DETAILED ACTION
Applicant's response, filed 25 June 2026, has been fully considered. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
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 .
Claim Status
Claims 1-20 are pending.
Claims 1-20 are rejected.
Priority
This application is a CON of PCT/US2021/025921, filed 04/06/2021, which claims benefit of application no. 63/008,196, filed 04/10/2020. The instant application has the effective filing date of 10 April 2020.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 02/15/2023 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement has been considered by the examiner.
Drawings
The drawings, submitted on 09/19/2022, are accepted by the examiner.
Claim Rejections - 35 USC § 112
Applicant’s arguments, see page 1, para. 3-4, with respect to U.S.C § 112(b) have been fully considered and are persuasive.
The rejection to claims 6-8 is withdrawn, in view of claim amendments.
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 1-20 are rejected under U.S.C 101 because the claimed invention is directed to abstract ideas without significantly more, as detailed in the analysis below.
Eligibility Step 1: Subject matter eligibility evaluation in accordance with MPEP § 2106:
Claims 1-18 are directed to a statutory category (method).
Claim 19 is directed to a statutory category (system).
Claim 20 is directed to a statutory category (product).
Therefore, in accordance with MPEP § 2106.03 all claims have patent eligible subject matter.
[Eligibility Step 1: YES]
Eligibility Step 2A: This step determines whether a claim is directed to a judicial exception in accordance with MPEP § 2106.
Eligibility Step 2A -- Prong One: Limitations are analyzed to determine if the claims recite any concepts that could equate to a judicial exception (i.e. abstract idea, law of nature, or natural phenomenon). Possible judicial exceptions are explored below.
Recitations of Judicial Exceptions:
Claims 1, 19, 20:
initializing a population of candidate solutions, wherein each of the candidate solutions is defined by a set of properties that include: an indication that a particular type of pre-processing is to be performed; an indication that a particular type of machine learning model is to be used; (mental process)
identification of a type of machine-learning model that is to be used; and/or a machine-learning model hyperparameter; (mental process)
filtering the population of candidate solutions by: determining, for each of the candidate solutions and for each of the data elements, a predicted sample characteristic by processing the spectrum of the data element with the set of properties, wherein processing the spectrum of the data element comprises pre-processing the spectrum using the particular type of pre-processing (mathematical concept)
selecting an incomplete subset of the population of candidate solutions based on an assessment of the predicted sample characteristic and known characteristic of the sample in the data set for each of the candidate solutions (mental process)
performing one or more additional generation iterations by: updating the population of candidate solutions to include a next-generation population of solutions identified using the incomplete subset of the population of candidate solutions and one or more genetic operators; (mental process)
repeating the filtering of the population of candidate solutions using the updated population of candidate solutions; (mathematical concept)
generating a processing pipeline based on the set of properties of a particular candidate solution in the incomplete subset of the population of candidate solutions selected during a last generation iteration of the one or more additional generation iterations. (mental process)
Claim 2: generating a predicted characteristic of the other sample by processing the other spectrum in accordance with the processing pipeline (mathematical concept)
Claim 4: wherein the set of properties for the particular candidate solution includes a hyperparameter for a particular type of machine- learning model, the particular type of machine-learning model including: partial least squares; random forest; or support vector machine (mathematical concept)
Claim 5: wherein the set of properties for the particular candidate solution includes a selection of or a hyperparameter for a particular type of machine-learning model, the particular type of machine-learning model being configured to generate classification outputs or numeric outputs. (mathematical concept)
Claim 8: wherein the predicted characteristic of the other sample characterizes: a concentration of one or more small-molecule analytes; a solvent; a prevalence of one or more protein variants; a protein higher-order structure; or large molecule impurities. (mathematical concept)
Claim 9: wherein the processing pipeline includes performing an asymmetric least squares technique to reduce or remove a baseline, and wherein the set of properties for the particular candidate solution includes at least one parameter for the asymmetric least squares technique (mathematical concept)
Claim 10: wherein the processing pipeline includes performing an smoothing technique to reduce or remove a baseline, and wherein the set of properties for the particular candidate solution includes at least one parameter for the smoothing technique. (mathematical concept)
Claim 11: wherein, for at least one sample of the plurality of samples, the plurality of data elements includes multiple data elements corresponding to the sample, the multiple data elements including different replicate spectrum generated using the sample. (mathematical concept)
Claim 12: partitioning the plurality of data elements into a training subset of the plurality of data elements and a testing subset of the plurality of data elements; (mental process)
wherein the at least some of the plurality of data elements for which the predicted sample characteristics are determined are defined as the testing subset of the plurality of data elements; (mental process)
and wherein filtering the population of candidate solutions further includes: learning one or more parameters using the testing subset of the plurality of data elements. (mental process)
Claim 13: wherein each of the plurality of samples corresponds to a same target chemical structure and to a same target formulation, wherein the plurality of samples includes multiple lot-specific subsets, each of the multiple lot-specific subsets including multiple samples manufactured during an individual lot, (mental process)
and wherein the partitioning of the plurality of data elements includes: partitioning the individual lots into the training subset and the testing subset; (mental process)
and partitioning the plurality of data elements based on the lot partitioning. (mental process)
Claim 14:
generating a predicted characteristic of the other sample by processing the other spectrum with the processing pipeline; (mathematical concept)
determining, based on the predicted characteristic, whether a quality-control condition is satisfied; (mental process)
when the quality control condition is satisfied, distributing the other sample to be administered to a subject; and when the quality control condition is not satisfied, inhibiting distribution of the other sample for subject administration (mental process).
Claim 15: when the quality control condition is not satisfied, dynamically adjusting one or more parameters associated with production of the other sample (mental process)
Claim 16: performing a feature-selection process that selects, from a set of intensities of the spectrum, one or more intensities for use in generating the predicted characteristic of the predicted sample, wherein the feature-selection processing is performed prior to generation of the predicted characteristic by the processing pipeline. (mental process)
Claim 17: wherein the feature-selection process includes: identifying, from the spectrum, a set of wavenumbers, each wavenumber being associated with an intensity value; defining a score for each wavenumber of the set of wavenumbers using a regression analysis; (mental process, mathematical concept)
sorting the set of wavenumbers according to the score of each wavenumber of the set of wavenumbers; (mental process)
performing one or more feature-selection iterations, wherein each feature-selection iteration includes: generating a subset of the set of wavenumbers by removing one or more wavenumbers of the spectrum having a lowest score; (mental process)
generating a model-validation score based on a cross-validation of the subset of the set of wavenumbers on the machine-learning model; (mathematical concept)
selecting, from the one or more feature-selection iterations, a particular feature- selection iteration of the one or more feature-selection iterations that includes a model-validation score that is closest to a threshold; (mental process)
selecting, for use in generating the predicted characteristic by the processing pipeline, intensities that correspond to the subset of the set of wavenumbers of the particular feature-selection iteration. (mental process)
Claim 18: generating a predicted characteristic of the other sample by processing the other spectrum in accordance with the processing pipeline; (mathematical concept)
determining, based on the predicted characteristic, whether a quality-control condition is satisfied; (mental process)
when the quality control condition is satisfied, initiating or completing one or more a manufacture process configured to manufacture additional samples; and when the quality control condition is not satisfied, terminating or modifying the one or manufacture process (mental process).
Step 2A- Prong One Analysis:
Selecting, sorting, and making determinations of data represent analysis techniques that require no more than mental observations and pen/paper. As such, limitations that recite these techniques fall into the mental process grouping of abstract ideas.
Generating secondary data based on machine learning algorithms (random forest, SVM, PLS) and mathematical calculations (asymmetrical least squares, smoothing, cross-validation, fitness metrics) represent analysis techniques that require transformations and organizations on data via mathematical formulas and thus fall under the mathematical concepts grouping of abstract ideas.
Eligibility Step 2A – Prong Two: A claim that integrates a judicial exception into a practical application will apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception. If the claim contains no additional claim elements beyond the abstract idea, the claim fails to integrate the abstract idea into a practical application (MPEP 2106.04(d)).
Eligibility Step 2B: Claim elements are probed for inventive concept equating to significantly more than the judicial exception (MPEP 2106.04(II)).
Additional Elements within the claimed invention include:
Claims 1, 19, and 20: inputting the pre-processed spectrum to the particular type of the machine learning model to generate the predicted sample characteristic; accessing a data set including a plurality of data elements, each of the data elements including: a spectrum generated based on an interaction between one sample of a plurality of samples and energy from an energy source; and a known characteristic of the sample
Claim 2: accessing another spectrum corresponding to another sample; outputting the predicted characteristic of the other sample.
Claim 3: wherein, for each data element of the plurality of data elements, the spectrum includes a Raman spectrum or an infrared spectrum.
Claim 6: wherein the other sample includes large molecules.
Claim 7: wherein the other sample includes small molecules.
Claims 14: accessing another spectrum corresponding to another sample
Claim 18: receiving the predicted characteristic; accessing another spectrum corresponding to another sample
The limitations above are directed to inputting, accessing, receiving, specifying, and outputting data necessary to complete the method of the claimed invention. As such, they recite mere data gathering actives that qualify as insignificant extra solution activities that do not separately, or as a whole claimed invention integrate the judicial elements into practical application per MPEP 2106.05(g).
[Eligibility Step 2A – Prong Two: YES]
The insignificant extra solution data gathering activities, as recited are also found to be well-understood, routine, and conventional per Mayo, 566 U.S. at 79, 101 USPQ2d at 1968; OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1092-93 (Fed. Cir. 2015), and Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016) for selecting information, based on types of information and availability of information.
[Eligibility Step 2B: NO]
Additional Elements that may be categorized differently include:
Claim 1: computer-implemented method
Claim 19: A system comprising: one or more data processors; and a non-transitory computer readable storage medium containing instructions which, when executed on the one or more data processors, cause the one or more data processors
Claim 20: computer-program product tangibly embodied in a non-transitory machine-readable storage medium, including instructions configured to cause data processors
The additional elements above represent generic computer components or implementations. When viewed separately or in the context of the whole claimed invention, they merely act as tools to carry out the judicial exceptions. Elements of this category do not integrate judicial exceptions into practical application per MPEP 2106.05(f).
[Eligibility Step 2A – Prong Two: YES]
The elements are further found to be well-understood, routine, and conventional per Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015), for storing and retrieving information in memory; and Bancorp Services v. Sun Life, 687 F.3d 1266, 1278, 103 USPQ2d 1425, 1433 (Fed. Cir. 2012), for performing repetitive calculations, Koljonen et al. (IDS reference; NPL; cite no. 24; 2008), which reviews genetic algorithms in near infrared spectroscopy and chemometrics; and Priya et al. (Renewable Sustainable Energy Reviews; Vol. 93; 2018), which reviews parameter optimization techniques, such as initializing and filtering candidate solutions.
[Eligibility Step 2B: NO]
As such, claims 1-20 are directed to judicial exceptions without significantly more and are rejected under 35 U.S.C 101.
Response to Arguments
Applicant argues the filtering step of claim 1 does not recite an abstract idea in the form of a mental process or mathematical process as the amended limitation of “pre-processing the spectrum using the particular type of pre-processing, and inputting the pre-processed spectrum to the particular type of machine learning model…” does not recite mental processes as it represents executing a multi-layered software architecture on physical sensor data (page 3, para. 1).
Examiner responds spectral pre-processing steps, using broadest reasonable interpretation, and in light of the specification [0056], are not limited to physically modifying sensor data and can include steps such as removing baselines, scaling the spectrum and/or smoothing the spectrum [0056]. Such steps qualify as mathematical concepts as indicated in the 101 recited herein per Digitech Image Techs., LLC v. Electronics for Imaging, Inc., 758 F.3d 1344, 1350, 111 USPQ2d 1717, 1721 (Fed. Cir. 2014) for organizing and manipulating information through mathematical correlations.
Examiner agrees that the amended limitation of inputting the pre-processed data does not recite a judicial exception as indicated in the 101 analysis herein.
Applicant argues the limitation of “selecting an incomplete subset” cannot practically be performed in the human mind because of the amended limitation of requiring it to be based on “an assessment of the predicted sample and known characteristic,” as this represents a comparative computational evaluation linking the previously generated physical data with the known ground truth data (page 3, para. 2); and due to the volume of data points to be selected (page 3, para. 2).
Examiner responds though the claims represent a “computational evaluation,” the courts do not distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer per MPEP 2106.04(a)(2)(III); Versata Dev. Group v. SAP Am., Inc., 793 F.3d 1306, 1335, 115 USPQ2d 1681, 1702 (Fed. Cir. 2015); and see also Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1318, 120 USPQ2d 1353, 1360 (Fed. Cir. 2016). Furthermore, the volume of data is not limited to an amount that cannot be practically performed in the human mind. Therefore, the limitation is still drawn to a mental process.
Applicant argues though the step of “performing one or more additional generation iterations” may involve math, it does not recite mathematical concepts itself as it is directed to processes such as algorithmically swapping out a pre-processing technique or mutating a machine-learning parameter (page 4, para. 1).
Examiner responds that the argument is found persuasive that the step does not particularly recite mathematical concepts, however swapping parameters and algorithms to be performed can be practically performed within the human mind and are thus still drawn to judicial exceptions in the form of abstract ideas (mental processes).
Applicant argues the step of “generating a processing pipeline” does not recite math and rather inherently requires the instantiation of an executable software architecture based on the surviving co-dependent set of properties (page 4, para. 2).
Examiner responds, this particular argument is found persuasive as generating a processing pipeline in the manner instantly described does not recite math, however upon further consideration, the step appears to read as a mental process under broadest reasonable interpretation to include steps commensurate in scope with organizing an application program interface for extracting and processing information from a diversity of types of hard copy documents per Content Extraction, 776 F.3d at 1345, 113 USPQ2d at 1356.
Applicant argues the claim 1 integrates any judicial exceptions into practical application through an improvement to technology via the “initialization of a programmatic population of candidate solutions where each solution dictates both a pre-processing and machine learning model type” (page 6, para. 1-2) with support for the improvement in [0004], [0005], and [0145]-[0278].
Examiner responds the specification or arguments fail to elaborate how the particular element of initializing a population of candidate solutions replaces the “subjective, manual trial-and-error with an automated, co- optimized pipeline architecture that empirically reduces predictive error,” rather than the judicial exceptions providing said improvement.
Applicant argues by “selecting an incomplete subset to update the population for the next generation” reduces computational demands of a brute force processing system that would process ever possible permutation of data, therefore reflecting an improvement to computer function as a whole (page 7, para. 1).
Examiner responds selecting an incomplete subset of solutions is classified as a judicial exception, in the form of an abstract idea (mental process); and MPEP 2106.05(a) recites: it is important to note, the judicial exception alone cannot provide the improvement. See the discussion of Diamond v. Diehr, 450 U.S. 175, 187 and 191-92, 209 USPQ 1, 10 (1981)) in subsection II, below
Applicant argues the claimed method provides a specific, structural improvement to the computer's memory and processing efficiency, which is directly analogous to the patent-eligible improvement recognized by the Federal Circuit in Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016) (page 7, para. 2).
Examiner responds the claimed method does not follow the same fact pattern of Enfish, LLC v. Microsoft Corp., 822 F.3d 1327 (Fed. Cir. 2016); and instead gathers and analyzes information using conventional techniques, consistent with TLI Communications, 823 F.3d at 612-13, 118 USPQ2d at 1747-48, as evidenced by the reviews of Koljonen et al. (IDS reference; NPL; cite no. 24; 2008) and Priya et al. (Renewable Sustainable Energy Reviews; Vol. 93; 2018) which teach the structural arrangement described.
Applicant argues the steps of “initializing a population of candidate solutions” (page 8, para. 2) and filtering them (page 8, para. 3) are not conventional; and even if the steps themselves are found conventional, the ordered combination of additional elements provide significantly more in a technical framework analogous to Bascom Global Internet Services, Inc. v. AT&T Mobility LLC, 827 F.3d 1291 (Fed. Cir. 2016), where the court found that an inventive concept can reside in the "non-conventional and non-generic arrangement of many of the known, conventional pieces." (page 9, para. 1).
Examiner responds the concept of initializing a population of candidate solutions…, and then filtering them by an evaluation or assessment of their fitness is a conventional arrangement as per Priya et al. (Renewable and Sustainable Energy Reviews; Vol. 93; 2018.), which recites an initialization phase (page 8, column 1), where every individual of population size represent a potential solution to the problem in a dimensional search space (page 8 column 1); and solutions with higher fitness values are retained and passed for further iterations. (page 9, column 1).
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 1-5, 7, 8, 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Koljonen et al. (IDS reference; NPL; cite no. 24; 2008) in view of Jarvis et al. (Bioinformatics; Vol. 21:7; 2005) and in further view of Costello et al. (npj Systems Biology and Applications; Vol. 4: 19; 2018).
Koljonen et al. reviews genetic algorithms in Near Infrared spectroscopy (NIR) and chemometrics.
Claims 1, 19, and 20 are directed to computer implemented method, systems, and computer readable mediums that perform the following steps.
access a dataset with spectrum, generated based on the interaction between a sample and its energy source; and a known characteristic of the sample;
initialize candidate solutions properties that include: indications that a particular type of pre-processing is to be performed; and an indication that a particular type of machine-learning model is to be used;
pre-process the spectrum using the particular type of pre-processing; and input the pre-processed spectrum into the machine learning model to predict a characteristic of the sample, which processes the spectrum;
use the filtered results and at least one genetic operator to generate a next generation of solutions;
filter the updated population of solutions, with the same method; to
generate a processing pipeline based on the solutions selected during the latest updated generation.
Koljonen et al. teaches that 18 wavelength variables of an NIR spectrum are (page 190, column 2) encoded using a binary alphabet (page 191, column 2).
Koljonen et al. further teaches that in a typical Genetic Algorithm (GA), shown below (page 190, fig. 1), after (i) creating an initial population, (ii) evaluating+ the fitness of each individual in the population (page 190, column 1), and (iii) checking the stopping condition, (iv) a new generation is generated (page 190, column 2) based off the findings (page 190, fig. 1).
PNG
media_image1.png
233
722
media_image1.png
Greyscale
Koljonen et al. further teaches that such GA’s are well suited for multi-criteria optimization, with properties such as wavelength inclusion, pre-processing steps, the number of latent variables in the model, and the regression (or classifier) model itself (page 195, column 1); and in this example, the machine learning model predicts the concentration of functional groups in a sample (page 191, column 1).
Koljonen et al. further teaches using a fitness function between the predicted and true organic compound sample concentration (page 191, column 1); a selection of parents, crossover, and mutation operators (page 190, column 2); and repeating steps ii to iv until the stopping condition is fulfilled (page 190, column 2).
Therefore Koljonen et al. teaches filtering the updated population of solutions using an assessment of the predicted and known sample characteristic via the fitness function.
Claim 2 is directed to accessing, generating, and outputting a predicted characteristic of another sample with the processing pipeline and spectrum data.
Claim 7 is directed to the sample including small molecules; and Claim 8 is directed to predicting one of the following characteristics: concentration of one or more small-molecule analytes; a solvent; a prevalence of one or more protein variants; a protein higher-order structure; or large molecule impurity.
Koljonen et al. teaches that another researcher compared GA and Partial Least Squares-bootstrap models from NIR spectrum data to analyze its predictive ability (page 192, column 1), specifically in regards to clavulanic acid concentration (page 192, column 1).
Regarding claim 3, Koljonen et al. teaches that Near Infrared Spectroscopy data sets were examined (page 193, column 1).
Regarding claim 4, Koljonen et al. teaches using the Partial Least Squares (PLS) model and denoting the number of components to be used within it (page 191, column 1).
Regarding claim 5, Koljonen et al. teaches the results from three NIR data sets showed comparable classification accuracy using Partial Least Squares (page 193, column 1).
Koljonen et al. does not explicitly teach the amended limitation of processes the spectrum by pre-processing the spectrum using the particular type of pre-processing; and inputting the pre-processed spectrum into the machine learning model to predict a characteristic of the sample (claims 1, 19, and 20).
Jarvis et al. describes genetic algorithm optimization for pre-processing and variable selection of spectroscopic data.
Jarvis et al. teaches most, if not all, calibration or classification studies of biological samples conducted with vibrational spectroscopic techniques use some form of data pre-processing prior to input into an appropriate mathematical model to compensate for experimental and machine variability (page 8, column 2); and supervised linear modelling techniques such as partial least squares (PLS) a popular calibration model, and discriminant function analysis (DFA) a powerful clustering algorithm, are ideally suited to the analysis of spectroscopic data (page 1, column 2), in which they can be calibrated to predict the level of the secondary metabolite gibberellic acid produced in an industrial bioprocess (page 2, column 1).
Therefore Jarvis et al. teaches a technique applicable to the method of Koljonen et al. As such, it would be obvious to one of ordinary skill in the art to apply the technique of Jarvis et al. to the method of Koljonen et al. with an expectation of predictable results and an improved system, with each element merely performing the same function as they do separately.
Koljonen et al. does not explicitly teach generating a processing pipeline based on the set of properties of a particular candidate solution in an incomplete subset of the population of candidate solutions selected during a last generation iteration of the one or more additional generation iterations (claims 1, 19, and 20).
Priya et al. provides a comprehensive review on parameter estimation techniques.
Priya et al. teaches differential evolution uses genetic operators such as mutation, crossover and selection to find optimal solution (page 7, column 2) via an initialization phase (page 8, column 1), where every individual of population size represent a potential solution to the problem in a dimensional search space (page 8 column 1); and a selection phase where solutions with higher fitness values are retained and passed for further iterations (page 9, column 1).
Priya et al. does not teach generating a processing pipeline based on the set of properties of a particular candidate solution in an incomplete subset of the population of candidate solutions selected during a last generation iteration of the one or more additional generation iterations.
Costello et al. describes a machine learning approach to predict metabolic pathway dynamics from time-series multiomics data.
Costello et al. teaches to find the best fit we used a differential evolution algorithm implemented in scipy; this global optimizer was chosen because its convergence is independent of the initial population choice and it tends to need less parameter tuning than other methods (page 12, column 1).
Costello et al. teaches a machine learning model must be selected to learn the relationship between input and outputs (Supplementary Fig. S1); TPOT uses genetic algorithms to find a model with the best cross-validated performance on the training set; the best performing models are mated to form a new population of models to test; this process is repeated for a fixed number of generations and the best performing model is returned to the user (page 11, column 1).
Costello et al. teaches after TPOT determines the optimal models associated with each metabolite, they are trained on the data set of interest and are ready for use to solve Eqs. 3 and 4 (page 11, column 2); once the models are trained, we can use them to predict metabolite concentrations by solving the following initial value problem using the same function f that was learned in Eqs. (1) and (2) by integrating the system forward in time numerically (page 11, column 2).
Costello et al. teaches new supervised learning techniques can be added to this approach by adding them to the scikit-learn library (page 9, column 1); and TPOT will automatically test them and use them if they provide more accurate predictions than the techniques used here (page 9, column 1).
Costello et al. further teaches the current approach uses tree-based pipeline optimization tool (TPOT) to combine, through genetic algorithms, 11 different machine learning regressors and 18 different preprocessing (feature selection) algorithms (page 9, column 1); and while the machine learning approach necessitates more data, it can be automatically applied to any pathway or host, leverages systematically new data sets to improve accuracy, and captures dynamic relationships which are unknown by the literature or have a different dynamic form than Michaelis–Menten kinetics (page 2, fig. 1).
Therefore Costello et al. in view of Priya et al. teach a method that initializes a set of candidate solutions include machine learning algorithms and preprocessing techniques, filters then into an incomplete subset over generations, and automatically trains and uses them within a metabolite prediction processing pipeline by integrating the system forward in time. As the method of Costello et al. further uses the genetic algorithm as an integral apart of its differential evolution approach, explained by Priya et al., it is applicable to the method of Koljonen et al. As such, the technique of Costello et al. in view of Priya et al. can be reasonably applied to the Koljonen et al.’s base method using the genetic algorithm for spectrum processing and characteristic prediction with a reasonable expectation of success and an improved system.
Claims 9 and 10 are rejected under 35 U.S.C. 103 as being unpatentable over Koljonen et al. (IDS reference; NPL; cite no. 24; 2008), in view of Jarvis et al. (Bioinformatics; Vol. 21:7; 2005) and Costello et al. (npj Systems Biology and Applications; Vol. 4: 19; 2018), as applied to claim 1-5, 7, 8, 19-20 above, and in further view of Liu et al. (IDS reference; NPL; cite no. 28; 2017).
Koljonen et al. teaches a genetic algorithm framework of predicting sample characteristics with machine learning and NIR spectral data.
Claims 9 and 10 are directed to the processing pipeline performing an asymmetric least squares (claim 9) or smoothing (claim 10) technique to reduce or remove a baseline; and the set of candidate solution properties including a parameter for the technique.
Koljonen et al. teaches utilizing the tendency of genetic algorithms to get stuck in local optima… to perform a type of noise filtering (page 192, column 1).
Koljonen et al. does not teach noise filtering is accomplished with a specific technique.
Liu et al. describes differences between deep convolutional neural networks and conventional machine learning techniques for Raman spectrum recognition.
Liu et al. teaches that conventional machine learning methods such as SVM and Random Forest are not capable of handling Raman signals which are not properly baseline corrected, and therefore require explicit baseline correction in their processing pipelines (page 9, column 1).
Liu et al. further teaches selecting another dataset which contains raw (uncorrected) spectra and six widely-used baseline correction methods, such as: modified polynomial fitting, rubber band, robust local regression estimation, iterative restricted least squares, asymmetric least square smoothing, and rolling ball. (page 10, column 1).
Liu et al. further teaches the difference between raw spectra and corresponding spectra, baseline corrected by asymmetric least squares, specifically, in Figure 4 (page 10, column 1).
Therefore, Koljonen et al. teaches that noise filtering is a known technique in genetic algorithm spectrum prediction frameworks. Liu et al. provides sufficient teachings for one of ordinary skill in the art to accomplish filtering in conventional machine learning models by including a mandatory baseline correction process parameter. Liu et al. further provides motivation for one of ordinary skill in the art to use an asymmetrical least square smoothing technique to enact baseline correction, as it is one of the most widely used techniques, specifically shown to impact spectra of conventional machine learning methods, analogous to Koljonen et al.
Claims 6, 11-13, 16, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Koljonen et al. (IDS reference; NPL; cite no. 24; 2008), in view of Jarvis et al. (Bioinformatics; Vol. 21:7; 2005) and Costello et al. (npj Systems Biology and Applications; Vol. 4: 19; 2018), as applied to claim 1-5, 7, 8, 19-20 previously, in view of Schwanninger et al. (J. Near Infrared Spectrosc; Vol. 19; 2011).
Koljonen et al. teaches a genetic algorithm framework of predicting sample characteristics with machine learning and NIR spectral data.
Claim 6 is directed to the sample including large molecules.
Koljonen et al. does not teach explicitly teach samples including large molecules.
Schwanninger et al. describes the determination of lignin content in spruce wood via Infrared spectroscopy and Partial least squares regression.
Schwanninger et al. teaches samples were taken from inner and outer [tree] rings (page 320, column 2); which had their total lignin content determined (page 320, column 2). Schwanninger et al. further teaches lignin is a major polymeric wood constituent (page 320, column 1).
Claim 11 is directed to the dataset including at least two different replicate spectrum of one or more samples.
Schwanninger et al. teaches that replicate spectra must be kept together in the dataset (page 321, column 2) and were treated as one sample (page 321, column 2).
Claim 12 is directed to splitting the data into training and testing sets; and using the testing set to determine a predicted sample characteristic and filter the solutions.
Regarding claim 12, Koljonen et al. teaches that a model is calibrated and evaluated for each trial with a training set; and the performance or fitness of the model is evaluated using an independent test set (page 192, column 1).
Therefore Koljonen et al. teaches splitting the data and using the testing set to filter the data, with the fitness function.
Koljenen et al. does not teach that the testing set is used to determine predicted sample characteristic.
Schwanninger et al. teaches when applying the PLS to determine lignin content, care should be taken that sample subsets for cross-validation (CV) and test-set (TS) validation are representative of the whole data set (page 321, column 2).
Schwanninger et al. further teaches splitting the reference data set: one half for CV (internal validation) and the other half for TS (external validation); then changing the groups; so, the one previously used for CV (CV1) was then used for TS (TS1) and the one first used for TS (TS2) served for CV (CV2) (page 321, column 2).
Therefore Schwanninger et al. uses the testing set in the prediction of sample characteristics.
Claim 13 is directed to the samples corresponding to an identical target chemical structure and formulation; including lot-specific samples, manufactured during an individual lot; and splitting the entire dataset into training and testing subsets based on the individual lots.
Schwanninger et al. teaches the samples were obtained from a field trial of 50 clones with two to five replicates grown at two sites in south Sweden (page 320, column 2); the samples were divided into two groups according to the two sites, into two groups according to ring numbers (4–6 or 11–13) (page 320, column 2); and based on the structure of softwood–lignin overtones (first and second) as well as combinations of vibrations of several groups (cH2, cH3, car–H stretching of the aromatic ring and o–H) are expected (page 325, column 2).
Claim 16 is directed to selecting at least one intensity from the spectrum data as features before predicting a characteristic of the sample.
Claim 17 is directed to selecting the intensity via the following steps: Identify a wavenumber associated with an intensity value; Use regression analysis to assign a “score” for each wavenumber; Sort the wavenumbers according to their score; Remove at least one wavenumber with the lowest score; Perform cross validation on the remaining wavenumbers via machine learning; Complete cross validation by selecting a model validation-score closest to a threshold; and select intensities that correspond to the remaining cross-validated wavenumbers as features.
Schwanninger et al. teaches associating wavenumber ranges with varying spectral intensities in a property weighting spectrum (PWS) (page 324, column 2); which, apart from a multiplicative factor, is identical to the first PLS vector (page 324, column 1). The first PLS vector (rank) was obtained by regressing the near infrared dataset against the total lignin content by means of full cross validation (page 321, column 2); and the rank with the smallest PRESS (predictive residual error sum of squares=sum of all squared differences between true and predicted values) was searched (page 321, column 2).
Schwanninger et al. further teaches selecting relevant spectral ranges meeting both high correlation coefficient and significant PWS signal criteria (page 326, column 2), with the goal of identifying a subset of wavenumbers that produce the smallest possible errors in the models for quantitative determinations; and using the removal of non-informative variables to produce better prediction and simpler models (page 325, column 1).
Therefore, Schwanninger et al. teaches using only a portion of wavenumbers after a cross-validation process and splitting lot specific sampled with identical target chemical structures and formulas into different data subsets (training, testing), based on the individual lot they were produced from. Though the samples are natural compounds with a formula instead of formulation, it is obvious for one of ordinary skill in the art to perform the same method of data partitioning for synthetically produced samples with manufactured formulations and remove non-informative variables determined from its ranking system using correlation coefficient determined thresholds.
As Koljonen et al. further teaches finding and selecting the most promising wavelet intervals (page 194, column 2), without a specific method of accomplishing the process, one of ordinary skill in the art would have sufficient motivation to apply the teachings of Schwanninger et al. with a reasonable expectation of success in selecting relevant spectral data features for Genetic Algorithm processing.
Claim 14 is rejected under 35 U.S.C. 103 as being unpatentable over Koljonen et al. (IDS reference; NPL; cite no. 24; 2008), in view of Jarvis et al. (Bioinformatics; Vol. 21:7; 2005) and Costello et al. (npj Systems Biology and Applications; Vol. 4: 19; 2018), as applied to claims 1-5, 7, 8, 19-20 previously, and in further view of Frano et al. (BW Tek; 2018).
Claim 14 is directed to accessing spectrum data of a sample; processing the data with the processing pipeline; predicting a characteristic of the sample; using the predicted characteristic to determine if a quality-control condition is met; and distributing or inhibiting sample distribution based on the condition being met or unmet, respectively.
Koljonen et al. teaches methods of using the genetic algorithm and infrared spectrum to predict characteristics of a sample via a machine learning pipeline.
Koljonen et al. does not teach using the processing pipeline to aid in a quality control decision making process.
Frano et al. describes Raman spectroscopy for at line content uniformity testing of pharmaceutical tablets.
Frano et al. teaches that Content uniformity (CU) testing is a crucial task in pharmaceutical manufacturing, as it ensures that each product that reaches a consumer contains a safe dosage of the active pharmaceutical ingredient (API) (page 1, column 1).
Frano et al. further teaches testing the prediction of the acetaminophen and lactose content using the models and collecting spectra for each sample (page 7, column 1). Frano et al. further teaches giving each model upper and lower limits, for a simple “Pass” or “Fail” result (page 5, column 1); and if the value calculated for either of the components is outside of the set lower and upper limits, the software will present a “Fail” message, signaling the user that a sample is not within the set guidelines (page 7, column 1); and if both components are found to be within the set lower and upper limits, the sample will “Pass” (page 7, column 1).
Therefore Frano et al. teaches using a spectrum processing pipeline to predict concentrations of a sample and applying the results as a quality control metric of determining if the finished pharmaceutical product is suitable to be released and distributed to consumers. As the processing pipeline taught by Frano et al. is also analogous to that taught by Koljonen et al., it is obvious for one of ordinary skill in the art to combine the techniques in order to use yield the predictable results of performing content uniformity testing to ensure distribution of a “tested and validated’ sample.
Claims 15 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Koljonen et al. IDS reference; NPL; cite no. 24; 2008) in view of Jarvis et al. (Bioinformatics; Vol. 21:7; 2005), Costello et al. (npj Systems Biology and Applications; Vol. 4: 19; 2018), and Frano et al. (BW Tek; 2018), as applied to claim 14 above, in further view of Esmonde-White et al. (Anal Bioanal Chem; Vol. 409; p.637-649; 2017).
Claim 15 is further directed to adjusting at least one production parameter of the sample, when the quality control condition is NOT satisfied.
Claim 18 is directed to further using the predicted characteristic of a sample to determine if a quality-control condition is met. If the condition is met, beginning or completing the manufacturing process of more samples; and if unmet; stopping or changing the manufacturing process.
Koljonen et al. in view of Frano et al. teach establishing pass/fail quality conditions related to the distribution of samples.
Frano et al. does not teach using the quality conditions to adjust production and manufacturing parameters of the sample.
Esmonde-White et al. describes Raman spectroscopy a process analytical technology for pharmaceutical manufacturing and bioprocessing.
Esmonde-White et al. teaches that Real-time, in-process analytics have an important role in ensuring quality product and enabling in-process corrections (page 640, column 2); and in a process description and batch sheet, a reaction of a sample would have been completed and collected at ~1250 min (page 641; column 1); but in situ Raman data showed reaction completion nearly 600 min before stipulated time (page 641, column 2). Esmonde-White et al. teaches that this the data suggests that batch cycle time could be reduced by several hours when moving up to the commercial manufacturing scale, improving process efficiency (page 641, column 2).
Therefore, Esmonde-White et al. teaches using a spectra processing pipeline analogous to that of Koljonen et al., to predict sample characteristics and adjust or change a parameter of production/manufacturing when a sample does not meet an expected quality control metric.
Though not explicitly taught, it would be obvious to one of ordinary skill in the art to adopt the pass/fail criteria taught by Frano et al. in order to complete the manufacturing process according to the batch sheet reaction time if the in-process analytics predicted a different time. As, Esmonde-White et. further teaches that the techniques can also be applied to in-line or off-line Raman measurements of content uniformity (page 642, column 1), one of ordinary skill in the art can reasonably combine the teachings with an expectation of success in using the framework for quality control relating to content uniformity and in process reactions before completing product distribution or manufacturing.
Response to Arguments
Applicant argues Koljonen et al. does not teach or suggest “initializing a population of candidate solutions, wherein each of the candidate solutions is defined by a set of properties that includes: an indication that a particular type of pre-processing is to be performed, and an indication that a particular type of machine-learning model is to be used” because the purpose of this initialized population as defined in the specification is to dynamically construct a cohesive processing pipeline; and Koljonen et al. teaches using the wavelength selection and number of components for a machine learning algorithm as a part of feature selection within a singular, static framework (page, para. 1-3).
Examiner responds, it is noted that the features upon which applicant relies (i.e., using the initialized population intended to dynamically construct and evaluate entire software architectures) are not recited in the rejected claims. Although the claims are interpreted in light of the specification, limitations or recitations of intended use from the specification, are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
Applicant argues Claim 1 further recites "filtering the population of candidate solutions by: determining, for each of the candidate solutions... a predicted sample characteristic by processing the spectrum of the data element with the set of properties, wherein processing the spectrum... comprises pre-processing the spectrum using the particular type of pre-processing, and inputting the pre-processed spectrum to the particular type of the machine-learning model, " in which the purpose of this filtering step is to actively execute the software pipeline generated during initialization. The claimed method physically transforms the raw spectrum using the specifically paired pre-processing technique and feeds it directly into the paired machine-learning model to evaluate the co-dependent efficacy of that unique combination (page 11, para. 1-3).
Examiner responds applicant’s argument has 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.
Applicant argues Koljonen et al. does not teach or suggest generating a processing pipeline based on the set of properties of a particular candidate solutions…” (page 11, para. 4).
Examiner responds though Koljonen et al. does not teach or suggest generating a processing pipeline, Koljonen et al. in view of the newly cited references teach all the components of said pipeline; and the newly cited reference (Costello et al.) provides sufficient motivation for one of ordinary skill in the art to use the genetic algorithm of Koljonen et al. to create a differential evolution approach which selects machine learning and preprocessing techniques and generates a processing pipeline to produce predictions about metabolites.
Applicant argues a person of ordinary skill in the art would not have had a reasonable expectation of success in combining the secondary references of Liu et al.’s CNN or Schwanninger’s PLS regression into the genetic algorithm of Koljonen et al.
Examiner responds Liu et al.’s teachings are directed to noise filtering techniques (previous office action, no. 78-80). Koljonen et al. teaches Nordling et al. have compared different types of Gas with different parameter values and showed that the influence of measurement noise, i.e. over-fitting, can be decreased by first generating a large pool of locally optimised solutions (page 4, column 1); their method is, therefore, utilising the tendency of Gas to get stuck in local optima, when the fitness landscape has very many local optima, to perform a type of noise filtering (page 4, column 1).
Similarly, Schwanninger et al.’s teachings are directed to wavenumber, intensity, and interval selection (previous office action, no. 98-100). Koljonen et al. teaches the most popular application for GAs in NIR spectroscopy is wavelength, or more generally speaking, variable selection (page 1, column 1).
Therefore Koljonen et al. provides sufficient teachings that performing a type of noise filtering and wavenumber and intensity (variable) selection are techniques that are common to and can be combined with the rest of its presented Genetic Algorithm techniques with each element merely performing the same function as they do separately yielding predictable results and a reasonable expectation of success.
Conclusion
No claims are currently 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.
Correspondence
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Milana Thompson whose telephone number is (571)272-8740. The examiner can normally be reached Monday - Friday, 9:00-6:00 ET.
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, Karlheinz Skowronek can be reached at (571) 272-1113. 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.
/M.K.T./Examiner, Art Unit 1687
/Karlheinz R. Skowronek/Supervisory Patent Examiner, Art Unit 1687
Prosecution Insights