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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 6/10/2026 has been entered.
Claim Status
Claims 1-3,5-9,11-15 and 17-18 are pending.
Claims 4, 10, and 16 are cancelled.
Claims 1-3,5-9,11-15 and 17-18 are rejected
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55. As such effective filing date of claims 1-3,5-9,11-15 and 17-18 is 5/26/2021.
Claim Rejections - 35 USC § 103
Response to Amendment
In view of applicant’s amendments to the claims, previous rejections under 35 U.S.C. 103 have been reviewed, updated, and provided below.
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.
Claims 1-3, 6-9, 12-15, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Feinberg et al. (US 20190272468 A1;), Jaeger et al. (Journal of Chemical Information and Modeling (2018) 27-35;), and Oulhote et al. (BioRxiv (2017) 1-27).
Claim 1 is directed to a method for training a compound property prediction model via the use of spatial structural information and training using said information.
Claim 7 is directed to an apparatus for training a compound property prediction model via the use of spatial structural information and training using said information.
Claim 13 is directed to a CRM for training a compound property prediction model via the use of spatial structural information and training using said information.
Feinberg et al. teaches in claim 1 “A method for predicting characteristics for molecules , wherein the method comprises: performing a first set of graph convolutions with a spatial graph representation of a set of molecules, wherein the first set of graph convolutions are based on bonds between the set of molecules; performing a second set of graph convolutions with the spatial graph representation, wherein the second set of graph convolutions are based on at least a distance between each atom and other atoms of the set of molecules; performing a graph gather with the spatial graph representation to produce a feature vector; and predicting a set of one or more characteristics for the set of molecules based on the feature vector”, in paragraph [0099] “In some embodiments, predicting a set of one or more characteristics comprises assessing data from the graph gather. Processes in accordance with numerous embodiments of the invention can receive atomic information for the set of one or more molecules. The atomic information can be but is not limited to: bond lengths within a molecule, bond strengths within a molecule, bond angles within a molecule”, in paragraph [0125] “The performance of deep neural network algorithms is highly sensitive to chosen hyper-parameters. Such sensitivity underscores the criticality of rigorous cross-validation”, reading on a method for training a compound property prediction model, the method comprising: for each first sample compound of first sample compounds, acquiring spatial structure information of a spatial structure formed by atoms and chemical bonds that constitute the first sample compound; training, using the first sample compounds as input samples and pieces of corresponding spatial structure information as output samples, to obtain a spatial structure prediction model; and fine-tuning the spatial structure prediction model, using second sample compounds as input samples and pieces of corresponding property information as output samples, to obtain the compound property prediction model, wherein an order of magnitudes of the second sample compounds labeled with the pieces of corresponding property information being less than an order of magnitudes of the first sample compounds that are not labeled with corresponding property information, and wherein the spatial structure information comprises at least one of three-dimensional coordinates, bond angles, or overall potential energy.
Feinberg et al. does not teach the spatial structure information comprising the specified types, property information comprising the specified types, and obtaining the spatial structure prediction model in the specified fashion.
Jaeger et al. teaches in the abstract “Inspired by natural language processing techniques, we here introduce Mol2vec, which is an unsupervised machine learning approach to learn vector representations of molecular substructures. Like the Word2vec models, where vectors of closely related words are in close proximity in the vector space, Mol2vec learns vector representations of molecular substructures that point in similar directions for chemically related substructures. Compounds can finally be encoded as vectors by summing the vectors of the individual substructures and, for instance, be fed into supervised machine learning approaches to predict compound properties. The underlying substructure vector embeddings are obtained by training an unsupervised machine learning approach on a so-called corpus of compounds that consists of all available chemical matter”, on page 29, column 2, paragraph 3 “The ESOL solubility data set was chosen to evaluate the performance of Mol2vec in a regression task to predict the aqueous solubilities of 1144 compounds”, on page 30, column 2, paragraph 3 “The evaluation of different parameters for Mol2vec revealed that the best settings overlap with those recommended for Word2vec in NLP on text data and comprise the Skip-gram model with a window size of 10 and 300-dimensional embeddings of Morgan substructures. The quality of the individual embeddings was assessed by using them as features in supervised ML tasks on the Tox21 data set”, reading on the first sample compounds that are not labeled with corresponding property information, wherein the pieces of corresponding property information comprise at least one of water solubility, toxicity, a matching degree with preset protein, compound reaction characteristics, stability, or degradability.
Oulhote et al. teaches in the abstract “We propose a novel approach using the G-formula, a maximum likelihood-based substitution estimator, combined with an ensemble learning technique (i.e. SuperLearner) to infer causal effect estimates for a multi-pollutant mixture… We compared the method with generalized linear and additive models, elastic net, random forests, and Extreme gradient boosting… The proposed method yielded the best average MSE across all the scenarios, and was therefore, able to adapt to the true underlying structure of the data. The method succeeded to detect the true predictors and interactions, and was less biased in all the scenarios”, where the superimposing of the plurality of models is merely a way of describing an ensemble of models and additionally it would therefore be obvious then if the ensemble was neural networks that a person skilled in the art would optimize the number of layers of each network thereby reading on wherein training the initial neural network to obtain the spatial structure prediction model comprises: training a plurality of single-layer spatial structure prediction models, and obtaining the spatial structure prediction model by superimposing the plurality of single-layer spatial structure prediction models, each of the plurality of single-layer spatial structure prediction models being configured to include features and spatial structures of neighbor molecules.
It would have been obvious at the time of first filing to modify the teachings of Feinberg et al. for the prediction of chemical characteristics using neural networks and specific chemical information with the teachings of Jaeger et al. for the use of unsupervised methods in chemical property prediction, and the teachings of Oulhote for the use of ensembles in chemical simulations. Feinberg et al., and Jaeger et al. are all directed to the same application of chemical property prediction using various methods and sets of data, with Jaeger et al. teaching on page 33, column 2, paragraph 3 “Results on common substructures as well as amino acids nicely illustrate that the derived substructure vectors of chemically related substructures and compounds occupy similar vector space. A thorough evaluation of Mol2vec on different chemical data sets showed that it can achieve state-of-the-art performance and compared with Morgan FPs seems to be especially suited for regression tasks. Additionally, Mol2vec combined with ProtVec (i.e., PCM2vec) performs well in proteo-chemometrics approaches and can be directly applied to data sets with unrelated targets with low sequence similarities”, showcasing the improvement of their respective methodologies. Oulhote et al. similarly showcases the improvement of their model over non-ensemble-based models in the abstract “The proposed method yielded the best average MSE across all the scenarios, and was therefore able to adapt to the true underlying structure of the data. The method succeeded to detect the true predictors and interactions, and was less biased in all the scenarios. Finally, we could correctly reconstruct the exposure-response relationships in all the simulations”. One would have had a reasonable expectation of success given that these methods do not preclude each other and are compatible, i.e. unsupervised neural network ensembles are not inherently impossible to create, but more specifically each of these methods are being used in similar context on similar data sets for improvements to the paradigm and would therefore be obvious to try to combine. Therefore, it would have been obvious at the time of first filing to have modified the teachings of each and to be successful.
Claim 2 is directed to the method of claim 1 but further specifies the acquiring of atoms and chemical bonds formed by said atoms for the compound and via experimental calculations or molecular dynamics simulation, determining the specified spatial characteristics of the molecule to use in the training.
Claim 8 is directed to the apparatus of claim 7 but further specifies the acquiring of atoms and chemical bonds formed by said atoms for the compound and via experimental calculations or molecular dynamics simulation, determining the specified spatial characteristics of the molecule to use in the training.
Claim 14 is directed to the CRM of claim 13 but further specifies the acquiring of atoms and chemical bonds formed by said atoms for the compound and via experimental calculations or molecular dynamics simulation, determining the specified spatial characteristics of the molecule to use in the training.
Feinberg et al. teaches in claim 2 “further comprising building a spatial graph representation of the set of molecules, wherein building the spatial graph representation comprises generating a distance matrix and an adjacency tensor, wherein the distance matrix denotes distances between atoms of the set of molecules and the adjacency tensor indicates a plurality of different edge types between atoms”, it would be inherent to any calculation of distance within a spatial context that both position of the atoms and the angles between the atoms would also be calculated, and therefore would read on wherein acquiring spatial structure information of the spatial structure formed by atoms and chemical bonds that constitute the first sample compound, comprises: acquiring the atoms and the chemical bonds, formed by the atoms, constituting the first sample compound; through a molecular dynamics simulation or an experimental calculation, determining three-dimensional coordinates of respective atoms, bond angles between different chemical bonds, atomic distances between the atoms, and an overall potential energy presented by the atoms and the chemical bonds; and using at least one of the three-dimensional coordinates, the bond angles, the atomic distances, and the overall potential energy as the spatial structure information of the first sample compound.
Claim 3 is directed to the method of claim 1 but further specifies that the compound property information comprise one of the elements listed in the group provided.
Claim 9 is directed to the apparatus of claim 7 but further specifies that the compound property information comprise one of the elements listed in the group provided.
Claim 15 is directed to the CRM of claim 12 but further specifies that the compound property information comprise one of the elements listed in the group provided.
Feinberg et al. teaches in paragraph [0121] “In some embodiments, the set of one or more characteristics comprises binding affinity. In some embodiments, the set of one or more characteristics comprises ligand conformation. In some embodiments, the set of one or more characteristics can be charge of the ligand, toxicity, absorption, distribution, metabolism, elimination, CYP450 subtype inhibition, metabolic stability, membrane permeability, oral bioavailability, quantum electronic properties, solubility, Log D, or a combination thereof”, reading on wherein the property information of a compound comprises at least one of water solubility, toxicity, a matching degree with preset protein, compound reaction characteristics, stability, or degradability.
Claim 6 is directed to the method of claim 1 but further specifies the acquiring of a to-be-determined compound with properties that are to be determined, and then using the model to predict property information of the compound.
Claim 12 is directed to the apparatus of claim 7 but further specifies the acquiring of a to-be-determined compound with properties that are to be determined, and then using the model to predict property information of the compound.
Claim 18 is directed to the CRM of claim 13 but further specifies the acquiring of a to-be-determined compound with properties that are to be determined, and then using the model to predict property information of the compound.
Feinberg et al. teaches in claim 1 “A method for predicting characteristics for molecules… and predicting a set of one or more characteristics for the set of molecules…”, reading on acquiring a to-be-determined compound with properties to be determined; and calling the compound property prediction model to predict property information of the to-be-determined compound.
Claims 5, 11, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Feinberg et al. (US 20190272468 A1; previously cited), Jaeger et al. (Journal of Chemical Information and Modeling (2018) 27-35; newly cited), and Oulhote et al. (BioRxiv (2017) 1-27; newly cited) as applied to claims 1-3, 6-9, 12-15, and 18 above, and further in view of Polino et al. (arXiv preprint (2018) 1-21; previously cited).
Claim 5 is directed to the method of claim 1 but further specifies distillating, in response to a complexity of the spatial structure prediction model exceeding a preset complexity, to obtain a lightweight spatial structure prediction model through a model distillation technology.
Claim 11 is directed to the apparatus of claim 7 but further specifies distillating, in response to a complexity of the spatial structure prediction model exceeding a preset complexity, to obtain a lightweight spatial structure prediction model through a model distillation technology.
Claim 17 is directed to the CRM of claim 13 but further specifies distillating, in response to a complexity of the spatial structure prediction model exceeding a preset complexity, to obtain a lightweight spatial structure prediction model through a model distillation technology.
Feinberg et al., Jaeger et al., Li et al., and Oulhote et al. teach the method of claims 1-3, 6-9, 12-15, and 18 as previously described.
Feinberg et al., Jaeger et al., Li et al., and Oulhote et al. do not teach model distillation.
Polino et al. teaches in the abstract “This paper focuses on this problem, and proposes two new compression methods, which jointly leverage weight quantization and distillation of larger networks, called “teachers,” into compressed “student” network. The first method we propose is called quantized distillation and leverages distillation during the training process, by incorporating distillation loss, expressed with respect to the teacher network, into the training of a smaller student network whose weights are quantized to a limited set of levels… We show that quantized shallow students can reach similar accuracy levels to state-of-the-art full-precision teacher models, while providing up to order of magnitude compression, and inference speedup that is almost linear in the depth reduction”, reading on distillating, in response to a complexity of the spatial structure prediction model exceeding a preset complexity, to obtain a lightweight spatial structure prediction model through a model distillation technology.
It would have been obvious at the time of invention to a person skilled in the art to modify the teachings of Feinberg et al. for compound property prediction, with the teachings of Polino et al. for the use of model distillation, as Polino et al. points out “One aspect of the field receiving considerable attention is efficiently executing deep models in resource-constrained environments”, and their paper shows “that quantized shallow students can reach similar accuracy levels to state-of-the-art full-precision teacher models, while providing up to order of magnitude compression, and inference speedup that is almost linear in the depth reduction”. One would have had a reasonable expectation of success given that the method of Feinberg et al. produces a feature vector which is then used to predict characteristics of molecules and Polino et al. uses scaling functions for such vectors. Therefore, it would have been obvious at the time of invention to incorporate the teachings of each and to be successful.
Response to Arguments
Applicant's arguments filed 6/10/2026 have been fully considered but they are not persuasive.
Applicant asserts on page 9 of the Remarks filed 6/10/2026 that the combination of references fails to produce the combined process required by features (a) and (b) of the amended claim. More specifically applicant asserts the references fail to disclose the training of a plurality of prediction models and generating a prediction model from superimposing multiple smaller models.
Applicants’ argument is inaccurate. Oulhote et al. explicitly teaches ensemble architectures, which in view of the structure prediction models of Feinberg et al., render obvious the layering of structure prediction models in an ensemble architecture. Additionally, both Feinberg et al. and Jaeger et al. teach the training of their models, paragraph [0020] “Systems and methods for training spatial convolution graph models in accordance with embodiments of the invention are illustrated” and page 29, column 1, paragraph 3 “The training was done using the gensim23 implementation of Word2vec, which is a shallow, two-layer neural network”, respectively.
Applicant asserts on page 10 of the Remarks filed 6/10/2026 that the combination of references fails to disclose the limitations of feature (b), specifically the fine tuning, the ratio of sample compounds to unlabeled compounds, and the specified property information.
Applicants’ argument is inaccurate. Feinberg et al. in paragraph [0125] and Jaeger et al. on page 30, column 2, paragraph 3, discuss hyperparameter training in view of the training of their model, i.e. fine-tuning. Additionally, Jaeger et al. explicity teaches the use of solubility data. Finally, the precise ratio of data for optimal model performance, i.e. the oders of magniture difference in data, is merely obvious to optimize (See MPEP 2144.05(II)(A) - Merck & Co. Inc. v. Biocraft Lab. Inc., 874 F.2d 804, 809, 10 USPQ2d 1843, 1848 (Fed. Cir. 1989), cert. denied, 493 U.S. 975 (1989)(Claimed ratios were obvious as being reached by routine procedures and producing predictable results); In re Kulling, 897 F.2d 1147, 1149, 14 USPQ2d 1056, 1058 (Fed. Cir. 1990)(Claimed amount of wash solution was found to be unpatentable as a matter of routine optimization in the pertinent art, further supported by the prior art disclosure of the need to avoid undue amounts of wash solution); and In re Geisler, 116 F.3d 1465, 1470, 43 USPQ2d 1362, 1366 (Fed. Cir. 1997)(Claims were unpatentable because appellants failed to submit evidence of criticality to demonstrate that that the wear resistance of the protective layer in the claimed thickness range of 50-100 Angstroms was "unexpectedly good")).
Lastly, applicant asserts that the cited references fail to disclose the sequential two-stage process taught by the amended claims.
Applicants’ argument is inaccurate. There is a difference between the recited limitations of the independent claim and scope of applicant’s arguments. Specifically, the claims recite training a model to obtain a super-imposed model, and fine-tuning the model, where fine tuning is done in training to optimize the hyperparameters of the model, and both Feinberg et al. in paragraphs [0020] and [0125] and Jaeger et al. on page 28, column 1, paragraph 3 and page 30, column 2, paragraph 3, discuss hyperparameter training, i.e. fine-tuning.
Conclusion
No claims are allowed.
All claims are identical to or patentably indistinct from, or have unity of invention with claims in the application prior to the entry of the submission under 37 CFR 1.114 (that is, restriction (including a lack of unity of invention) would not be proper) and all claims could have been finally rejected on the grounds and art of record in the next Office action if they had been entered in the application prior to entry under 37 CFR 1.114. Accordingly, THIS ACTION IS MADE FINAL even though it is a first action after the filing of a request for continued examination and the submission under 37 CFR 1.114. See MPEP § 706.07(b). 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 KEENAN NEIL ANDERSON-FEARS whose telephone number is (571)272-0108. The examiner can normally be reached M-Th, alternate F, 8-5.
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 571-272-9047. 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.
/K.N.A./ Examiner, Art Unit 1687
/LARRY D RIGGS II/ Supervisory Patent Examiner, Art Unit 1686
Prosecution Insights