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 .
Interview Summary
An interview summary for the interview conducted on 09/11/2025 is attached.
Claim Status
Claims 1, 5, 7-23, and 25-28 are currently pending and under examination herein.
Claims 2-4, 6, and 24 are cancelled.
Priority
Acknowledgment is made of applicant's claim for foreign priority to GB1819498.5 filed 11/29/2018, under 35 U.S.C. 119 (a)-(d). Receipt is acknowledged of certified copies of papers required by 37 CFR 1 .55. Accordingly, the effective filing date of the claimed invention is 11/29/2018.
Withdrawn Rejection/Objections
Rejections and/or objections not reiterated from previous office actions are hereby
withdrawn in view of the amendments filed 12/01/2025.
All rejections of claims 3 and 4 are hereby withdrawn; their cancelation moots the
rejections.
The following rejections and/or objections are either maintained or newly applied. They
constitute the complete set presently being applied to the instant application.
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.
1, 3-5, 7-23, and 25-28 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
The Supreme Court has established a two-step framework for this analysis, wherein a claim does not satisfy § 101 if (1) it is “directed to” a patent-ineligible concept, i.e., a law of nature, natural phenomenon, or abstract idea, and (2), if so, the particular elements of the claim, considered “both individually and as an ordered combination,” do not add enough to “transform the nature of the claim into a patent-eligible application.” Elec. Power Grp., LLC v. Alstom S.A., 830 F.3d 1350, 1353 (Fed. Cir. 2016) (quoting Alice, 134 S. Ct. at 2355). Applicant is also directed to MPEP 2106.
Step 1: The instantly claimed invention is directed to methods (claims 1, 5,7-19, 23 and 26 being representative) and an apparatus (claims 20-22 being representative). Therefore, the instantly claimed invention falls into one of the four statutory categories. [Step 1: YES]
Step 2A: First it is determined in Prong One whether a claim recites a judicial exception, and if so, then it is determined in in Prong Two if the recited judicial exception is integrated into a practical application of that exception.
Step 2A, Prong 1: Under the MPEP § 2106.04, the Step 2A (Prong 1) analysis requires determining whether a claim recites an abstract idea, law of nature, or natural phenomenon.
Claims 1, 3-5, 7-23, and 25-28 recite the following steps which fall under the mathematical concepts, mental processes, and/or certain methods of organizing human activity groupings of abstract ideas:
Claims 1 and 26 recite a method of training a neural network to learn ligand binding similarities between protein binding site, and using said model. The limitation training a neural network fall into mathematical concepts groupings of abstract idea since the model uses mathematical calculations as claimed in claims 11-14 (Loss functions).
Claims 1 and 26 further recite projecting the representation of the first binding site and the representation of the second binding site into a latent space; the limitation projecting representations into a latent space uses mathematical calculations to transform data into a lower-dimensional space, and as such falls within mathematical concepts groupings of abstract ideas.
Claims 1 and 26 further recite performing a comparison between the similarity indicator and the label; the limitation performing a comparison between the similarity indicator and the label is a mathematical calculation as disclosed in the present specification (similarity indicator is a distance [0050]; the label comprises a binary value [0021]; using contrastive loss function to compare the similarity indicator [0057]), and as such, falls within mathematical concepts grouping of abstract ideas; said limitation is also, a mental process because human mind is able to compare values.
Claim 1 further recite updating the neural network based on the comparison; the limitation updating the machine learning model based on the comparison falls into mathematical concepts groupings of abstract idea since it uses mathematical equations/formulas as claimed in claim 12 (loss function). Also, a mental process because human mind is able to compare values and update a model based on the comparison.
Claim 26 further recites converting the representation of the first binding site and the representation of the second binding site into respective grids of voxels, each voxel representing occupancy and pharmacophoric properties of a different volume of the corresponding protein binding site; the limitation converting a representation into voxels involves mathematical calculations, and as such, falls into mathematical concepts groupings of abstract ideas.
Claim 26 further recites determining whether the different protein binding sites bind the same ligand; the limitation determining is considered a mathematical relationship as disclosed in the present specification (“the label may comprise a score indicating a likelihood that the binding sites bind the same ligand, but in the example system 200 of Figure 2 the label is binary and indicates a yes or no answer” [0048]; “the label takes binary values of 0 and 1 indicating whether or not two binding sites bind the same ligand. As a result, if a label takes a value of 0 indicating that the binding sites do not bind the same ligand, then the loss function may require the projected binding sites to be separated by a minimum distance in latent space” [0059]). As such the recited limitation falls into mathematical concepts grouping of abstract ideas. See MPEP 2106(a)(2) I. A. Additionally, said limitation can be performed in human mind (mental process), since human mind is capable of determining whether binding site binds the same ligand based on the result of an analysis.
Claim 26 further recites training the neural network comprises minimising a contrastive loss function; the limitation training a neural network by minimizing a loss function is considered a mathematical calculation, and as such, falls within mathematical concepts grouping of abstract ideas.
Claims 11 recites performing the comparison comprises minimizing a loss function; the limitation performing the comparison falls into mathematical concepts groupings of abstract idea since it uses mathematical formulas/equations. Also, a mental process because human mind is able to perform the recited function.
Claim 12-14 recite updating the machine learning model comprises performing back propagation using the minimized loss function. The limitation updating the machine learning model falls into mathematical concepts groupings of abstract idea since it uses mathematical equations/formulas (loss function). Also, a mental process because human mind is able to compare values and update a model based on the comparison.
Claim 15 recite jittering the binding sites in input space; the limitations jittering the binding site can be practically performed in human mind (mental process) because human mind is able to adjust the atomic positions of a molecular structure.
Claim 23 recites detecting a ligand binding similarity score between two protein binding sites using a neural network trained according to the computer-implemented method of claim 1; the limitation detecting a ligand binding similarity score can be practically performed in human mind (mental process), since human mind is able to detect/perceive similarity based on a result of a mathematical algorithm.
Claim 25 recites that the neural network provides translationally and rotationally invariant descriptions of the first binding site and the second binding site in latent space; the limitation providing invariant descriptions is considered a mathematical calculation, as disclosed in the present specification (using Van der Waals functions [0061-0064]), and as such, falls within mathematical concepts groupings of abstract ideas.
Claim 28 recites that centroids of first and second binding sites are jittered by up to 2 angstroms before inputting the representations of the first and second binding sites into the neural network; the limitation jittering the representation is considered a mathematical calculation (specification [0062]: using binding site detection algorithms), and as such, falls into mathematical concepts groupings of abstract ideas.
Therefore, the above limitations fall under the “Mental Process” and/or “Mathematical
concept” groupings of abstract idea. See MPEP 2106.04 (a)(2) III and MPEP 2106.04(b) I. As such, claims 1-25 recite an abstract idea. [Step 2A, Prong 1: YES]
Response to Applicants Arguments
Applicant in Remarks submitted 12/01/2025 states:
The Office has impermissibly failed to consider dependent claims 5, 7-10, and 16-22 independently under its subject matter eligibility analysis.
It is respectfully submitted that this is not persuasive. As currently drafted, there are no active steps recited in claims 5, 7-10, and 16-22, and as such, they merely provide more information about the inputting step and the type of model.
Applicant further submits that:
Independent claim 1 is directed towards a method of training a neural network. Example 39 of the USPTO Subject Matter Eligibility Examples provides that a method of training a neural network does not recite a mental process.3 While the underlying subject matter of claim 1 as amended and Example 39 may be different, claim 1 as amended mirrors Example 39. For example, claim 1 is directed towards training a neural network to learn ligand binding similarities while Example 39 is directed towards training a neural network for facial detection. However, both claim 1 and Example 39 are directed towards methods of training a neural network. Furthermore, both claim 1 as amended and Example 39 recite analogous method steps including collecting or inputting data, transforming or projecting the data, and training or updating the neural network. As Example 39 states "the claim does not recite a mental process because the steps are not practically performed in the human mind."4 Thus, for at least the same reasons as provided by the Office in Example 39, claim 1 as amended also does not recite a mental process since it cannot be practically performed in the human mind.
Additionally, Applicant submits that at least the step of updating the neural network
based on the comparison cannot practically be performed in the human mind. As acknowledged in Example 39, the analogous step of training the neural network was held to not recite a mental.
Applicant respectfully submits that the Office's assertion that claim 1 require analogous steps directed to a practical application under the Step 2A, prong one of the subject matter eligibility analysis is erroneous. Per MPEP § 2106.04(II)(A)(1), "[i]n [Step 2A] Prong One examiners evaluate whether the claim recites a judicial exception, i.e. whether a law of nature, natural phenomenon, or abstract idea is set forth or described in the claim."7 Thus, Applicant submits that the proper analysis under Step 2A, prong 1, does not require a determination that claim 1 is integrated into a practical integration. Instead, this is reserved for the Step 2A, prong 2 analysis.8 Accordingly, Applicant submits that the Office's assertion is erroneous and claim 1 as amended does not recite a mental process for at least the reasons outlined above.
The Applicant remarks are directed to Step 2A Prong One of 101 analysis, specifically that whether the claims recite a judicial exception.
It is respectfully submitted that this is not persuasive. Example 39 is directed to training a neural network for facial detection. In contrast, the instant claims do not include any analogous steps directed to analyzing digital facial images (not a judicial exception). Instant claims are directed to training a model using binding site representation that comprise data, for example, structural information. As stated above, the broadest reasonable interpretation of claims 1 and 26 in light of the specification encompasses mental and/or mathematical concepts for at least the steps of training, performing a comparison, projecting, updating, converting, and determining. See MPEP 2106.04(a)(2) C.
With regards to Applicant stating that the step of updating the neural network based on the comparison cannot practically be performed in the human mind since it is analogous to step of training the neural network, Examiner states that in contrast to instant claimed invention, Example 39 does not recite any updating based on comparison steps (mental and/or mathematical steps). The limitation updating based on a comparison, given the plain meaning of “updating” is considered a mental process since the plain meaning of “updating” includes observation, evaluation, judgment, and opinion (See MPEP 2106.04(a)(2), subsection III.) that can practically be performed in human mind, for example, by using a pen and paper or computer.
With regards to Applicant stating that under the Step 2A, prong one of the subject matter eligibility analysis is erroneous, Examiner submits that in the mentioned response, whether the claims were directed to the similar subject matter was analyzed, not the practicality of the invention. Specifically, in the 06/30/2025 Office Action Examiner responded that “Instant application is more similar to Example 47, which is directed to training a neural network that is used to determine and evaluate anomaly in data… Similar to example 47, instant claimed training, performing a comparison, projecting, updating, converting, and determining encompass performing mathematical calculations ...”.
As such, the claimed invention recites abstract ideas.
Applicant further states that:
unlike claim 2 of Example 47, claim 1 as amended does not recite or encompass any specific mathematical calculations. Put differently, claim 1 as amended does not recite specific mathematical formulas, equations, or calculations unlike claim 2 of Example 47. Instead, claim 1 as amended may merely involve underlying mathematical concepts, similar to the ANN of claim 1 of Example 47. However, as acknowledged by the Office in its analysis of claim 1 of Example 47, this is insufficient to determine that a claim recites a mathematical concept.
Applicant submits that the assertion that claim 1 as amended does not recite a mathematical concept is supported by the Office's August 4, 2025 Memorandum entitled "Reminders on evaluating subject matter eligibility of claims under 35 U.S.C. 101." Specifically, this Memorandum highlights the distinction between "claims that recite an exception (which require further eligibility analysis) from claims that merely involve an exception (which are eligible and do not require further eligibility analysis)."
It is respectfully submitted that this is not persuasive. It is important to note that words used in a claim operating on data to solve a problem can serve the same purpose as a formula or mathematical relationship/mathematical concepts (MPEP 2106.04(a)(2)).
Additionally, claim limitations are analyzed given their broadest reasonable interpretation in light of the specification (MPEP 21006.04 (a)(2) C). For example, in Example 47, it was analyzed that “Step (a) recites “training . . . the ANN based on input data . . . to generate a trained ANN.” …. When given their broadest reasonable interpretation in light of the disclosure, the backpropagation algorithm and gradient descent algorithm are mathematical calculations.”, in other words, the limitations of the claim were analyzed in light of specification. Similarly, for at least “training a neural network … comprising … projecting … updating …” in light of specification are considered mathematical concepts (see specification, minimizing a contrastive loss function [0050]; updating the neural network by back propagation using the minimized contrastive loss; The similarity indicator 220 may for example be based on a distance between the binding sites in a Euclidean latent space [0049], and so on). Similar to Example 47 limitations of “performing the training using a backpropagation algorithm and a gradient descent algorithm” instant claims limitations of “training comprising “projecting” and “updating” recite mathematical concepts in light of specification.
Step 2A: Prong 2: Under the MPEP § 2106.04, the Step 2A, Prong 2 analysis requires identifying whether there are any additional elements recited in the claim beyond the judicial exception(s), and evaluating those additional elements to determine whether they integrate the exception into a practical application of the exception. This judicial exception is not integrated into a practical application for the following reasons.
The additional elements of claims include the following.
Claims 1, and 18-22 recite inputting a representation of a first binding site; a representation of a second binding site. Claim 1 further recites outputting a similarity indicator based on the representations of the first and second binding sites.
Claims 20 recites an apparatus comprising a processor, a memory unit and a communication interface to implement the recited method.
Claim 21 and 22 recite a computer-readable medium comprising data or instruction code representative of a machine learning model generated according to the method.
The additional elements of an apparatus comprising a processor, a memory, a communication interface, and a non-transitory computer-readable storage medium are generic computer components and/or processes. The courts have found the use of a computer or other machinery in its ordinary capacity for economic or other tasks (e.g., to receive, store, or transmit data) or simply adding a general-purpose computer or computer components after the fact to an abstract idea (e.g., a fundamental economic practice or mathematical equation) does not integrate a judicial exception into a practical application. See MPEP 2106.05(f). Furthermore, the additional elements of inputting and outputting data serve to collect the information for use by the abstract idea.
Therefore, the additionally recited elements amount to insignificant extra-solution activity and, as such, the claims as a whole do no integrate the abstract idea into practical application. Thus, claims 1, 3-5, and 7-23, and 25-28 are directed to an abstract idea. [Step 2A, Prong 2: NO]
Response to Applicants Arguments
Applicant in Remarks submitted 12/01/2025 states:
claim 1 as amended integrates any allegedly abstract ideas into the practical application of identifying similarities in protein binding sites for use in drug discovery. As outlined in the specification there is a need for identifying similarities in protein binding sites that are not based on intuition. Existing techniques are biased and "place restrictions on their ability to accurately identify binding sites with similar ligand binding characteristics,"20 particularly since "some pairs of binding sites may bind the same ligand despite looking quite different to the human observer."21 Claim 1 as amended thus provides this improvement in the practical application of recognizing similarities in protein binding sites for use in drug discovery since the added elements provide for a method for training a neural network to recognize similarities in protein binding sites with improved accuracy.
It is respectfully submitted that this is not persuasive. The Applicant remarks are directed to Step 2A Prong Two of 101 analysis, specifically whether the additional elements integrate the recited judicial exception into a practical application of the exception. As stated above, the claims do not recite any additional elements that meet any of the relevant considerations for evaluating integration of the judicial exception into a practical application. Taken as a whole, the claims are interpreted as being directed to an improved algorithm related to finding similarity. It is important to note, the judicial exception alone cannot provide the improvement (See MPEP 2106.04(d) III). The improvement must be provided by one or more additional elements. See the discussion of Diamond v. Diehr, 450 U.S. 175, 187 and 191-92, 209 USPQ 1, 10 (1981)) in subsection II, below. In addition, the improvement can be provided by the additional element(s) in combination with the recited judicial exception. See MPEP § 2106.04(d) (discussing Finjan, Inc. v. Blue Coat Sys., Inc., 879 F.3d 1299, 1303-04, 125 USPQ2d 1282, 1285-87 (Fed. Cir. 2018)).
Applicant further states:
claim 1 as amended also recites the additional steps of "projecting, by the neural network, the representation of the first binding site and the representation of the second binding site into a latent space" and "outputting from the neural network a similarity indicator based on the representations of the first and second binding sites, the similarity indicator indicating whether or not the first and second binding sites bind the same ligand, and wherein the similarity indicator comprises a measure of distance between the representations of the first binding site and the second binding site in the latent space" none of which recite a mathematical concept for at least the reasons outlined above and thus should be considered in the Office's eligibility analysis under Step 2A, prong two.
Furthermore, claim 1 has been amended to recite that the "structural information comprises volumetric information and wherein the representations of the first and second binding sites each comprise an encoded three-dimensional grid of voxels, each voxel being associated with an occupancy value indicating whether an atom is present." These added limitations further integrate any alleged abstract ideas into the practical application of recognizing similarities in protein binding sites for use in drug discovery.
These additional limitations reflect the improvement in recognizing similarities in protein binding sites for use in drug discovery by allowing the model to identify binding sites with similar structural binding characteristics regardless of the visual similarities of the binding sites and without making assumptions on the similarities between the binding sites. As a result, the model trained using the method of claim 1 "provides improved rates of accurate predictions of whether or not binding sites bind the same ligand,"23 which "has positive implications for fields such as drug discovery where such information helps to answer questions such as which ligands bind a particular protein, and what is the function of a particular protein."24
As a result, claim 1 as amended provides a method that allows for an improved neural network to identify similarities in protein binding sites for use in drug discovery. As a result, claim 1 as amended integrates the judicial exception into a practical application and is patent eligible under under Step 2A prong 2 of the Alice Mayo test.
It is respectfully submitted that the above statements are not persuasive. The Applicant remarks are directed to Step 2A Prong Two of 101 analysis, specifically whether the additional elements integrate the recited judicial exception into a practical application of the exception. As stated above, the step of projecting is an abstract idea. Furthermore, the limitations of "structural information comprises volumetric information and wherein the representations of the first and second binding sites each comprise an encoded three-dimensional grid of voxels, each voxel being associated with an occupancy value indicating whether an atom is present” further describes the inputting step that amounts to necessary data gathering, as such, insignificant extra-solution activity and does not integrate the judicial exception into a practical application (see MPEP 2106.05(g)). Additionally, the outputting step amounts to necessary data outputting that does not integrate the judicial exception into a practical application.
With regards to applicant stating that the “added limitations further integrate any alleged abstract ideas into the practical application of recognizing similarities in protein binding sites for use in drug discovery”, Examiner stated that claimed invention as currently drafted does not recite any limitations regarding any drug discovery.
Therefore, the additionally recited elements amount to insignificant extra-solution activity and, as such, the claims as a whole do no integrate the abstract idea into practical application.
Step 2B: In the second step it is determined whether the claimed subject matter includes additional elements that amount to significantly more than the judicial exception. See MPEP § 2106.05.
The claims do not include any additional steps appended to the judicial exception that are sufficient to amount to significantly more than the judicial exception.
The additional elements of claims include the following.
Claims 1, and 18-22 recite inputting a representation of a first binding site; a representation of a second binding site. Claim 1 further recites outputting a similarity indicator based on the representations of the first and second binding sites.
Claims 20 recites an apparatus comprising a processor, a memory unit and a communication interface to implement the recited method.
Claim 21 and 22 recite a computer-readable medium comprising data or instruction code representative of a machine learning model generated according to the method.
The additional elements of an apparatus, a processor, a memory, a communication interface, and a non-transitory computer-readable storage medium, are well-understood, routine, and conventional computer components and/or processes. The courts have found the use of a computer or other machinery in its ordinary capacity for economic or other tasks (e.g., to receive, store, or transmit data) or simply adding a general-purpose computer or computer components after the fact to an abstract idea (e.g., a fundamental economic practice or mathematical equation) does not provide significantly more. See Affinity Labs v. DirecTV, 838 F.3d 1253, 1262, 120 USPQ2d 1201, 1207 (Fed. Cir. 2016) (cellular telephone); TU Communications LLC v. AV Auto, LLC, 823 F.3d 607,613,118 USPQ2d 1744, 1748 (Fed. Cir. 2016) (computer server and telephone unit).
Furthermore, the additional elements of inputting and outputting data amounts to necessary data gathering and outputting. See MPEP 2106.05(g)(3).
Therefore, the additional element is not sufficient to amount to significantly more than the judicial exception.
Taken alone, the additional elements do not amount to significantly more than the above-identified judicial exception(s). Even when viewed as a combination, the additional elements fail to transform the exception into a patent-eligible application of that exception. Thus, the claims as a whole do not amount to significantly more than the exception itself. [Step 2B: NO]
Therefore, the instantly rejected claims are not drawn to eligible subject matter as they are directed to an abstract idea (and/or natural correlation) without significantly more. For additional guidance, applicant is directed generally to applicant is directed generally to the MPEP § 2106.
Response to Applicants Arguments
Applicant in Remarks submitted 12/01/2025 states:
Even if amended claim 1 includes a step that could be considered to be directed to an abstract idea, it is eligible at least under Step 2B of the Alice Mayo test because it recites additional elements that amount to significantly more than any alleged abstract idea. In particular, claim 1 as amended recites "inputting to the neural network (a) a representation of a first binding site, (b) a representation of a second binding site, wherein the representations of the first and second binding sites comprise structural information relating to three-dimensional structures of the first and second binding sites, wherein the structural information comprises volumetric information and wherein the representations of the first and second binding sites each comprise an encoded three-dimensional grid of voxels, each voxel being associated with an occupancy value indicating whether an atom is present." Applicant respectfully submits that projecting representations of first and second protein binding sites that are based on structural volumetric of two different binding sites into a latent space was not well-understood, routine, and conventional at the time the application as filed. In fact, the Office has not correctly identified these elements anywhere in the prior art. Thus, these additional elements provide an inventive concept that makes amended claim 1 patent eligible under Step 2B of the Alice Mayo test.
Therefore, claim 1 as amended is patent eligible for at least the above reasons. Claims 5, 7-23, 25 and 28 depend from claim 1 as amended and are eligible for at least the same reasons.
It is respectfully submitted that the above statements are not persuasive. The Applicant remarks are directed to Step 2B of 101 analyses, specifically evaluating additional elements to determine whether they amount to an inventive concept by considering them both individually and in combination to ensure that they amount to significantly more than the judicial exception itself.
As stated above, the limitations of “wherein the structural information comprises volumetric information and wherein the representations of the first and second binding sites each comprise an encoded three-dimensional grid of voxels, each voxel being associated with an occupancy value indicating whether an atom is present” while not reciting any active steps, provide additional information about the inputting step which is considered an extra solution activity that does not amount to significantly more. Additionally, the projecting step, as stated above, is an abstract idea, not an additional element to be considered in Step 2B.
Same analysis applies to corresponding limitations of claim 26.
Therefore, the instantly rejected claims are not drawn to eligible subject matter as they are directed to an abstract idea without significantly more.
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.
Claims 1, 3-5, 7-23, and 25-28 are rejected under 35 U.S.C. 103 as being unpatentable over Brylinski (Brylinski M (2014) eMatchSite: Sequence Order-Independent Structure Alignments of Ligand Binding Pockets in Protein Models. PLoS Comput Biol 10(9): e1003829. doi: 10.1371/journal.pcbi.1003829) in view of Heifets et al. (US 9373059 B1).
Regarding claims 1 and 26, Brylinski discloses a computer-implemented method of training a model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document) , the method comprising: inputting to the machine learning model: a representation of a first binding site; a representation of a second binding site (P. 6-7, section: Pocket Similarity Score: alignments of pairs of ligand binding pockets; also, the entire document), wherein the representations of the first and second binding sites comprise structural information (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; also, p. 4, col. 2, Section: Secondary Structure Score; also, p. 3, Section: Design and Implementation: Evolutionary information is included as sequence profiles and entropy, as well as secondary structure profiles; also, the entire document) ; and a label comprising an indication of ligand binding similarity between the first binding site and the second binding site (p. 6, col. 2, para. 3: “We follow a similar 6-fold cross-validation protocol (label) as described above for assessing the inter-residue distance prediction. Machine learning for the estimation of pocket similarity is implemented using the Support Vector Machines algorithm for classification problems provided by libSVM 3.14; it is considered implicitly disclosed that for said cross- validation a label (the known pocket similarity score) is used);
Brylinski further discloses outputting from the machine model a similarity indicator based on the representations of the first and second binding sites; performing a comparison between the similarity indicator and the label; and updating the machine learning model based on the comparison (p. 6, col. 2, para. 3: “The training and validation of the machine learning model used to assess similarities between pairs of pockets”, which implicitly entails comparing the machine learning output with known instances and adjusting the model until input and output are equal with a certain precision). Brylinski further discloses validation of the fold-independent matching of ligand binding sites requires specific datasets of proteins that bind chemically similar ligands despite having different sequences and structures (pg.3, col. 2, para. 1)).
Brylinski further discloses that the similarity indicator comprises a measure of distance between the representations of the first binding site and the second binding site (In pocket matching calculations, we used only those proteins, for which the center of each of the best of top five binding sites is predicted within a distance of 8 A˚ from the experimental pocket center (pg. 4, col. 1, para. 3)).
Brylinski further discloses that the structural information relates to three-dimensional structure of the binding sites (p. 3, Section: Design and Implementation: “Evolutionary information is included as sequence profiles and entropy, as well as secondary structure profiles (Secondary structure of a protein refers to the three-dimensional structure of a protein)”).
Brylinski further discloses that the structural information relates to three-dimensional structure of the binding sites (p. 3, Section: Design and Implementation: “Evolutionary information is included as sequence profiles and entropy, as well as secondary structure profiles (Secondary structure of a protein refers to the three-dimensional structure of a protein)”).
Further regarding claims 1 and 26, Brylinski does not expressly disclose that the machine learning architecture is a neural network, input structural details and projecting, by the neural network, the representation of the first binding site and the representation of the second binding site into a latent space.
Heifets discloses a computer-implemented method of using a neural network to predict whether a different protein binding sites bind the same ligand (A computer system for characterization of a test object using spatial data, the computer system (claim 1)).
Heifets further discloses training the neural network on a labelled dataset that includes functional information comparing ligand binding abilities along with binding site structures for protein binding sites (evaluating spatial data and training object including structural data and active site information (col. 6, last para.); see also, the spatial coordinates of the target objects are determined using modeling methods such as density functional methods (col. 10, para. 3)).
Heifets further discloses labelled dataset comprising a representation of a first protein binding site have a first structure (the geometric data is normalized by choosing the origin of the X, Y and Z coordinates to be the center of mass of a binding site of the target object (col. 18, para. 6)), a representation of a second protein binding site having a second structure different than the first structure (the target object is a polymer having an active site, and the sampling samples the test object or training object (col. 18, para. 6); the test objects are large polymer, such as antibodies (col. 11. Para. 3)). Heifets further discloses
Heifets further discloses a label indicating whether the first protein binding site and second protein binding site bind the same ligand (In an embodiment, the neural network may optionally, where training data is labeled (e.g., with the binding data), tune the weights within the network to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations (col. 30, para. 4)).
Heifets et al. discloses voxel maps of the test object and target object from their three-dimensional structures (Figure 2A, 208: the target object with three-dimensional coordinates; Figure 2B, 216: Docking test object; Figure 2B: voxel maps; Summary, p. 31, col. 2, Creating a Voxel Map & also, the entire document).
Heifets further discloses voxel maps of the test object and target object from their three-dimensional structures (Figure 2A, 208: the target object with three-dimensional coordinates; Figure 2B, 216: Docking test object; Figure 2B: voxel maps; Summary, p. 31, col. 2, Creating a Voxel Map & also, the entire document).
Regarding claim 5, Brylinski discloses the limitations of claim 4 as stated above. Brylinski further discloses a representation of a first binding site; a representation of a second binding site, wherein the representations of the first and second binding sites comprise structural information, as stated above. Brylinski does not expressly disclose a voxel associated with a further value. However, HIROKAZU discloses each voxel is associated with a further value indicating a property selected from the set of hydrophobicity, aromaticity, acceptance or donation of a hydrogen bond, positive or negative ionizability, and being metallic (p. 30, col. 1, para. 2-3: test and training objects have hydrogen bond donor/acceptor, aromatic rings, …; also, the entire document).
Regarding claim 7, Brylinski discloses the limitations of claim 6 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network comprising one or more convolutional layers. However, Heifets et al. discloses machine learning model comprises a neural network with plurality of convolutional layers (p.41, col. 1, para. 2: Structure-Based Deep-Convolutional Neural Network…; (p. 27, col. 1, last paragraph: The convolutional network comprises plurality of convolutional layers).
Regarding claim 8, Brylinski discloses the limitations of claim 6 or 7 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network comprising one or more max-pooling layers. However, Heifets discloses that the neural network comprises pooling layers (p. 37, col. 1, para. 1).
Regarding claim 9, Brylinski discloses the limitations of claims 6, 7, or 8 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network comprising a steerable three-dimensional convolutional neural network. However, Heifets discloses that the neural network comprises a steerable three-dimensional convolutional neural network (p. 36, col. 2, L. 30-46: the neural network is configured to develop three-dimensional convolutional layers. …Biological activity maybe invariant under rotation, as well as translation, so the network may be optionally configured to generate rotated feature maps that take advantage of the rotational symmetries of space partitioning. For example, if the system was configured to use cubes to partition the input data, the system could be configured to generate rotated feature maps by tying the weights of the function computations together after a 90-degree rotation; also, the entire document (according to the disclosed specification [0061]: steerable convolutional blocks are used to ensure that projected binding sites are rotationally invariant as well as translationally invariant).
Regarding claim 10, Brylinski discloses the limitations of claims 6 or 9 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network comprising a deep learning neural network. However, Heifets discloses that the neural network comprises a deep learning neural network (p. 38, col. 2, L. 31-37: a deep neural network is implemented; also, the entire document).
Regarding claim 11, Brylinski discloses the limitations of any preceding claims as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network performing the comparison comprises minimizing a loss function. However, Heifets discloses a neural network that performs a comparison by minimizing a loss function (p. 39, col. 2, neural network classification of a plurality of training objects is compared to the binding data … L. 33: the neural network may optionally, tune the weights within the network to potentially minimize the error between the neural network's predicted binding affinities and/or categorizations and the training data's reported binding affinities and/or categorizations; also, the entire document).
Regarding claim 12, Brylinski discloses the limitations of claim 11 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network updating the model by backpropagation. However, Heifets discloses a neural network that the machine learning model comprises a neural network updating the model by performing back propagation using the minimized loss function (p. 39, col. 2, … error is back-propagated through the neural network; also, the entire document).
Regarding claim 13, Brylinski discloses the limitations of claims 11 or 12 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network performing a loss function and the loss function comprises a contrastive loss. However, Heifets discloses a neural network that the machine learning model comprises a neural network performing a loss function and the loss function comprises a contrastive loss representing a loss between the similarity indicator and the label. (p. 39, col. 2, … using the contrastive divergence algorithm; also, the entire document).
Regarding claim 14, Brylinski discloses the limitations of claims 11 or 12 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a neural network performing a loss function and the loss function comprises a triplet loss. However, Heifets discloses a neural network that the machine learning model comprises a neural network performing a loss function and the loss function comprises a triplet loss based on a pair of binding sites, a reference binding site and the label. (p. 32, col. 2, last paragraph-p. 33, col. 1, para. 1-3: … a structural protein-ligand interaction fingerprint (SPLIF) score … A SPLIF implicitly 65 encodes all possible interaction types that may occur between interacting fragments of the test (or training) object and the target object; also, the entire document).
Regarding claim 15, Brylinski discloses the limitations of any preceding claims as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises jittering the binding site. However, Heifets discloses a neural network that the machine learning model that comprises jittering the binding sites in input space (p. 41, col. 1, para. 2: Input Presentation: … multiple poses within the binding site cavity are sampled; also, the entire document).
Regarding claim 16, Brylinski discloses the limitations of any preceding claims as stated above. Brylinski further discloses a label comprising an indication of ligand binding similarity between the first binding site and the second binding site (p. 6, col. 2, para. 3: “We follow a similar 6-fold cross-validation protocol (label) as described above for assessing the inter-residue distance prediction. Machine learning for the estimation of pocket similarity is implemented using the Support Vector Machines algorithm for classification problems provided by libSVM 3.14; it is considered implicitly disclosed that for said cross- validation a label (the known pocket similarity score) is used). Brylinski does not disclose that the label comprises a binary value indicating whether the first and second binding sites bind structurally similar ligands. However, Heifets discloses a binary value indicating whether the first and second binding sites bind structurally similar ligands. (p. 32, col. 1, first paragraph: the characteristic of the atom is encoded in the voxel as a binary categorical variable … one channel within each voxel may represent carbon whereas another channel within each voxel may represent oxygen. When a given atom type is found in the three-dimensional grid element corresponding to 10 a given voxel, the channel for that atom type within the given voxel is assigned a first value of the binary categorical variable, such as "1", and when the atom type is not found in the three-dimensional grid element corresponding to the given voxel, the channel for that atom type is assigned a second 15 value of the binary categorical variable, such as "0" within the given voxel; also, the entire document; also, the entire document).
Regarding claim 17, Brylinski discloses the limitations according to any one of claims 6 to 10 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a convolutional neural network. However, Heifets discloses a neural network model obtained from a computer implemented method according to any one of claims 6 to 10 (Figure 1, obtaining a model is implicitly shown: after the training 66, a model is obtained and the obtained model is tested in 70; also, the entire document).
Regarding claim 18, Brylinski discloses the limitations according to any one of claims 6 to 10 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a convolutional neural network. However, Heifets discloses a computer-implemented method of using a neural network model, wherein the neural network model is obtained from a computer implemented method according to any one of claims 6 to 10, the method of using the neural network model comprising: inputting to the neural network model respective representations of third and fourth binding sites; and using the neural network model to output a ligand binding similarity indicator (Figure 1, Test object Evaluation Library 70; claim 1, test object and target object; also, the entire document).
Regarding claim 19, Brylinski discloses the limitations of claim 18 as stated above. Brylinski further discloses a computer-implemented method of training a machine learning model (abstract: eMatchSite a software/ a new method for constructing sequence order-independent alignments of ligand binding sites in protein models, ; also, the entire document) to learn ligand binding similarities between protein binding sites (p. 4 , col.1, Implementation of eMatch Site: eMatchSite is capable to estimate pairwise Ca-Ca distances between binding residues upon the alignment of two pockets using machine learning and a set of seven residue-level scores; P. 6-7, section: Pocket Similarity Score, discloses the machine learning method to calculate a similarity score; also, the entire document). Brylinski does not disclose that the machine learning model comprises a convolutional neural network. However, Heifets discloses a computer-implemented method of claim 18 as discussed above wherein the ligand binding similarity indicator comprises an indication of whether the first and second binding sites are likely to bind structurally similar ligands (Figure 2E, Using the plurality of scores to characterize the test object 278; claim 1, (D), (E), and (f); also, the entire document).
Regarding claim 20, Brylinski discloses an apparatus comprising a processor, a memory unit and a communication interface, wherein the processor is connected to the memory unit and the communication interface, wherein the processor and memory are configured to implement the computer-implemented method according to any one of claims 1 to 16, 18 or 19 (claims 1-29, Figure 1-2F; p. 28, para. 3: a computer system 100 for classification of a test object 72 (or training object) using spatial data; a general processor 74, general memory 90/92; also, the entire document).
Regarding claim 21, Brylinski discloses a computer-readable medium comprising data or instruction code representative of a machine learning model generated according to the method of any one of claims 1 to 16, 18 or 19, which when executed on a processor causes the processor to implement the machine learning model (Claim 1: the general memory storing at least one program for execution by the at least one general processor, the at least one program comprising instructions for …).
Regarding claim 22, Brylinski discloses a computer-readable medium comprising data or instruction code which, when executed on a processor, causes the processor to implement the computer-implemented method of any of claims 1 to 16, 18 or 19. (Claim 1: the general memory storing at least one program for execution by the at least one general processor, the at least one program comprising instructions for …).
Regarding claim 23, Heifets discloses a systems and methods for test object classification are provided in which the test object is modeled with a target object in a plurality of different poses to form voxel maps. obtaining spatial coordinates for a target object wherein the target object is a polymer and the spatial coordinates are a set of three-dimensional coordinates for a crystal structure of the polymer (claim 10). The voxel maps are vectorized and sequentially fed into a convolutional neural network. In some embodiments, a structural protein-ligand interaction fingerprint (SPLIF) score is generated for each pose of a given test object (or training object) to a target object and this SPLIF score is used as additional input into the underlying neural network or is individually encoded in the voxel map (col. 16, para. 2). The convolutional neural network comprises an input layer, a plurality of individually weighted, sequentially connected convolutional layers, and an output scorer (Summary) the scorer comprises a clustering algorithm, a nearest neighbor analysis (claim 28).
Heifets further discloses that the computational prediction of off-target effects can be used to identify compounds that could be used to treat alternative diseases (col. 41, para. 1).
Regarding claim 25, Heifets discloses that biological activity may be invariant under rotation, as well as translation, so the network may be optionally configured to generate rotated feature maps that take advantage of the rotational symmetries of space partitioning (col. 24, para.3).
Regarding claim 27, Heifets discloses that biological activity may be invariant under rotation, as well as translation, so the network may be optionally configured to generate rotated feature maps that take advantage of the rotational symmetries of space partitioning (col. 24, para.3). Heifets further discloses that the neural network is a 3D convolutional neural network (col. 34, para. 1); reading on limitations of the neural network is a three-dimensional convolutional neural network (3D CNN), comprising one or more steerable convolutional blocks to ensure that projected binding sites are rotationally invariant as well as translationally invariant.
Regarding claim 28, Heifets discloses that the grid spacing is a three-dimensional fixed grid spacing of 1 Å to form corresponding voxels of a voxel map (col. 20, last para.). Heifets further discloses that the spacing between the dimensional points may be randomly chosen or may be moved closer together, or farther (col. 20, para. 5). Heifets further discloses that multiple poses within the binding site cavity are sampled (p. 41, col. 1, para. 2). Heifets further discloses that the convolutional neural network is configured to adapt for dynamic systems, such as the alternative positions that may be encountered as both the target object and the test object move (col. 24, para. 1). Heifets further discloses that the target object is a polymer and the spatial coordinates are an ensemble of ten or more, twenty or more or thirty or more three-dimensional coordinates for the polymer determined by nuclear magnetic resonance where the ensemble has a backbone RMSD of 1.0 Å or better, 0.9 Å or better, 0.8 Å or better, 0.7 Å or better, 0.6 Å or better, 0.5 Å or better, 0.4 Å or better, 0.3 Å or better, or 0.2 Å or better (col. 9, last para.).
In KSR Int 'l v. Teleflex, the Supreme Court, in rejecting the rigid application of the teaching, suggestion, and motivation test by the Federal Circuit, indicated that “The principles underlying [earlier] cases are instructive when the question is whether a patent claiming the combination of elements of prior art is obvious. When a work is available in one field of endeavor, design incentives and other market forces can prompt variations of it, either in the same field or a different one. If a person of ordinary skill can implement a predictable variation, § 103 likely bars its patentability.” KSR Int'l v. Teleflex lnc., 127 S. Ct. 1727, 1740 (2007).
Applying the KSR standard to Brylinski and Heifets, the examiner concludes that the combination of Brylinski and Heifets represents applying a known technique to a known method. Both Brylinski and Heifets are directed to classifying molecules/complexes/polymers. Brylinski contained the base machine learning method of binding site similarity indicator that input presentation of first and second binding sites into the model. In the same field of research, Heifets provided the specific presentation of the 3D structures/voxels of two structures/complexes/polymers to compare their shape, volume, and chemical properties. Combining the ligand similarity of Brylinski with 3D voxelization technique of Heifets would have allowed for identification of similarly structured binding site. One ordinary skilled in the art before he effective filing data of the claimed invention would have had a reasonable expectation of success at combining the method of Brylinski and Heifets. This combination would have been expected to have provided a more accurate identification of similarly structured binding site and downstream drug discovery. Therefore, the invention would have been prima facie obvious to one of skill in the art before the effective filing date of the claimed invention, absent evidence to the contrary.
Response to Applicants Arguments
Applicant in Remarks submitted 12/01/2025 states:
First, Heifets does not disclose representations of first and second protein binding sites that each comprise an encoded three-dimensional grid of voxels, each voxel being associated with an occupancy value indicating whether an atom is present as recited in claim 1 as amended.
Instead, Heifets discloses generating voxel maps of different poses between test objects docked onto target objects. Heifets states that "test objects 72 and/or training objects 66 are modeled with the target objects 58 in each pose of a plurality of different poses."25 Heifets states that "[i]n some embodiments, the target object 58 is a polymer with an active site, the test object (or training object) is a chemical compound, and the modeling comprises docking the test object into the active site of the polymer (216)."26 Heifets then goes on to state that "after generation of each of the poses for each of the target and/or test objects, a voxel map 40 is created of each pose."27
Thus, at best Heifets discloses generating voxel maps of the final docking state between a test training object(s) docked onto a target object. However, Heifets is silent on generating voxel grid representations of first and second protein binding sites themselves prior to docking (e.g., binding) with a test object (e.g., a ligand). As will be understood by one of ordinary skill in the art, binding between a protein and a ligand may be accompanied by conformational changes in the protein binding sites themselves.
Second, modifying Brylinski to include Heifets's neural network and volumetric
information to create vectors to be used in the neural network would change the principle of operation of Brylinski and is thus impermissible under MPEP § 2143.01(VI).29 As best understood, the Office is proposing modifying Brylinski's machine learning model to include
25 Heifets at Column 11, lines 35-37 (emphasis in original).26Id. at Column 11, lines 37-41 (emphasis in original).
It is respectfully submitted that this is not persuasive. As stated above, the combination of Brylinski and Heifets represents applying a known technique of 3D voxelization of two structures/compounds/polymers to a known method of using machine learning model to indicate binding site similarity. Brylinski contained the base machine learning method of binding site similarity indicator that inputs presentation of first and second binding sites into the model upon which the 3d voxelization of structures of Heifets in a neural network architecture can be seen as an improvement to provide a more accurate binding site identification. One ordinary skilled in the art would have been motivated to use 3D voxelization in a NN of Heifets as oppose to the computationally expensive SOI alignment in a SVM of Brylinski for screening and binding site identification, since 3D voxelization is faster and screens and ranks many sites in a relatively short period of time.
Conclusion
No claims are allowed.
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 GHAZAL SABOUR whose telephone number is (703)756-1289. The examiner can normally be reached M-F 7:30-5:00.
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, Larry D. Riggs can be reached at (571) 270-3062. 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.
/G.S./Examiner, Art Unit 1686
/LARRY D RIGGS II/Supervisory Patent Examiner, Art Unit 1686