendments to claims should Applicant amend. The purpose of this is to reduce potential 35 U.S.C. §112(a) or §112 1st paragraph issues that can arise when claims are amended without support in the specification. The Examiner thanks the Applicant in advance.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1, 10, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Dey et al. (Dey et al., “Impact of Clinical and Genomic Factors on COVID-19 Disease Severity,” Feb 21, 2022, pp. 378-387) in view of Kramer et al. (Kramer et al., “Reconstructing Nonlinear Dynamical Systems from Mult-Modal Time Series,” 2022, 39th International Conference on Machine Learning, pp. 1-21).
Regarding claims 1, 10, and 16
(clam 1) A method for unsupervised disease progression modeling, the method comprising:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
Dey et al. teaches:
Extract (receiving) clinical and genomic features (data)…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, para. 3)
Example of computed (therefore, using a computer)…
“We interpret the log-odds of the PU-Adapted logistic regression model to denote the importance of a particular feature toward COVID-19 severity. Figure 2 shows the feature importance of top 30 statistically significant features associated with COVID-19 severity computed by the final combined clinico-genomic model using the PU learning ramework. The statistical significance of a feature is computed using bootstrapping with 100 runs and the confidence interval is represented as error bars in the figure which are beyond zero (the vertical line).” (pg. 383, para. 1) Inherent with computed and runs is a computer executing instructions.
Example of computed (therefore, using a computer)…
“We interpret the log-odds of the PU-Adapted logistic regression model to denote the importance of a particular feature toward COVID-19 severity. Figure 2 shows the feature importance of top 30 statistically significant features associated with COVID-19 severity computed by the final combined clinico-genomic model using the PU learning ramework. The statistical significance of a feature is computed using bootstrapping with 100 runs and the confidence interval is represented as error bars in the figure which are beyond zero (the vertical line).” (pg. 383, para. 1) Inherent with computed and runs is a computer executing instructions.
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; and
Build and use clinical and genomic features…
“To build our severity prediction models, we used several clinical and genomic features that may have an impact on COVID-19 severity. For the clinical features, we manually curated some of the most relevant features which may have association with COVID-19 for further analysis. Our clinical features include demographics, lifestyles, comorbidities based on patient’s prior history, self-reports, and their chronic disease histories curated from prior inpatient records.” (pg. 380, para. 6)
Used (building) predictive learning (personalized disease progression) framework (also called Clinico-Genomic Framework) for COVID-19 (disease progression)…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, para. 3)
Example of age, ethnicity, smoking as patient groupings…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
determining a disease progression pattern for each of the one or more patient groupings by a personalized disease progression model using a neural network based state-space model, wherein the personalized disease progression model utilizes unsupervised computational approaches in leveraging the personalized disease progression framework; and
Disease with symptoms and progression (progression pattern)…
“The disease exhibits a wide range of symptoms ranging from asymptomatic to mild to rapid progression to critical stage2. In general, patients with mild symptoms recover easily whereas severe COVID-19 patients require further treatments such as ventilation in ICU. Such heterogeneity in terms of symptoms poses great challenges for understanding the disease etiology and designing treatment protocols. Characterizing the factors that impact the severity of this illness has been an open research question that we tackle in this work. Typically, severe effects of infectious diseases like COVID-19 are hypothesized to be associated with the variations of host genome3. GWAS (Genome wide association studies) have been reported in literature to understand the impact of host genomics on the disease4, 5. In addition to viral load environmental factors such as demographics, socio-economic status, lifestyle may also have an impact on the spread, exposure, and severity. The availability of rich Electronic Health Records (EHR) provides an opportunity for picking apart clinical factors associated with COVID-19 2, 6, 7.” (pg. 378, para. 3)
Clincio-genomic model and COVID-19 severity (disease progression) with patient age, ethnicity, prior conditions (patient groupings)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in8, and prior conditions like diabetes mellitus, hypertension have been reported in7, 22, 23. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies8 reported it with increased risk factors, while others22 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
Unlabeled (unsupervised) learning…
“We propose in this work a combined framework for finding the clinical and genomic factors associated with COVID-19 severity using a large COVID-19 dataset from the UK Biobank (UKBB). To address the above mentioned challenges, we first use the COVID-19 related hospitalization as a surrogate outcome for defining severe COVID-19 cases. Second, we use a machine learning technique called positive-unlabeled (PU) learning to address the noise and reporting bias present in COVID-19 severity labels. Moreover, we use a recently proposed genomic analysis framework entitled RubricOE9 to select the set of genomic factors after adjusting for the common prior comorbid conditions which may act as potential confounders. Finally, we aim to assess how the importance of the extracted biomarkers evolve over the pandemic marked by events such as the introduction of using Dexamethazone as treatment for COVID-19 severity and the emergence of the reportedly more contagious B.1.1.7 SARS-CoV-2 (alpha) strain.” (pg. 379, para .1)
Example of characterize clinical characteristics for COVID-19 (determining disease) and patterns…
“Simultaneously, there have been concerted efforts to characterize the clinical characteristics of COVID-19 including the patho-physiology and the risk factors for COVID-19. Several large-scale meta-analysis studies were conducted either on previous studies2 or by summarizing published articles6, 7 to identify common clinical symptoms and lab abnormalities among severe COVID-19 patients. Some common risk factors observed among critical/mortal COVID-19 patients are socio-demographic information such as male, older than 65, smoking, and prior history of hypertension, diabetes, cardiovascular disease, respiratory diseases, kidney diseases, obesity, immunosuppresion, and cancer. In addition to these prior conditions, depression as well as other cognitive and neurological disorders were also found as risk factors by a nationwide cohort study in Israel 14, although this study found that smoking and presence of respiratory diseases do not significantly increase the risk of complications. These factors were also found in a separate study 15 performed on 40% of all patients in England. In addition, socio-economic deprivation, diabetes, severe asthma, Black and South Asian ethnicity were observed as high risk factors. Research has also been conducted to characterize risk among sub-populations such as obesity, heart failure, and Parkinson’s Disease16–18 concurrently. Mathew et al.19 found three immunotypes revealing different patterns of lymphocyte responses among hospitalized COVID-19 patients.” (pg. 379, para. 4)
Pattern of number of medication taken per day with high severity risk factors (disease progression)…
“In addition, our study found a few more features such as number of medication taken per day to be significant. Such features are related to the overall health status of the patient and correlate indirectly to high severity risk factors. Interestingly, distance from home to workplace has also been reported with lower risk of severity, which may be confounded by people commuting by car for longer distance instead of public transport or cycling for shorter commutes.” (pg. 383, para. 3)
Leveraging unlabeled (unsupervised) machine learning algorithms with framework…
“To date, there have been 180 million confirmed cases of COVID-19, with more than 3.8 million deaths, reported to WHO worldwide. In this paper we address the problem of understanding the host genome’s influence, in concert with clinical variables, on the severity of COVID-19 manifestation in the patient. Leveraging positive unlabeled machine learning algorithms coupled with RubricOE, a state-of-the-art genomic analysis framework, on UK BioBank data we extract novel insights on the complex interplay. The algorithm is also sensitive enough to detect the changing influence of the emergent B.1.1.7 SARS-CoV-2 (alpha) variant on disease severity, and, changing treatment protocols. The genomic component also implicates biological pathways that can help in understanding the disease etiology. Our work demonstrates that it is possible to build a robust and sensitive model despite significant bias, noise and incompleteness in both clinical and genomic data by a careful interleaving of clinical and genomic methodologies.” (Abstract)
Artificial intelligence (neural network) used successfully and machine learning with clinic-genomic dataset…
“Besides finding either clinical or genomic risk factors of severe COVID-19, several machine learning (ML) and artificial intelligence (AI) have also been applied successfully to solve a wide range of needs, ranging from fast ML based COVID-19 infection prediction 20 to routine blood-tests21 as fast alternatives to costly and time-consuming PCR tests22 to predicting disease state of individual cells given their transcriptomes23. However, to the best of our knowledge, we did not find any machine learning model built on combined clinico-genomic dataset to assess the COVID-19 severity.” (pg. 379, para. 5)
See Neural Network State and Unsupervised below.
comparing an actual disease progression pattern to a personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
Clinical-genomic model (personalized disease progression trajectory) associated with (comparing) COVID-19 severity (actual disease progression pattern)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
Where clinic-genomic framework predicts (personalized disease progression trajectory) severe COVID-19 cases…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, last para.)
Neural Network State and Unsupervised
Dey et al. teaches machine learning. They do not specifically teach neural network
Kramer et al. also in the business of machine learning :
Interest to use machine learning for dynamical systems in biology and medicine and in unsupervised way…
“Empirically observed time series in physics, biology, or medicine, are commonly generated by some underlying dynamical system (DS) which is the target of scientific interest. There is an increasing interest to harvest machine learning methods to reconstruct this latent DS in a data-driven, unsupervised way. In many areas of science it is common to sample time series observations from many data modalities simultaneously, e.g. electrophysiological and behavioral time series in a typical neuroscience experiment.” (Abstract, marked U-1 in Kramer)
State space model using RNN (recurrent neural network)…
“For our approach we build on a nonlinear state space model framework introduced previously in (Durstewitz, 2017b; Koppe et al., 2019). The latent process used for approximating the unknown DS f(y; t) is modeled by a Gaussian piecewise-linear (PL) RNN of the form…” (pg. 3, col. 2, para. 3, marked U-2 in Kramer)
“Furthermore, we tested cross-modal inference, i.e. whether the trained mmPLRNN would be able to predict class labels from BOLD signals alone on left-out test data. While here this mainly serves to examine whether cross-modal links have been built within the model’s latent space, it is also a relevant application case. Specifically, we ran a cross-validation protocol where each 20% segment of the time series was left out in turn for training, and unseen class labels were predicted on these left-out test sets (see Appx. B.3 for details). Fig. 3C summarizes the agreement between true and predicted behavioral class labels across all test sets from successful training runs (see Appx. B.3) in a confusion matrix. These results were on par with those produced by various classifiers trained directly on the same 20-dimensional BOLD signal vectors and associated class labels as used for the mmPLRNN (multi-class F1 scores for logistic regression: 0:47, linear discriminant analysis: 0:48, support vector machines (RBF kernel, C = 1): 0:47, mmPLRNN: 0:45). This confirms that the mm-PLRNN has learned the connections between the two data modalities within its latent space in an about optimal manner, i.e., without much loss of information as judged by this comparison.” (pg. 8, col. 1, para. 2, marked U-3 in Kramer)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the Dey the ability to use neural network with state space and unsupervised data as taught by Kramer et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Kramer et al. who teaches increasing interest harvesting machine learning methods for unsupervised use in medicine.
Claims 2, 6, 11, 14, 17, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (11) above in further view of Krittangawong et al. (Krittanawong et al, “Artificial Intelligence and Cardiovascular Genetics,” Feb. 14, 2022, Life, pp. 1-28).
Regarding claims 2, 11, and 17
(claim 2) The method of claim 1, wherein the genetic data includes genome-wide association study (GWAS) data, and wherein the one or more patient groupings are heterogeneous patient groups defined by genetics and determined by clustering the GWAS data of a plurality of patients using Variational Auto-Encoder (VAE) and the state-space model wherein the genetic data is utilized as input data.
{
From Applicant’s specification regarding VAE…
“The personalized progression module 150 may learn the genetic markups using a deep-learning algorithm based on a variational auto-encoder (V AE) as will be described in greater detail below and with respect to FIG. 4. The deep learning model may be comprised of a Variational Auto-Encoder (VAE) and a SSM wherein genetic factors may be utilized as input. The genetic factors may be comprised of data from the GWA data received by the personalized progression module 150 at step 202, which may include genetic factors such as Single Nucleotide Polymorphisms (SNPs). The genetic data which may be collected in GWA studies may measure the genetic variation of an individual (e.g., patient) at a single base position of the individual's Deoxyribonucleic acid (DNA)….” [0069]
}
Dey et al. teaches:
Example of GWAS on pruned on this (clustering) dataset……
“….We further pruned for linkage disequilibrium (LD), removing correlated SNPs with r2 > 0:2 with a variance inflation factor of 10 on a window size of 50 kb. We performed a Genome-wide Association Study (GWAS) on this pruned dataset for a sanity check of the QC protocol and observe significant loci associated with the severity of COVID-19 infection. GWAS was performed by the package SAIGE26, while correcting for population structure, sex and age. We computed the top 20 principal components (PCs) with TeraPCA27 as a proxy for population structure and used them as covariates along with the biomarkers.” (pg. 381, para. 1)
The combined references teach machine learning and genomics. They do not teach heterogeneous and autoencoder.
Krittanawong et al. also in the business of machine learning and genomics teaches:
Autoencoders as primary DL (deep learning) used in genomics…
“Using these techniques, DL has been successfully applied within genomics in several major projects, including DeepSEA (a DL-based sequence analyzer that can predict the epigenetic state in multiple cell types), and a subsequent DragoNN primer online training in academic institutions globally [27,85]. To date, convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and GANs have been the primary DL techniques used in genomics (Table 2). These approaches have been implemented for several tasks, including functional assessments of variants [28], AI-guided multiethnic polygenic risk score (PRS) generation [86]. and variant calling optimization [87]. Interestingly, the number of layers within DL architectures used in genomics has generally been far less than those used for image recognition, and, thus far, typically consist of only a few layers [27,79,88] with many hundreds to thousands of parameters [89].” (pg. 12, para. 4, marked V-1 in Krittanawong)
Table 2 teaches…
“Variational Autoencoders [55] Autoencoder Predicts drug response” (pg. 10, marked V-2 in Krittanawong)
GWAS studies…
“Since it is origins in the 1940s, ML has used algorithmic and statistical techniques to process data for a variety of purposes and applications [64,65]. ML concepts, such as supervised machine learning (e.g., support vector machines to distinguish between cases and controls) and unsupervised machine learning (e.g., a variety of models to reduce highly dimensional data into lower dimensional space), are common tools in genomewide association studies (GWAS). In contrast to these types of ML, DL is a time- and resource-intensive subtype of ML that can achieve higher performance via its ability to learn complex representations from the data, depending on the task. Recently, advancements in computational power have enabled the application of DL onto large data sets (i.e., “big data”) to build extremely expressive and complex multi-layer artificial neural networks (ANN) [66]. The initial success of DL began in image processing and recognition, where it can be used to recognize objects without explicitly defining the relevant features. For example, instead of trying to identify the specific contours of the nose, eyes, or mouth, the DL algorithm categorizes an object as a “face”, which is recognized through a more abstract representation automatically learned from prior training on a dataset. In CVD, DL has been applied to non-imaging data, improving the accuracy of patient risk stratification and relationship prediction in comparison to traditional models, such as the Framingham Risk Score; although, typically, DL outperforms other models only on non-tabular data where there are complex nonlinear features that can be learned in a highly connected model [67,68].” (pg. 11, para. 1, marked V-3 in Krittanawong)
Genetic data with multiethnic polygenic (heterogeneous patient groups) PRS…
“Using these techniques, DL has been successfully applied within genomics in several major projects, including DeepSEA (a DL-based sequence analyzer that can predict the epigenetic state in multiple cell types), and a subsequent DragoNN primer online training in academic institutions globally [27,85]. To date, convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and GANs have been the primary DL techniques used in genomics (Table 2). These approaches have been implemented for several tasks, including functional assessments of variants [28], AI-guided multiethnic polygenic risk score (PRS) generation [86]. and variant calling optimization [87]. Interestingly, the number of layers within DL architectures used in genomics has generally been far less than those used for image recognition, and, thus far, typically consist of only a few layers [27,79,88] with many hundreds to thousands of parameters [89].” (pg. 12, para. 4, marked V-1 in Krittanawong)
Multi-ethnicity (heterogeneous patient groups) PRS…
“A recent study showed that polygenic risk powerfully modifies the risk conferred by monogenic risk variants [158]. However, incorporating these loci into clinical practice is not well established and PRS has limitations in complex disease predictions because of its dependency on linear regression, a lack of phenotype differentiation [159], and a variation in the numbers of SNVs in PRS [160]. A recent quantitative experiment demonstrated some improvement in prediction accuracy using multi-ethnic PRS (mixing training data from Europeans, South Asians, and Africans) [86]. Zhao and Zou investigated PRS, both empirically and theoretically, and found that accuracy can vary dramatically depending on how sparse true genetic signals are [161]. Therefore, an important future research direction is to use AI to explore non-linear PRS relationships, handle interactive high-dimensional data, and randomize selection of SNVs and genetic signals. AI could also be applied to multi-ethnic cohorts to elucidate the role that PRS and ML models, such as GANs, and could potentially play a role in creating a multi-ethnicity PRS. Despite the challenges, some steps have been taken to increase the diversity of WES and WGS samples with efforts such as the Trans-Omics for Precision Medicine (TOPMed) Program, the Million Veteran Program, the Atherosclerosis Risk in Communities (ARIC) Study, the MultiEthnic Study of Atherosclerosis (MESA) [162], and the Multiethnic Variation in Recovery: Role of Gender on Outcomes of Young AMI Patients (VIRGO) Study [163]. Nevertheless, the “unknown unknowns” of modifier genes or polygenic influences in CVDs remains to be explored. Although ML models in the PRS field remain in a developmental phase and have not yet been clinically tested in cardiovascular genetics, AI is poised to overcome current challenges by integrating ethnicity into genomic research.” (pg. 17, last para. – pg. 18, first para., marked V4 in Krittanawong)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use heterogeneous patient groups as taught by Krittanawong et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Krittanawong et al. who teaches the advantages of analyzing different groups for prediction and the combined references benefit as they analyze patient data for prediction purposes.
Regarding claims 6, 14, and 20
(claim 6) The method of claim 5, further comprising:
retraining the personalized disease progression model based on the comparing of the actual disease progression pattern to the personalized disease progression trajectory for the corresponding patient grouping to improve an accuracy of future personalized disease progression trajectories for new patient data.
Dey et al. teaches:
Data repository updated…
“Dataset. We analyzed the large prospective cohort of patients from the UK collected in the UKBB24 repository. The dataset contains diverse information including demographics, diagnosis, medications, lab tests, and genomic information of approximately half a million patients. In addition, the national SARS-CoV-2 laboratory test data were made available in UKBB through the Public Health England (PHE). This COVID-19 dataset contains a flag indicating specimen origin of COVID-19 tests: hospital inpatient origin vs other settings. It also notes the specimen collection date and the specimen result in addition to a few other information about how the specimen was collected. Data were available from Mar 16, 2020 and the repository has been updated once or twice every month. In our current study, we used data from Mar 16, 2020 to Jan 21, 2021.” (pg. 379, para. last – pg. 380, para 1)
Dataset evolved over time…
“Preparing the outcome and feature sets. Although the COVID-19 dataset in UKBB is a prospective study, it did not collect the disease severity explicitly. The nature of this dataset is very diverse depending on national testing strategy which evolved over time as the pandemic progressed. For example, in UK, testing was initially restricted to those with symptoms in hospital and thus under this assumption, using the positive tests of these subjects in hospital can be a reasonable proxy for severity. However, when the testings were expanded covering even asymptomatic patients from diverse facilities, defining the severity required analyzing information regarding the specimen sources of test.” (pg. 380, para. 2)
Nested test-training set and working data for further train-test splits (retraining data) to rank and iteratively finds stable set (improve accuracy)…
“2. Stable Feature Extraction: The learning rubric employs a nested test-training set configuration on the genomic data after QC and pruning for LD. The outer test-training set split reserves the test set (“validation”) for final SNP evaluation and denotes the training set as “working” data. Within the latter, further train-test splits are performed to rank the features by their Youden index28 (also known as J statistic). RubricOE is applied on the genomic data, accounting for top 20 PCs, sex, year of birth and Diabetes Mellitus (DM) as covariates. It iteratively finds a “stable” set of SNPs which persistently remains highly ranked with their respective Youden index. Annotation, Gene Ontology (GO) and enriched pathway analysis of the “stable” set of SNPs produced by RubricOE has been evaluated using the R package clusterProfiler29 v3.12.” (pg. 381, para. 2)
Example of evaluate and accuracy…
“Model Evaluation. We evaluate the effectiveness of PU-learning algorithms with a comparison with baseline traditional classification algorithm using logistic regression with a L1 loss. Computing traditional metrics such as precision, recall, F1 score, and accuracy is challenging in the positive unlabeled scenario since the only information available relates to the positive label while no sample from the negative class is clearly provided. One straightforward solution to this problem is to assume that the unlabeled data are negative. However, this is not fully accurate. A modified F1 score (we call it as Fpu 1 score) has been proposed to address this issue34. This score is computed as r2 p(^y=1) , where r is the recall and p(^y = 1) is the prevalence of the predicted positive label. As shown in34, the Fpu 1 score can take values greater than one but shares the same property as F1 score; it is high when both precision and recall are high and small when either one is small. We also report metrics used to evaluate traditional classification methods for completeness although these are not recommended for our setting. They include accuracy, precision, recall, and F1 score.” (pg. 382, para. 3)
Clinical-genomic model (personalized disease progression trajectory) associated with (comparing) COVID-19 severity (actual disease progression pattern)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
The combined references teach machine learning. They do not teach retrain.
Krittanawong et al. also in the business of machine learning teaches:
Collecting enough (additonal) training data…
“The next step is to train the network. During training, the model parameters are learned by the network from the training data provided relative to the labeled examples using backpropagation and other related gradient descent learning techniques. The major
challenge of this task is collecting enough training data and optimizing the hyperparameters (e.g., initialization strategy, learning rate, regularization techniques) so that the network can learn a robust set of parameters for the given prediction. It may also be necessary to reconsider the overall network architecture if the performance remains low. Importantly, given enough training data, sufficient computational resources, and an appropriate network architecture, nearly any mathematical function can be learned, including highly abstract functions from genomics data or image data to a disease state.” (pg. 11, last para. – pg. 12, top para., marked V-5 in Krittanawong)
Example of feedback (additional data) and iteratively train (therefore, retrain)…
“Once training is complete, the creation and analysis of artificial nucleotide sequences, such as the creation of artificial human genomes [82] or artificial enhancers (“synthetic DNA”), using approaches such as generative adversarial networks (GANs), can be considered [83]. GANs are DL models that include two primary components: a generator and a discriminator. Generated DNA sequences are used as inputs for the discriminator to analyze if the model has generated a convincingly real biological sequence. This feedback is used to iteratively train the generator model to produce artificial sequences with increasingly realistic properties. For example, a recent study used a type of GAN (an auxiliary classifier generative adversarial network) to generate synthetic participants that closely resembled real participants from the SPRINT trial (Systolic Blood Pressure Trial) to facilitate exploratory analyses [84].” (pg. 12, para. 3, marked V-6 in Krittanawong)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to retrain as taught by Krittanawong et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Krittanawong et al. who teaches the advantages of using feedback to iteratively train models for exploratory analysis. This would improve model accuracy over time.
Claims 4, 12, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (11) above in further view of Pub. No. US 2018/0233233 to Sharma et al.
Regarding claims 4, 12, and 18
(claim 4) The method of claim 1, further comprising:
displaying, in a user interface, the disease progression pattern for each of the one or more patient groupings using the personalized disease progression trajectory for each of the one or more patient groupings, wherein the personalized disease progression trajectory is continuously updated as the additional patient data is received.
[No Patentable Weight is given to non-functional descriptive claim language of “displaying… the disease stage progression pattern…”]
Dey et al. teaches:
Updating data every month (continuously)…
“The dataset contains diverse information including demographics, diagnosis, medications, lab tests, and genomic information of approximately half a million patients. In addition, the national SARS-CoV-2 laboratory test data were made available in UKBB through the Public Health England (PHE). This COVID-19 dataset contains a flag indicating specimen origin of COVID-19 tests: hospital inpatient origin vs other settings. It also notes the specimen collection date and the specimen result in addition to a few other information about how the specimen was collected. Data were available from Mar 16, 2020 and the repository has been updated once or twice every month. In our current study, we used data from Mar 16, 2020 to Jan 21, 2021.” (pg. 379 last para – pg. 380 first para.)
The combined references teach progression. They do not teach display.
Sharma et al. also in the business of progression teaches:
Example of disease progression…
“In another embodiment, the prediction of the disease progression for the patient based on the state variable can be performed using a trajectory clustering algorithm. In particular, a trajectory clustering algorithm can be used to group the database of disease trajectories into several clusters. A difference metric, such as the Euclidean norm, is used to compare the current state vector of the patient or an initial disease trajectory defined by the state vector values over an initial time period to the clusters in the database to and assign the current patient to one of the clusters by finding the most similar cluster. The patient's disease trajectory is then determined using the average disease trajectory for the assigned cluster in the database of disease trajectories. FIG. 2 illustrates disease progression modeling using trajectory clustering according to an embodiment of the present invention. As shown in FIG. 2, P.sub.1, P.sub.2, P.sub.3, . . . , P.sub.n are average disease trajectories for clusters 1-n in a database of disease trajectories. It can be noted that in the trivial case, the trajectories P.sub.1, P.sub.2, P.sub.3, . . . , P.sub.n may represent clusters of size 1, i.e., individual patients rather than a group of patients, but in an advantageous implementation each of these trajectories represents a cluster of multiple patients. Q represents the patient's disease trajectory, with 202 being the known initial disease trajectory between t.sub.0 and t.sub.p, and 204 being the predicted disease trajectory between t.sub.p and t.sub.f. The initial disease trajectory 202 of Q is compared to the initial portions of P.sub.1-P.sub.n, and it the most similar cluster to Q is determined to be P.sub.3. In order to estimate the predicted disease trajectory 204 of Q, the evolution of the state variables in Q between t.sub.p and t.sub.f match the evolution of the state variables in P.sub.3 between t.sub.p and t.sub.f. For the trajectory clustering, the whole state vector or some predefined subset of state variables in the state vector may be used to find the most similar cluster.” [0023]
Example of displaying disease trajectories…
“Returning to FIG. 1, at step 108, an optimal therapy recommendation and predicted disease trajectories are output. For example, the optimal therapy type and timing can be output by displaying a recommendation including the optimal therapy type and timing on a display device of a computer system, sending this information in a message (e.g., email), and/or storing the optimal therapy type and timing recommendation on a storage or memory of a computer system. In addition, various predicted disease trajectories for the patient can be output. For example, a predicted disease trajectory for a “do nothing” scenario and predicted disease trajectories for one or more therapy options including different types and timings of therapies can be displayed on a display device of a computer system or stored in a memory or storage of a computer system. These output results can be incorporated in a Clinical Decision Support (CDS) system. These output results can be integrated into an Electronic Medical Record (EMR) system. These output results can be incorporated in a radiologist's reading system at the time when a diagnosis is being mage to inform the radiologist of the potential outcomes the patient might have. These output results can be included in a report that is accessible to the patient so the patient can make informed decisions.” [0044]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to display progression trajectory as taught by Sharma et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Sharma et al. and the health benefits of displaying trajectories and recommendations as this helps to visually determine if progress is being made with respect to health.
Claims 8, 9, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (11) above in further view of Pub. No. US 2021/0057107 to Solomon et al.
Regarding claims 8 and 15
(claim 8) The method of claim 1, further comprising:
generating one or more recommendations for each of the one or more patient groupings based on the disease progression pattern corresponding to each patient grouping, wherein the one or more recommendations includes at least a set of early intervention protocols.
Dey et al. teaches:
Example of before and after with drug approved (recommendation)…
“Is the clinico-genomic model sensitive to treatment protocols? UKBB does not have direct information on COVID-19 drugs and detailed treatment protocols that a COVID-19 patient went through, so we assess the impact of a COVID-19 drug indirectly using its date of approval. In particular, we study the change in coefficients of the top features (Figure 2) before and after the drug approval date. Dexamethazone was approved for use in UK on Jun 15, 2020. We observe (see Figure 3) that most of the significant factors from the whole dataset still had similar impact after the drug being used, but only 5 features– age, current smokers, DM, gene MACROD2 and SNP rs10733122– had significant impact before the approval date. MACROD2 is known to be associated with ischemic stroke and large artery atherosclerosis36, both of which are correlated with COVID-19 severity37. Similarly, age, DM and smoking status are also shown to be affected by corticosteroids including Dexamethazone38–40.” (pg. 383, para. 4)
The combined references teach patterns. They do not literally teach recommendations.
Solomon et al. also in the business of patterns teaches:
Recommendations of treatments or tests…
“An embodiment of a recommender system for recommending a treatment or diagnostic test may comprise at least one processor configured to perform: selecting, from a database of genetic markers, a set of genetic markers corresponding to a disease or immune state; collecting a set of genetic marker values by parsing a patient electronic health record (EHR); generating one or more imputed genetic marker values associated with one or more of the collected genetic marker values; generating a set of genome-specific trained model predicted outcomes by: for each collected genetic marker value: selecting, from a set of genome-specific trained models, at least one genome-specific trained model corresponding to the collected genetic marker value; executing each of the selected the genome-specific trained model(s) to generate a genome-specific trained model predicted outcome, thereby generating a set of genome-specific trained model predicted outcomes; executing a second trained model to identify genome-specific trained model predicted disease state and/or outcomes; and recommending one or more of the identified treatments and/or tests on the basis of the predicted disease state and/or outcomes.” [0031]
“EHR extractor/encoder program (2250) receives, from collected data database (1100), historical patient EHRs, e.g., EHRs from a cohort of patients comprising a particular patient population of interest, and mines the patient EHRs datasets in order to generate datasets including DX, RX, outcomes, and genetic data associated with the EHRs data, which is stored in training dataset storage database (2275).” [0171]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to provide recommendations as taught by Solomon et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Solomon et al. and the health benefits providing recommendations to improve trajectories or patterns. The combined references further benefit as they are analyzing data for improving health outcomes.
Regarding claim 9
The method of claim 8, further comprising:
monitoring an effectiveness of the one or more recommendations including the set of early intervention protocols; and
Dey et al. teaches:
Model is sensitive to treatment protocols (monitoring effectiveness)…
“While we observe that the model is sensitive to treatment protocols in general, it should be noted that this analysis may be confounded due to small sample size or other environmental factors linked with the use of corticosteroids. Also, medication data were not directly available in the cohort we used - as such this analysis has a strong assumption that the treatment strategy was majorly adopted for severe COVID-19 patients after it was approved. We aim to explore the effect of medication and treatment procedure more systematically in future.” (pg. 383, last para. – pg. 384, para. 1)
revising the early intervention protocols for future recommendations.
Explore the effect (revising) the medication and treatment (protocols) in the future…
“While we observe that the model is sensitive to treatment protocols in general, it should be noted that this analysis may be confounded due to small sample size or other environmental factors linked with the use of corticosteroids. Also, medication data were not directly available in the cohort we used - as such this analysis has a strong assumption that the treatment strategy was majorly adopted for severe COVID-19 patients after it was approved. We aim to explore the effect of medication and treatment procedure more systematically in future.” (pg. 383, last para. – pg. 384, para. 1)
Claim 26 is rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (12) above in further view of Pub. No. US 2021/0082092 to Sargent et al.
Regarding claim 26
The method of claim 2, wherein the plurality of patients are selected by a user within a user interface based on a chronic disease and patient characteristics, and wherein the patient data is anonymized by applying perturbations and adversarial patches to visual patient data.
{
From Applicant’s specification on adversarial patches…
“The computing device may train the second generative adversarial network to generate a patch for the first image with the local artifact removed based on the second image of the patient. The computing device may train the second generative by receiving a first image of a patient captured at a first period of time containing a local artifact, receiving a second image of the patient captured at a second period of time containing no artifact, pairing the first image and the received image of the patient, and inputting the paired first image and second image of the patient into the second generative adversarial network. Additionally, the personalized progression module 150 may also anonymize the patient data, for example, the personalized progression module 150 may apply perturbations, adversarial patches, and/or other noise to visual patient data, such as, but not limited to, include medical images, medical videos, amongst other visual content, which may include, but is not limited to including Magnetic Resonance Imaging (MRI) scans, Computed Tomography (CT) scans, X-ray images, ultrasounds, Positron Emission Tomography (PET) scans, Arthrograms, Myelograms, amongst other visual patient data. The personalized progression module 150 may only receive data relevant to the specific disease being modeled.” [0053]
Therefore, patches to images.
}
Dey et al. teaches:
Curate (select) chronic disease…
“To build our severity prediction models, we used several clinical and genomic features that may have an impact on COVID-19 severity. For the clinical features, we manually curated some of the most relevant features which may have association with COVID-19 for further analysis. Our clinical features include demographics, lifestyles, comorbidities based on patient’s prior history, self-reports, and their chronic disease histories curated from prior inpatient records.” (pg. 380, para. 6)
Interface
The combined references teach select. They do not teach user interface.
Sargent et al. also in the business of select teaches:
“The user interface 122 includes components used to receive input from a user on the user device 120 and transmit the input to the artifact removal program 136 residing on the server 130, or conversely to receive information from the artifact removal program 136 and display the information to the user on user device 120. In an example embodiment, the user interface 122 uses a combination of technologies and devices, such as device drivers, to provide a platform to enable users of the user device 120 to interact with the artifact removal program 136. In an example embodiment, the user interface 122 receives input, such as but not limited to, textual, visual or audio input received from a physical input device, such as, but not limited to, a keypad and/or a microphone.” [0027]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to provide a user interface as taught by Sargent et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by the combined references that interact with computer systems. It would be obvious to provide a user interface to access data.
Patch
The combined references teach image. They do not teach adversarial patch.
Sargent et al. also in the business of image teaches:
Example of train (select) and generate a patch (patch is a perturbation) for a patient’s image…
“The computing device may train the second generative adversarial network to generate a patch for the first image with the local artifact removed based on the second image of the patient. The computing device may train the second generative by receiving a first image of a patient captured at a first period of time containing a local artifact, receiving a second image of the patient captured at a second period of time containing no artifact, pairing the first image and the received image of the patient, and inputting the paired first image and second image of the patient into the second generative adversarial network.” [0005]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use a patch as taught by Sargent et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Sargent et al. who teaches the benefits of using a patch for patient data. Further, this provides benefits of reducing health privacy concerns.
Claim 27 is rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (12) above in further view of Pub. No. US 2023/0326542 to Montserrat et al.
Regarding claim 27
The method of claim 2, wherein the clustering of the GWAS data of the plurality of patients further comprises:
identifying different genetic signatures by grouping the GWAS data to lower dimensions using the VAE based matrix factorization techniques;
See Matrix below.
projecting Single Nucleotide Polymorphisms (SNPs) to lower dimensions for interpretability; and
See Matrix below.
regularizing the input data to include existing domain knowledge, wherein the existing domain knowledge includes a plurality of disease progression pathways.
See Matrix below.
Matrix
The combined references teach GWAS. They do not teach lower dimension.
Montserrat et al. also in the business of GWAS teaches:
Genome-wide association studies (GWAS)…
“Various information can be deduced from the correlations between neighboring sites along the genome. For example, local-ancestry inference uses the pattern of variation observed at various sites along an individual's genome to estimate the ancestral origin of an individual's DNA. In addition, correlations along the genome can influence polygenic risk scores (PRS), genome-wide association studies (GWAS), and many other aspects of precision medicine. Given that the correlations between neighboring genetic variants are ancestry dependent, applying the results of these analyses to an individual's genome may require knowledge of the individual's ancestry at each site along the genome.” [0004]
Variational autoencoder and reduced number of dimensions based on transform the input (factorization technique)…
“In some examples, the generative machine learning model comprises an encoder and a decoder configured as a class-conditional variational autoencoder (CVAE). Both the encoder and the decoder can be implemented as neural network models. The encoder can transform the input vector in a variant segment space to a multi-dimensional probability distribution of embedding vectors in a latent space having a reduced number of dimensions, e.g., by mapping to a mean and width (variance) of the distribution for each of the reduced number of dimensions. For an isotropic distribution, the variance would be the same for each dimension. The distributions in the reduced space can represent variations of the input variant segment. The decoder can obtain samples of embedding vector from the probability distribution, which are then reconstructed to form the output vectors from the samples, the output vectors having the same dimension as the input vector and representing the simulated variant segments.” [0010]
SNP with reduced number of dimensions…
“In some examples, the generative machine learning model comprises a first sub-model and a second sub-model, both of which can be implemented as a neural network model. The first sub-model can include an encoder configured to map the input vector a multi-dimensional probability distribution of embedding vectors in a latent space. The latent space can have a reduced number of dimensions relative to the number of SNP sites represented in the input SNP segment. While reducing the number of dimensions, the mapping can still retain information indicative of the pattern of SNP variants of the input vector in the embedding vector. In a case where the probability distribution comprises a Gaussian distribution, the encoder can determine, based on the pattern of SNP variants in the input vector, a mean and a variance of a distribution for each dimension of the embedding vector. Different probability distributions (e.g., different Gaussian distributions having different means and variances for different dimensions of the latent space) can be determined for different SNP sequences. In some examples, the ancestral origin can be input to the encoder with the input vector generate the parameters of a distribution of embedding vectors for that ancestral origin. Multiple probability distributions can be generated by the encoder for different ancestral origins.” [0033]
Associate (regularizing) genetic variants (input data) with predisposition to illness (disease progression pathway)…
“Local-ancestry inference can be helpful for genome-wide association studies (GWAS). GWAS is a study of a genome-wide set of genetic variants in different individuals to see if any variant is associated with a trait, such as predisposition to certain illness, certain biomedical traits (e.g., blood pressure, cholesterol level, etc.). Accordingly, such studies can associate specific genetic variations with particular diseases. Knowing a predisposition of a particular ancestral origin for a particular disease can help identify whether certain variations are associated with the particular disease.” [0055]
“In addition to summation and activation function processing, each node of hidden layer 354 can also perform a batch normalization process to normalize the outputs of the hidden layer to, for example, increase the speed, performance, and stability of neural network model 350. The normalization process can include, for example, subtracting a mean of the outputs from each output of the hidden layer node, and dividing by the subtraction results by the standard deviation of the outputs, to generate a normalized output at each hidden layer node. In some examples, the normalization operation can be performed prior to applying the activation function. Based on the activation function processing and batch normalization processing, node 354a generates an intermediate output ie.sub.0, node 354b generates an intermediate output ie.sub.1, whereas node 354m generates an intermediate output ie.sub.m.” [0089]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to lower dimension as taught by Montserrat et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Montserrat et al. who teaches the benefits of using a reduced matrix for SNP data. A lower or reduced matrix would also allow for faster computation of data.
Claim 28 is rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (11) above in further view of Pub. No. US 2021/0090694 to Colley et al.
Regarding claim 28
The method of claim 1, wherein the unsupervised computational approaches includes a unsupervised model utilizing one or more maximum-likelihood detection algorithms to learn hidden stages and a most likely sequence of hidden states of a Hidden Markov Model (HMM).
The combined reverences teach unsupervised model. They do not teach Markov or maximum-likelihood detection algorithms.
Colley et al. also in the business of unsupervised models teaches:
Hidden Markov models…
“MLAs include supervised algorithms (such as algorithms where the features/classifications in the data set are annotated) using linear regression, logistic regression, decision trees, classification and regression trees, Naïve Bayes, nearest neighbor clustering; unsupervised algorithms (such as algorithms where no features/classification in the data set are annotated) using Apriori, means clustering, principal component analysis, random forest, adaptive boosting; and semi-supervised algorithms (such as algorithms where an incomplete number of features/classifications in the data set are annotated) using generative approach (such as a mixture of Gaussian distributions, mixture of multinomial distributions, hidden Markov models), low density separation, graph-based approaches (such as mincut, harmonic function, manifold regularization), heuristic approaches, or support vector machines.” [2590]
Expectation (likelihood) maximization algorithm….
“… These methods work by applying the observation that certain haplotypes are common in certain genomic regions. Therefore, given a set of possible haplotype resolutions, these methods choose those that use fewer different haplotypes overall. Methods for calculating this may be based on combinatorial approaches (parsimony) or likelihood functions based such as the Hardy-Weinberg principle, the coalescent theory model, or perfect phylogeny. The parameters in these models may be estimated using algorithms such as the expectation-maximization algorithm (EM), Markov chain Monte Carlo (MCMC), or hidden Markov models (HMM). Once the patient's haplotype is determined, overall tumor ploidy may be calculated.” [2714]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use Markov model as taught by Colley et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Colley et al. who teaches the benefits of using Markov models for unsupervised algorithms including calculating benefits such as parsimony.
Claims 21 and 24 are rejected under 35 U.S.C. 103 as being unpatentable over Dey et al. (Dey et al., “Impact of Clinical and Genomic Factors on COVID-19 Disease Severity,” Feb 21, 2022, pp. 378-387) in view of Krittangawong et al. (Krittanawong et al, “Artificial Intelligence and Cardiovascular Genetics,” Feb. 14, 2022, Life, pp. 1-28).
Regarding claims 21 and 24
(claim 21) A method for supervised disease progression modeling, the method comprising:
receiving patient data, wherein the patient data is comprised of at least genetic data and clinical observations;
Dey et al. teaches:
Extract (receiving) clinical and genomic features (data)…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, para. 3)
Example of computed (therefore, using a computer)…
“We interpret the log-odds of the PU-Adapted logistic regression model to denote the importance of a particular feature toward COVID-19 severity. Figure 2 shows the feature importance of top 30 statistically significant features associated with COVID-19 severity computed by the final combined clinico-genomic model using the PU learning ramework. The statistical significance of a feature is computed using bootstrapping with 100 runs and the confidence interval is represented as error bars in the figure which are beyond zero (the vertical line).” (pg. 383, para. 1) Inherent with computed and runs is a computer executing instructions.” (pg. 383, first para.)
building a personalized disease progression framework using one or more patient groupings and one or more disease progression pathways derived from the patient data; and
Build and use clinical and genomic featues…
“To build our severity prediction models, we used several clinical and genomic features that may have an impact on COVID-19 severity. For the clinical features, we manually curated some of the most relevant features which may have association with COVID-19 for further analysis. Our clinical features include demographics, lifestyles, comorbidities based on patient’s prior history, self-reports, and their chronic disease histories curated from prior inpatient records.” (pg. 380, para. 6)
Used (building) predictive learning (personalized disease progression) framework (also called Clinico-Genomic Framework) for COVID-19 (disease progression)…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, para. 3)
determining a personalized disease progression trajectory for each of the one or more patient groupings using a personalized disease progression model; and
Clincio-genomic model and COVID-19 severity (disease progression) with patient ethnicity, prior conditions (patient groupings)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in8, and prior conditions like diabetes mellitus, hypertension have been reported in7, 22, 23. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies8 reported it with increased risk factors, while others22 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
comparing an actual disease progression pattern to the personalized disease progression trajectory for a corresponding patient grouping as additional patient data is received.
Clinical-genomic model (personalized disease progression trajectory) associated with (comparing) COVID-19 severity (actual disease progression pattern)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
Where clinic-genomic framework predicts (personalized disease progression trajectory) severe COVID-19 cases…
“Clinico-Genomic Framework. We used a predictive learning framework called Positive-unlabeled (PU) learning for extracting the most significant clinical and genomic features associated with COVID-19 outcome. Note that our dataset is more robust in terms of defining the severe COVID-19 patients than the non-severe cases. Hence, using a classical binary classification algorithm will not be appropriate and able to cope with such noisy class labels. PU learning algorithms have alternatively been reported to be especially suitable for learning from a noisy negative class by treating instances in this class as unlabeled while predicting the positive class (severe COVID-19 cases).” (pg. 381, last para.)
Supervised
Dey teaches classification and machine learning. They do not literally teach supervised.
Krittanawong et al. also in the business of machine learning teaches:
Supervised machine learning is a common tool for genomewide association studies…
“Since it is origins in the 1940s, ML has used algorithmic and statistical techniques to process data for a variety of purposes and applications [64,65]. ML concepts, such as supervised machine learning (e.g., support vector machines to distinguish between cases and controls) and unsupervised machine learning (e.g., a variety of models to reduce highly dimensional data into lower dimensional space), are common tools in genomewide association studies (GWAS). In contrast to these types of ML, DL is a time- and resource-intensive subtype of ML that can achieve higher performance via its ability to learn complex representations from the data, depending on the task. Recently, advancements in computational power have enabled the application of DL onto large data sets (i.e., “big data”) to build extremely expressive and complex multi-layer artificial neural networks (ANN) [66]. The initial success of DL began in image processing and recognition, where it can be used to recognize objects without explicitly defining the relevant features. For example, instead of trying to identify the specific contours of the nose, eyes, or mouth, the DL algorithm categorizes an object as a “face”, which is recognized through a more abstract representation automatically learned from prior training on a dataset. In CVD, DL has been applied to non-imaging data, improving the accuracy of patient risk stratification and relationship prediction in comparison to traditional models, such as the Framingham Risk Score; although, typically, DL outperforms other models only on non-tabular data where there are complex nonlinear features that can be learned in a highly connected model [67,68].” (pg. 11, para. 1, marked V-3 in Krittanawong)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use supervised training as taught by Krittanawong et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Dey et al. who teaches using available classification data, and supervised training uses labeled data.
Claims 22 and 25 are rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (18) above in further view Kramer et al. (Kramer et al., “Reconstructing Nonlinear Dynamical Systems from Mult-Modal Time Series,” 2022, 39th International Conference on Machine Learning, pp. 1-21).
Regarding claims 22 and 25
(claim 22) The method of claim 21, wherein the personalized disease progression trajectory includes at least identifying a current stage from a plurality of disease stages and a progression pattern for each of the one or more patient groups, wherein the personalized disease progression trajectory is based on assigning a state at each of a plurality of time points by the personalized disease progression model using forward temporal model designed as a Recurrent Neural Network (RNN) to model a representation of the one or more patient groupings based on the patient data.
Dey et al. teaches:
Example of initially restricted (identifying) symptoms in hospital (current stage)…
“Although the COVID-19 dataset in UKBB is a prospective study, it did not collect the disease severity explicitly. The nature of this dataset is very diverse depending on national testing strategy which evolved over time as the pandemic progressed. For example, in UK, testing was initially restricted to those with symptoms in hospital and thus under this assumption, using the positive tests of these subjects in hospital can be a reasonable proxy for severity. However, when the testings were expanded covering even asymptomatic patients from diverse facilities, defining the severity required analyzing information regarding the specimen sources of test.” (pg. 380, para. 2)
Disease stages…
“The disease exhibits a wide range of symptoms ranging from asymptomatic to mild to rapid progression to critical stage2. In general, patients with mild symptoms recover easily whereas severe COVID-19 patients require further treatments such as ventilation in ICU. Such heterogeneity in terms of symptoms poses great challenges for understanding the disease etiology and designing treatment protocols. Characterizing the factors that impact the severity of this illness has been an open research question that we tackle in this work. Typically, severe effects of infectious diseases like COVID-19 are hypothesized to be associated with the variations of host genome3. GWAS (Genome wide association studies) have been reported in literature to understand the impact of host genomics on the disease4, 5. In addition to viral load environmental factors such as demographics, socio-economic status, lifestyle may also have an impact on the spread, exposure, and severity. The availability of rich Electronic Health Records (EHR) provides an opportunity for picking apart clinical factors associated with COVID-19 2, 6, 7.” (pg. 378, para. 3)
Finding (identifying) severity (disease stages)…
“We propose in this work a combined framework for finding the clinical and genomic factors associated with COVID-19 severity using a large COVID-19 dataset from the UK Biobank (UKBB). To address the above mentioned challenges, we first use the COVID-19 related hospitalization as a surrogate outcome for defining severe COVID-19 cases. Second, we use a machine learning technique called positive-unlabeled (PU) learning to address the noise and reporting bias present in COVID-19 severity labels. Moreover, we use a recently proposed genomic analysis framework entitled RubricOE9 to select the set of genomic factors after adjusting for the common prior comorbid conditions which may act as potential confounders. Finally, we aim to assess how the importance of the extracted biomarkers evolve over the pandemic marked by events such as the introduction of using Dexamethazone as treatment for COVID-19 severity and the emergence of the reportedly more contagious B.1.1.7 SARS-CoV-2 (alpha) strain.” (pg. 379, para. 1)
Recurrent Neural Network
Dey teaches machine learning. They do not teach Recurrent Neural Network.
Krittanawong et al. also in the business of machine learning teaches:
Recurrent neural networks (RNN) for genomics….
“Using these techniques, DL has been successfully applied within genomics in several major projects, including DeepSEA (a DL-based sequence analyzer that can predict the epigenetic state in multiple cell types), and a subsequent DragoNN primer online training in academic institutions globally [27,85]. To date, convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, and GANs have been the primary DL techniques used in genomics (Table 2). These approaches have been implemented for several tasks, including functional assessments of variants [28], AI-guided multiethnic polygenic risk score (PRS) generation [86]. and variant calling optimization [87]. Interestingly, the number of layers within DL architectures used in genomics has generally been far less than those used for image recognition, and, thus far, typically consist of only a few layers [27,79,88] with many hundreds to thousands of parameters [89].” (pg. 12, para. 4, marked V-1 in Krittanawong)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use recurrent neural network as taught by Krittanawong et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Dey et al. who teaches use of machine language including artificial intelligence in the genomic area. Also, Krittanawong et al. teaches about the advancements in computational power to build neural networks and the benefits of using neural networks for prognostic guidance.
Space State and Time
The combined references teach neural network. They do not teach space state and time points.
Kramer et al. also in the business of machine learning :
Interest to use machine learning for dynamical systems in biology and medicine and in unsupervised way…
“Empirically observed time series in physics, biology, or medicine, are commonly generated by some underlying dynamical system (DS) which is the target of scientific interest. There is an increasing interest to harvest machine learning methods to reconstruct this latent DS in a data-driven, unsupervised way. In many areas of science it is common to sample time series observations from many data modalities simultaneously, e.g. electrophysiological and behavioral time series in a typical neuroscience experiment.” (Abstract, marked U-1 in Kramer)
State space model using RNN (recurrent neural network)…
“For our approach we build on a nonlinear state space model framework introduced previously in (Durstewitz, 2017b; Koppe et al., 2019). The latent process used for approximating the unknown DS f(y; t) is modeled by a Gaussian piecewise-linear (PL) RNN of the form…” (pg. 3, col. 2, para. 3, marked U-3 in Kramer)
“Furthermore, we tested cross-modal inference, i.e. whether the trained mmPLRNN would be able to predict class labels from BOLD signals alone on left-out test data. While here this mainly serves to examine whether cross-modal links have been built within the model’s latent space, it is also a relevant application case. Specifically, we ran a cross-validation protocol where each 20% segment of the time series was left out in turn for training, and unseen class labels were predicted on these left-out test sets (see Appx. B.3 for details). Fig. 3C summarizes the agreement between true and predicted behavioral class labels across all test sets from successful training runs (see Appx. B.3) in a confusion matrix. These results were on par with those produced by various classifiers trained directly on the same 20-dimensional BOLD signal vectors and associated class labels as used for the mmPLRNN (multi-class F1 scores for logistic regression: 0:47, linear discriminant analysis: 0:48, support vector machines (RBF kernel, C = 1): 0:47, mmPLRNN: 0:45). This confirms that the mm-PLRNN has learned the connections between the two data modalities within its latent space in an about optimal manner, i.e., without much loss of information as judged by this comparison.” pg. 8, col. 1, para. 2, marked U-3 in Kramer)
RNN with time series…
“Our complete model setup is illustrated in Fig. 1. We assume that we have observed time series X = fxtg generated by some unknown DS dy=dt = f(y; t) sampled at discrete time points t according to some output distribution p(X j Y), e.g. Gaussian, Poisson, or categorical. In fact, as illustrated in Fig. 1, here we assume that the unknown DS is observed through several such output channels (data modalities) simultaneously, from which we would like to infer the underlying DS which is approximated by a sufficiently expressive RNN. Here we assume that all observed time series were generated by the same underlying DS and hence are naturally aligned via their common time labels (reflecting the most common scenario in the natural sciences where measurements across different modalities are taken simultaneously).” (pg. 3, col. 2, para. 2, marked U-4 in Kramer)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the Dey the ability to use neural network with state space and unsupervised data as taught by Kramer et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Kramer et al. who teaches increasing interest in using machine learning in an unsupervised way.
Claim 23 is rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (18) above in further view of Pub. No. US 2024/0047081 to Welch et al.
Regarding claim 23
The method of claim 21, wherein the personalized disease progression model utilizes supervised computational approaches when a disease onset is given in leveraging the personalized disease progression framework, wherein a true state of a disease is utilized as an additional loss function with regular Evidence Lower Bound (ELBO) based loss.
Dey et al. teaches:
Machine learning model with label noise (supervised learning) and inferred (leveraging) severity…
“Pursuing such joint analysis to find clinical and genomic factors associated with COVID-19 that can be used as biomarkers comes with a few challenges. First, defining COVID-19 severity from EHR is often challenging, as it is not directly collected in these datasets. Hence, the severity has to be inferred using appropriate clinical knowledge as a surrogate phenotype which in turn may introduce label noise and data collection bias. Thus, building machine learning models on clinico-genomic dataset needs special adjustments8 to handle such noise and bias associated with COVID-19 severity. Second, assessing the impact of genomic factors on COVID-19 severity may be confounded by several other clinical factors such as the prior comorbidities of patients and treatment protocols. The effect of such confounding factors has to be addressed carefully when conducting the combined clinico-genomic studies. Third, the impact of clinico-genomic factors may also vary depending on the variations of diverse COVID-19 strains that have been observed due to genetic mutations.” (pg. 378, para. last)
Leveraging standard classification methods…
“Among several state-of-the-art PU learning techniques, we use PU Adapter31, uPU32, and nnPU33 for learning the COVID-19 severity. All of these PU learning techniques have the basic assumption that the observed samples are drawn uniformly from the positive distribution, which enables PU learning techniques to leverage standard binary classification methods with minor modification of data or algorithm.” (pg. 382, para. 1)
The combined references teach disease. They do not teach ELBO loss.
Welch et al. also in the business of disease teaches:
Disease state…
“In some examples, it may be desirable to identify a perturbation that transitions a cell in a diseased state (e.g., a starting cell state) to a healthy cell state (e.g., a target cell state). To this end, some embodiments input a starting cell state and a target cell state into a trained machine learning algorithm, which identifies a perturbation that causes the starting cell state to transition to the target cell state.” [0103]
“The deep learning era brought much more powerful generative models that are constructed with deep neural networks from latent variables to data samples, and are stochastically optimized. Some of the early deep generative models such as restricted Boltzmann machines (Hinton, G. E. (2002), Training products of experts by minimizing contrastive divergence, Neural computation, 14(8), 1771-1800) and sigmoid belief networks (Neal, R. M. (1992), Connectionist learning of belief networks, Artificial intelligence, 56(1), 71-113) were very efficient in data modeling (Welling, M., M. Rosen-Zvi, and G. E. Hinton (2004), Exponential family harmoniums with an application to information retrieval, Advances in neural information processing systems, 17) and their extensions have been applied to different types of data (Hinton, G. E., S. Osindero, and Y.-W. Teh (2006), A fast learning algorithm for deep belief nets, Neural computation, 18(7), 1527-1554; Hinton, G. E., and R. R. Salakhutdinov (2009), Replicated softmax: an undirected topic model, Advances in neural information processing systems, 22; Salakhutdinov, R., and H. Larochelle (2010), Efficient learning of deep boltzmann machines, in Proceedings of the thirteenth international conference on artificial intelligence and statistics, pp. 693-700, JMLR Workshop and Conference Proceedings; Mnih, A., and K. Gregor (2014), Neural variational inference and learning in belief networks, in International Conference on Machine Learning, pp. 1791-1799, PMLR). Three commonly-used classes of deep generative models are variational autoencoders (VAEs) (Kingma, D. P., and M. Welling (2013), Auto-encoding variational bayes, arXiv preprint arXiv:1312.6114), generative adversarial networks (GANs) (Goodfellow, I., J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio (2014), Generative adversarial nets, in Advances in neural information processing systems, pp. 2672-2680) and auto-regressive models such as PixeICNN and PixeIRNN (Oord, A. v. d., N. Kalchbrenner, L. Espeholt, O. Vinyals, A. Graves, et al. (2016a), Conditional image generation with pixelcnn decoders, Advances in neural information processing systems, 29; Oord, A. v. d., N. Kalchbrenner, and K. Kavukcuoglu (2016b), Pixel recurrent neural networks, in International conference on machine learning, pp. 1747-1756, PMLR; Oord, A. v. d., S. Dieleman, H. Zen, K. Simonyan, O. Vinyals, A. Graves, N. Kalch-brenner, A. Senior, and K. Kavukcuoglu (2016c), Wavenet: A generative model for raw audio, arXiv preprint arXiv:1609.03499).” [0123]
Using ELBO loss with GenotypeVAE…
“Some implementations utilized the ELBO loss and trained GenotypeVAE with a learning rate of 10-4 for 300 epochs. During the training, some implementations considered a probability of 0.5 for an iteration with a batch of genes to be genetic perturbations with single target genes, and another probability of 0.5 for the iteration to be used with the next batch of genes to be genetic perturbations with double target genes. Some implementations then evaluated the GenotypeVAE latent spaces for the genetic perturbations of the GI, LINCS-Gene and GSPS datasets. Both GI and GSPS have integer count scRNA-seq samples, while LINCS-Gene has normally distributed microarray samples. Some implementations therefore obtained their cellular representations by training scVl (Lopez, R., J. Regier, M. B. Cole, M. I. Jordan, and N. Yosef (2018), Deep generative modeling for single-cell transcriptomics, Nature methods, 15(12), 1053-1058) on the GI and GSPS data, as well as VAE on the LINCS. FIG. 18A shows the UMAP plots of perturbation representations and cellular representations for the two selected genetic perturbations in each of the three datasets. Both pairs of (KLF1/ctrl, SLC4A1/ctrl) in the GI data and (ERG, ERBB3) in the LINCS-Gene data have very different perturbation representations and cell state distributions. The perturbation representations of the pair of (RPL3, PINK1) in the GSPS data show distinctive distributions, while the difference between their cellular representations is less obvious. The perturbation representations have meaningful mappings to the cellular representations for the GI and LINCS-Gene datasets. The smaller difference in the cellular representation for the GSPS data might be due to its high batch effects (Replogle, J. M., et al. (2021), Mapping information-rich genotype-phenotype landscapes with genome-scale perturb-seq, bioRxiv).” [0228]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to use ELBO as taught by Welch et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Welch et al. who teaches the advantages of using ELBO with disease states. This also provides for better results.
The combined references teach training. They do not explicitly teach true state of a disease. However, Welch et al. also teaches using experts to minimize divergence.
It would have been obvious to one of ordinary skill in the art before the effective filing date of Applicant’s filing to modify the combined references with the knowledge available to such an artisan that experts are used to minimize divergence by verifying (teach true) data. This would have been known work in the field of endeavor prompting variations of it in the same field based on use of providing better results and would provide predictable results.
Claims 29 and 30 are rejected under 35 U.S.C. 103 as being unpatentable over the combined references in section (18) above in further view of Pub. No. US 2024/0047081 to Sharma et al.
Regarding claim 29
The method of claim 22, further comprising:
modifying the personalized disease progression model to include a time gap between successive data points of the plurality of time points, wherein attentions are learned during the personalized disease progression model training, and wherein the personalized disease progression model assumes that a transition between states is dependent on treatment patterns.
The combined references teach treatment and disease trajectories. They do not teach time points
Sharma et al. also in the business of treatment teaches:
Training disease trajectories…
“In another embodiment, a machine-learning based algorithm can be used to automatically predict the disease trajectory for the patient based on the state vector of the patient. Using a machine-learning algorithm, a database of disease trajectories may be used to train a machine-learning based classifier to estimate a disease trajectory based on the state variables of a patient. The trained machine-learning based classifier can then be used to estimate a disease trajectory for a new patient based on the state variables for that patient. Any type of supervised, semi-supervised, or unsupervised machine-learning algorithm can be used to train the machine-learning based classifier. For example, machine-learning algorithms, such as support vector machine (SVM), a machine-learning based regressor, deep learning, or deep reinforcement learning, can be used, but the present invention is not limited thereto. FIG. 3 illustrates predicting a disease trajectory for a patient using a machine-learning based algorithm according to an embodiment of the present invention. As shown in FIG. 3, a patient database 302, which included a large number of disease trajectories of different patients is used to train a machine-learning based algorithm 304. Once the machine-learning based algorithm 304 is trained, the known state variables (at a current time or over an initial time period) for a new patient 306 are input to the trained machine-learning based algorithm 304, and the trained machine-learning based algorithm 304 estimates a predicted trajectory 308 for the new patient.” [0024]
Disease trajectory and change over time (include time gap between time points) based on disease progression and treatment…
“Returning to FIG. 1, at step 106, an optimization is performed to determine an optimal therapy and timing for the patient based on the predicted disease progression. An intervention or therapy, such as drug intervention, surgery, or a device therapy, will cause the system dynamics to change both acutely and over a period of time. Some common examples of interventions include:… “ [0025]
Continuous (successive time points) and discrete dynamics and switching signal with time-dependent and state-dependent…
“Depending on the nature of the intervention, the change in dynamics caused by the intervention may be either gradual or sudden. According to an advantageous embodiment of the present invention, to account for such changes to the dynamics of the state variables, interventions are modeled using hybrid dynamical systems (also referred to as switching systems). In particular, a hybrid dynamical systems approach is used to model the system dynamics, by coupling the continuous dynamics to the discrete intervention events. This can be expressed as:
{dot over (x)}(t)=f.sub.σ(t)(x(t))
where x(t) denotes the continuous dynamics and σ(t) denotes the discrete dynamics that represent the switching signal. The switching used to represent the interventions can be time-dependent or state-dependent. Additionally, the switching can either be intrinsic to the system (i.e., autonomous) or controlled (i.e., the intervention that leads to the switching is designed).” [0033]
State variable and timing of drug (treatment)…
“In some cases of drug therapy, the dose and timing of the drug may be based on the value of a particular biomarker, which is one of the state variables. Once the therapy starts or stops, the biomarker itself is subject to change due to the therapy. The interaction between the continuous dynamics of the system variables and the hybrid switching to another regime based on a rule (e.g., biomarker above or below a particular threshold) can be captured using a state-dependent switching system.” [0035]
Timings changed (time gap) based on therapy…
“By using switching systems (e.g., time-dependent or state-dependent) to model interventions, different types of therapies and different timings for the therapies can be changed to find the type and timing for a therapy that results in the best disease progression for the patient based on the predicted evolution of the state vector. One potential therapy that can be considered is a “do nothing” (or “watchful waiting”) therapy, which can be compared to other therapy types and timings.” [0037]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to change time (time points/gap) for treatment as taught by Sharma et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Sharma et al. who teaches the benefits of disease progression modeling with continuous and discrete interventions and changing time (time gap) based on treatment and treatment progress for a patient data.
Regarding claim 30
The method of claim 29, further comprising:
displaying, in a user interface, the personalized disease progression trajectory to a user for each of the one or more patient groupings, wherein the personalized disease progression trajectory for each of the one or more patient groupings are continuously updates as the additional patient data is received, and wherein the display for each of the one or more patient groupings includes at least a current stage, a progression pattern, and early intervention protocols for each of the one or more patient groupings; and
[No Patentable Weight is given to non-functional descriptive language of
“wherein the display for each of the one or more patient groupings includes at least a current stage, a progression pattern, and early intervention protocols for each of the one or more patient groupings”]
Dey et al. teaches:
Updating data every month (continuously) when repository updated (data received)…
“The dataset contains diverse information including demographics, diagnosis, medications, lab tests, and genomic information of approximately half a million patients. In addition, the national SARS-CoV-2 laboratory test data were made available in UKBB through the Public Health England (PHE). This COVID-19 dataset contains a flag indicating specimen origin of COVID-19 tests: hospital inpatient origin vs other settings. It also notes the specimen collection date and the specimen result in addition to a few other information about how the specimen was collected. Data were available from Mar 16, 2020 and the repository has been updated once or twice every month. In our current study, we used data from Mar 16, 2020 to Jan 21, 2021.” (pg. 379 last para – pg. 380 first para.)
See Display below.
retraining the personalized disease progression model based on the comparing of the actual disease progression pattern to the personalized disease progression trajectory for the corresponding patient grouping.
Data repository updated…
“Dataset. We analyzed the large prospective cohort of patients from the UK collected in the UKBB24 repository. The dataset contains diverse information including demographics, diagnosis, medications, lab tests, and genomic information of approximately half a million patients. In addition, the national SARS-CoV-2 laboratory test data were made available in UKBB through the Public Health England (PHE). This COVID-19 dataset contains a flag indicating specimen origin of COVID-19 tests: hospital inpatient origin vs other settings. It also notes the specimen collection date and the specimen result in addition to a few other information about how the specimen was collected. Data were available from Mar 16, 2020 and the repository has been updated once or twice every month. In our current study, we used data from Mar 16, 2020 to Jan 21, 2021.” (pg. 379, para. last – pg. 380, para 1)
Dataset evolved over time…
“Preparing the outcome and feature sets. Although the COVID-19 dataset in UKBB is a prospective study, it did not collect the disease severity explicitly. The nature of this dataset is very diverse depending on national testing strategy which evolved over time as the pandemic progressed. For example, in UK, testing was initially restricted to those with symptoms in hospital and thus under this assumption, using the positive tests of these subjects in hospital can be a reasonable proxy for severity. However, when the testings were expanded covering even asymptomatic patients from diverse facilities, defining the severity required analyzing information regarding the specimen sources of test.” (pg. 380, para. 2)
Nested test-training set and working data for further train-test splits (retraining data) to rank and iteratively finds stable set (improve accuracy)…
“2. Stable Feature Extraction: The learning rubric employs a nested test-training set configuration on the genomic data after QC and pruning for LD. The outer test-training set split reserves the test set (“validation”) for final SNP evaluation and denotes the training set as “working” data. Within the latter, further train-test splits are performed to rank the features by their Youden index28 (also known as J statistic). RubricOE is applied on the genomic data, accounting for top 20 PCs, sex, year of birth and Diabetes Mellitus (DM) as covariates. It iteratively finds a “stable” set of SNPs which persistently remains highly ranked with their respective Youden index. Annotation, Gene Ontology (GO) and enriched pathway analysis of the “stable” set of SNPs produced by RubricOE has been evaluated using the R package clusterProfiler29 v3.12.” (pg. 381, para. 2)
Example of evaluate and accuracy…
“Model Evaluation. We evaluate the effectiveness of PU-learning algorithms with a comparison with baseline traditional classification algorithm using logistic regression with a L1 loss. Computing traditional metrics such as precision, recall, F1 score, and accuracy is challenging in the positive unlabeled scenario since the only information available relates to the positive label while no sample from the negative class is clearly provided. One straightforward solution to this problem is to assume that the unlabeled data are negative. However, this is not fully accurate. A modified F1 score (we call it as Fpu 1 score) has been proposed to address this issue34. This score is computed as r2 p(^y=1) , where r is the recall and p(^y = 1) is the prevalence of the predicted positive label. As shown in34, the Fpu 1 score can take values greater than one but shares the same property as F1 score; it is high when both precision and recall are high and small when either one is small. We also report metrics used to evaluate traditional classification methods for completeness although these are not recommended for our setting. They include accuracy, precision, recall, and F1 score.” (pg. 382, para. 3)
Clinical-genomic model (personalized disease progression trajectory) associated with (comparing) COVID-19 severity (actual disease progression pattern)…
“Our final combined clinico-genomic model finds the significant features associated with COVID-19 severity (Figure 2), which may act as potential biomarkers. Some of the top significant clinical features have already been found in previous studies. For example, age and black ethnicity have been reported in7, and prior conditions like diabetes mellitus, hypertension have been reported in6, 14, 15. Note that our study finds smoking status (past smoker, current smoker, pack years of smoking) is related to severe COVID-19 patients, which has ambiguous evidence in literature. A few studies7 reported it with increased risk factors, while others14 did not find it to be a significant risk factors for COVID-19 severity.” (pg. 383, para. 2)
See Retrain below.
Display
The combined references teach progression. They do not teach display.
Sharma et al. also in the business of progression teaches:
Example of disease progression…
“In another embodiment, the prediction of the disease progression for the patient based on the state variable can be performed using a trajectory clustering algorithm. In particular, a trajectory clustering algorithm can be used to group the database of disease trajectories into several clusters. A difference metric, such as the Euclidean norm, is used to compare the current state vector of the patient or an initial disease trajectory defined by the state vector values over an initial time period to the clusters in the database to and assign the current patient to one of the clusters by finding the most similar cluster. The patient's disease trajectory is then determined using the average disease trajectory for the assigned cluster in the database of disease trajectories. FIG. 2 illustrates disease progression modeling using trajectory clustering according to an embodiment of the present invention. As shown in FIG. 2, P.sub.1, P.sub.2, P.sub.3, . . . , P.sub.n are average disease trajectories for clusters 1-n in a database of disease trajectories. It can be noted that in the trivial case, the trajectories P.sub.1, P.sub.2, P.sub.3, . . . , P.sub.n may represent clusters of size 1, i.e., individual patients rather than a group of patients, but in an advantageous implementation each of these trajectories represents a cluster of multiple patients. Q represents the patient's disease trajectory, with 202 being the known initial disease trajectory between t.sub.0 and t.sub.p, and 204 being the predicted disease trajectory between t.sub.p and t.sub.f. The initial disease trajectory 202 of Q is compared to the initial portions of P.sub.1-P.sub.n, and it the most similar cluster to Q is determined to be P.sub.3. In order to estimate the predicted disease trajectory 204 of Q, the evolution of the state variables in Q between t.sub.p and t.sub.f match the evolution of the state variables in P.sub.3 between t.sub.p and t.sub.f. For the trajectory clustering, the whole state vector or some predefined subset of state variables in the state vector may be used to find the most similar cluster.” [0023]
Example of displaying disease trajectories…
“Returning to FIG. 1, at step 108, an optimal therapy recommendation and predicted disease trajectories are output. For example, the optimal therapy type and timing can be output by displaying a recommendation including the optimal therapy type and timing on a display device of a computer system, sending this information in a message (e.g., email), and/or storing the optimal therapy type and timing recommendation on a storage or memory of a computer system. In addition, various predicted disease trajectories for the patient can be output. For example, a predicted disease trajectory for a “do nothing” scenario and predicted disease trajectories for one or more therapy options including different types and timings of therapies can be displayed on a display device of a computer system or stored in a memory or storage of a computer system. These output results can be incorporated in a Clinical Decision Support (CDS) system. These output results can be integrated into an Electronic Medical Record (EMR) system. These output results can be incorporated in a radiologist's reading system at the time when a diagnosis is being mage to inform the radiologist of the potential outcomes the patient might have. These output results can be included in a report that is accessible to the patient so the patient can make informed decisions.” [0044]
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to display progression trajectory as taught by Sharma et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Sharma et al. and the health benefits of displaying trajectories and recommendations as this helps to visually determine if progress is being made with respect to health.
Retrain
The combined references teach machine learning. They do not teach retrain.
Krittanawong et al. also in the business of machine learning teaches:
Collecting enough (additonal) training data…
“The next step is to train the network. During training, the model parameters are learned by the network from the training data provided relative to the labeled examples using backpropagation and other related gradient descent learning techniques. The major
challenge of this task is collecting enough training data and optimizing the hyperparameters (e.g., initialization strategy, learning rate, regularization techniques) so that the network can learn a robust set of parameters for the given prediction. It may also be necessary to reconsider the overall network architecture if the performance remains low. Importantly, given enough training data, sufficient computational resources, and an appropriate network architecture, nearly any mathematical function can be learned, including highly abstract functions from genomics data or image data to a disease state.” (pg. 11, last para. – pg. 12, top para., marked V-5 in Krittanawong)
Example of feedback (additional data) and iteratively train (therefore, retrain)…
“Once training is complete, the creation and analysis of artificial nucleotide sequences, such as the creation of artificial human genomes [82] or artificial enhancers (“synthetic DNA”), using approaches such as generative adversarial networks (GANs), can be considered [83]. GANs are DL models that include two primary components: a generator and a discriminator. Generated DNA sequences are used as inputs for the discriminator to analyze if the model has generated a convincingly real biological sequence. This feedback is used to iteratively train the generator model to produce artificial sequences with increasingly realistic properties. For example, a recent study used a type of GAN (an auxiliary classifier generative adversarial network) to generate synthetic participants that closely resembled real participants from the SPRINT trial (Systolic Blood Pressure Trial) to facilitate exploratory analyses [84].” (pg. 12, para. 3, marked V-6 in Krittanawong)
It would have been obvious to one of ordinary skill in the art before the effective filing date to include in the method and system of the combined references the ability to retrain as taught by Krittanawong et al. since the claimed invention is merely a combination of old elements and in the combination each element merely would have performed the same function as it did separately, and one of ordinary skill in the art would have recognized that the results of the combination were predictable. Further motivation is provided by Krittanawong et al. who teaches the advantages of using feedback to iteratively train models for exploratory analysis and this improves the model.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KENNETH BARTLEY whose telephone number is (571)272-5230. The examiner can normally be reached Mon-Fri: 7:30 - 4:00 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, SHAHID MERCHANT can be reached at (571) 270-1360. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/KENNETH BARTLEY/Primary Examiner, Art Unit 3684
Prosecution Insights