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 .
Request for Continued Examination (RCE)
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 11/21/25 has been entered.
Priority
This application has no priority claims, thus the filing date of 3/8/2022 is acknowledged.
Status
Claims 1-2, 4, 7-13, 15-21 and 23-24 are pending. Claims 23-24 were newly presented.
New Claim Rejections - 35 USC § 103
Claims 1-2, 4, 7-13, 15-21 and 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (ACS Appl. Mater. Interfaces 2020, 12, p. 734−743, p. S1-S22) in view of the secondary references of Dureckova et al. (J. Phys. Chem. C 2019, 123, 4133−4139), Burner et al. (J. Phys. Chem. C 2020, 124, 27996−28005), Radeck (“Automated deployment of machine learning applications to the cloud”, Thesis Heidelberg University, (26.10.2020), 100 pages), Bucior et al. (Cryst. Growth Des. 2019, 19, 6682−6697), Marchi et al.(ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), DOI: 10.1109/ICASSP40776.2020, 4-8 May 2020, p. 956-960), Wigh et al. (WIREs Comput Mol Sci. 2022, published: 18 February 2022;12:e1603, 19 pages), and RDKit Documentation (RDKit Documentation, https://rdkit.org/docs/ retrieved from archive.org with public available date 2021-10-26).
Zhang teaches machine learning enabled design of metal-organic frameworks (MOFs) for carbon capture (Title; Abstract; p. 739: “Our algorithm was also successfully applied to design novel MOFs promising for CO2-capture application.”). Zhang’s teachings correspond to claim 1 as follows:
A computer-implemented method the method comprising:
selecting a set of chemical groups relating to carbon dioxide capture characteristics, wherein the chemical groups are represented by simplified molecular-input line-entry system arbitrary target specification (SMARTS) sub-structure descriptors;
generating a structure-based key defining a fixed number of bits corresponding to the selected set of chemical groups relating to carbon dioxide capture characteristics;
(Zhang p. 735: “In the SMILES model, a string (S) is in a vector form of S = {s1, s2, . . ., sT}, where symbol si represents chemical information such as atoms, ions, bonds, the number of rings, branch, or terminal symbol”; Fig 1:
PNG
media_image1.png
648
953
media_image1.png
Greyscale
Zhang p. 741: section “Training the Recurrent Neural Network Model”; p. 735: “the following 40 symbols were identified in our definition of SMILES strings; si ∈ {“\n”, “&”, “C”, “(“, “)”, “c”, “1”, “2”, “o”, “=”, “O”, “N”, “3”, “F”, “[C@@H]”, “n”, “-”, “#”, “/”, “[nH]”, “Br”, “[C@H]”, “S”, “s”, “4”, “Cl”, “[C@]”, “[C@@]”, “\\”, “5”, “[S@@]”, “[S@]”, “6”, “7”, “I”, “P”, “[P@@]”, “[PH2]”, “[P@]”, “[PH]”}. The meaning of each symbol is provided in Table S1.”; p. 739: “On the Diversity of Newly Designed MOFs. Our algorithm is directed with one clear mission of generating a set of MOFs enriched with structures which are promising for a target application … topological data analysis (TDA) was conducted. For TDA, we used a similarity measure comparing linkers in MOFs by noting that a metal node and net are the same for each case. For a similarity measure, an organic linker is represented using a topological fingerprint which generates a bit vector based on the molecular fragments.”);
generating, based on the structure-based key, fixed bit fingerprints for screening, directly on a cloud-based instance, simplified molecular-input line- entry system (SMILES) structure descriptors of candidate molecules based on carbon dioxide capture capacity,
(Zhang p. 735: “It begins with providing input information about a metal node, topology, and target application, and then an organic linker is generated with the Monte Carlo tree search (MCTS)26 implemented in the ChemTS”; p. 739-740: “for each pair of structures, a distance was estimated by utilizing the Tanimoto coefficient as follows … = 1 – c/(a+b-c) where c is bits set in common in the two fingerprints, and a (or b) represents bits set for each fingerprint. The pairwise distances were used to construct a distance matrix that works as an input for generating mapper plots using TDA.”) wherein the generating the fixed bit fingerprints comprises:
converting the SMARTS sub-structure descriptors and the SMILES structure descriptors into internal graph representations of the candidate molecules and the chemical groups;
(Zhang p. 735: S = {s1, s2, . . ., sT}; Fig 1; RNN; p. 741: section “Training the Recurrent Neural Network Model”, “We chose to utilize the Simplified Molecular Input Line Entry
System (SMILES)29 representation which uses sequences of strings as a compact grammar for molecules based on principles of molecular graph theory.”; p. 739: “For a similarity measure, an organic linker is represented using a topological fingerprint31 which generates a bit vector based on the molecular fragments”; Zhang’s cited reference 31 is to RDKit which as evidenced by RDKit p. 27 converts SMARTS and SMILES to fingerprints and RDK Book p. 28 describes the algorithm as “The fingerprinting algorithm identifies all subgraphs in the molecule”);
executing a parallelized search of the internal graph representations to determine a presence or absence of each of the chemical groups in each of the candidate molecules, wherein the parallelized search batches the candidate molecules or the chemical groups while maintaining an order of output results that matches an input order of the chemical groups; and
(Zhang Fig. 1: “Monte-Carlo Tree Search”. As evidenced by RDKit documentation, the modules are parallel with numthreads option and functions on molecules in the same input/output order)
generating, based on the output results of the parallelized search, a featurization of each of the candidate molecules as a fixed bit fingerprint with a number of bits matching the structure-based key, wherein each bit in the fixed bit fingerprint indicates the presence or absence of a corresponding one of the chemical groups defined by the structure-based key; and
(p. 735: “Given a partial string of symbols (s1, s2. . ., si) at level i, the distribution of the next symbol si+1 is predicted according to the conditional probability which is trained by the recurrent neural network (RNN) for the SMILES strings in a database.”; p. 739: “For a similarity measure, an organic linker is represented using a topological fingerprint31 which generates a bit vector based on the molecular fragments.”) and
applying the fixed bit fingerprints to a next step in a workflow involving materials for carbon dioxide capture (Fig. 1: “Selection” and “Output”; Table 3).
Zhang does not teach “screening directly on a cloud-based instance”.
Relatedly, Dureckova and Burner teach computer implemented machine learning models for predicting carbon dioxide capture using structure-based descriptors of MOFs to generate high-performing materials. For example, Burner summarizes the method as:
PNG
media_image2.png
405
645
media_image2.png
Greyscale
.
Radeck teaches known and commercially available cloud systems where machine learning applications can be deployed that one of skill in the art would have considered given the computational requirements for implementing screening methods such as taught by Zhang. In view of the success of Zhang, Dureckova, and Burner one of ordinary skill in the art would have considered implementing the method in the cloud and arrive at the claimed invention with a reasonable expectation of success.
Bucior teaches schemes of representing MOF structures for use in searching and machine learning where the structures are derived from SMILES-derived SMARTS substructure descriptions such as MOFkey/id (p. 6682, 6689-6693):
PNG
media_image3.png
513
976
media_image3.png
Greyscale
Marchi teaches machine learning featurization with molecular fingerprints having a fixed bit vector from SMILES (p. 956).
Wigh review molecular representation in machine learning with molecular fingerprints bit vectors and molecular graphs (p. 1-9).
Regarding claim 2, Zhang teaches the use of a parallel computing library ChemTS (p. 735: “Our tool is built on the basis of the ChemTS16 using a Monte Carlo tree search (MCTS)26 which showed remarkable success as a heuristic decision algorithm in computer Go27 and recurrent neural network (RNN) which is a class of artificial neural networks (ANN) generating a series of neurons connected in a certain order to process data.”).
Regarding claim 4 relating to the variety of chemical groups, Zhang teaches a diversity in generation of molecular fragments (p. 739-740, section “On the Diversity of Newly Designed MOFs”).
Regarding claim 8, Zhang teaches a regression model (p. 741: section “Training the Recurrent Neural Network Model”, “In a training procedure, the weight matrices of the neural network are fitted by minimizing the loss function”).
Regarding claim 9, Zhang teaches use of the neural network model (p. 741: section “Training the Recurrent Neural Network Model”; Fig. 1).
Regarding claim 10, Radeck teaches known and commercially available cloud systems where machine learning applications can be deployed that one of skill in the art would have considered given the computational requirements for implementing screening methods such as taught by Zhang.
Regarding claim 11, Zhang teaches providing an entry in a molecule database for a fixed bit fingerprint and searching as claimed (p. 735: “Given a partial string of symbols (s1, s2. . ., si) at level i, the distribution of the next symbol si+1 is predicted according to the conditional probability which is trained by the recurrent neural network (RNN) for the SMILES strings in a
database.”; p. 739: “For a similarity measure, an organic linker is represented using a topological fingerprint31 which generates a bit vector based on the molecular fragments.”).
Regarding claim 12 to a computer system performing the method of claim 1, Zhang teaches implementation on a computer system and the same corresponding components (p. 735: “In this work, we aim to develop a generative model-based computational framework for the tailor-made generation of MOF which is promising for targeted applications. Our tool is built on the basis of the ChemTS16 using a Monte Carlo tree search (MCTS)26 which showed remarkable success as a heuristic decision algorithm in computer Go27 and recurrent neural network (RNN) which is a class of artificial neural networks (ANN) generating a series of neurons connected in a certain order to process data. We test our tool by performing in silico experiments to design high-performing MOFs for methane-storage and carbon-capture applications.”).
Regarding claim 13, Zhang teaches the use of a parallel computing library ChemTS (p. 735: “Our tool is built on the basis of the ChemTS16 using a Monte Carlo tree search (MCTS)26 which showed remarkable success as a heuristic decision algorithm in computer Go27 and recurrent neural network (RNN) which is a class of artificial neural networks (ANN) generating a series of neurons connected in a certain order to process data.”).
Regarding claim 15 relating to the variety of chemical groups, Zhang teaches a diversity in generation of molecular fragments (p. 739-740, section “On the Diversity of Newly Designed MOFs”).
Regarding claim 16-17, Zhang teaches using an algorithm which includes the same data (p. 735, section “Construction of Metal−Organic Frameworks Using the Monte Carlo Tree Search”).
Regarding claim 18, Radeck teaches known and commercially available cloud systems where machine learning applications can be deployed that one of skill in the art would have considered given the computational requirements for implementing screening methods such as taught by Zhang.
Regarding claim 19, Zhang teaches providing an entry in a molecule database and searching as claimed (p. 735: “Given a partial string of symbols (s1, s2. . ., si) at
level i, the distribution of the next symbol si+1 is predicted according to the conditional probability which is trained by the recurrent neural network (RNN) for the SMILES strings in a
database.”).
Regarding claim 20 to a computer program product for performing the method of claim 1, Zhang teaches implementation on a computer system and the same corresponding components (p. 735: “In this work, we aim to develop a generative model-based computational framework for the tailor-made generation of MOF which is promising for targeted applications. Our tool is built on the basis of the ChemTS16 using a Monte Carlo tree search (MCTS)26 which showed remarkable success as a heuristic decision algorithm in computer Go27 and recurrent neural network (RNN) which is a class of artificial neural networks (ANN) generating a series of neurons connected in a certain order to process data. We test our tool by performing in silico experiments to design high-performing MOFs for methane-storage and carbon-capture applications.”).
Regarding claim 21, Zhang teaches the use of a parallel computing library ChemTS (p. 735: “Our tool is built on the basis of the ChemTS16 using a Monte Carlo tree search (MCTS)26 which showed remarkable success as a heuristic decision algorithm in computer Go27 and recurrent neural network (RNN) which is a class of artificial neural networks (ANN) generating a series of neurons connected in a certain order to process data.”).
Regarding claim 23, Zhang teaches batching the data set (p. 741) as well as parallel tools, RDKit. Regarding claim 24, the bit vectors taught by the prior art have the same value for each bit – “bit” is short for binary digit, i.e. 0/1.
Regarding the “fixed bit” language and graph search, Zhang teaches a bit vector representation of SMILES defined structures and symbols representing chemical groups (p. 735, 739) which one of ordinary skill in the art would recognize as equivalent computer-based representations of chemical structures and would use them interchangeable to optimize the ability to perform the computer methodology represented by Zhang’s Figure 1 (p. 735: “Given a partial string of symbols (s1, s2. . ., si) at level i, the distribution of the next symbol si+1 is predicted according to the conditional probability which is trained by the recurrent neural network (RNN) for the SMILES strings in a database.”; p. 739: “For a similarity measure, an organic linker is represented using a topological fingerprint31 which generates a bit vector based on the molecular fragments.”).
One of ordinary skill in the art following Zhang’s teaching of using ML to train and screen for MOFs for carbon dioxide capture would have been motivated to further develop and refine the method for improved performance. The level of skill in the art of computational chemistry is very high as evidenced by Zhang, Dureckova, Burner, Bucior, Marchi and Wigh which are all in the same field of endeavor. Those of ordinary skill in the art would have reasonably considered utilizing fingerprints, SMARTS structure descriptors and molecular/internal graphs for efficiently searching chemical groups as was routine in the art and implemented in RDKit. One of ordinary skill in the art would have known the importance of matching input/output order to preserve data integrity. One of ordinary skill in the art would have considered the same bit vector as taught by Marchi as a molecular representation.
With each of the claims, the level of skill in the art is very high such that one of ordinary skill in the art would consider routine the combination of elements from the teaching of the art. One of ordinary skill in the art would have recognized that the results of the combination would be predictable due to the well-known nature and optimizations routinely performed in the art. Thus, one of ordinary skill in the art would have arrived at the invention as claimed before the effective filing date with a reasonable expectation of success.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (ACS Appl. Mater. Interfaces 2020, 12, p. 734−743, p. S1-S22) in view of the secondary references as applied to claims 1-2, 4, 7-13, 15-21 and 23-24 above and further in view of Kuenemann et al. (Mo!. Inf. 2017, 36, 1600143, 13 pages).
Regarding claim 7, Zhang does not specify that the chemical groups are amine-based carbon dioxide solvents to target candidate molecules in the form of carbon dioxide capturing amine molecules.
Kuenemann modeling solutions of amines based on experimental data (p. 2: “Herein, we assembled the largest dataset of experimental measurements on CO2 absorption properties for aqueous solutions of amines”) and applied machine learning techniques to generate a model for screening amine with desired CO2 absorption profiles (p.13).
One of ordinary skill in the art would have readily considered combining the teaching of Zhang with Kuenemann because they are in the same field of endeavor of computation modeling of CO2 absorption. One of ordinary skill in the art would have had a reasonable expectation of success because the two techniques are closely related machine learning models where the input parameters could be readily changed to accommodate the amines vs. MOFs. Thus, the claim is prima facie obvious.
Response to Remarks - 35 USC § 103
Applicant argues that the cited references do not teach the as amended claimed structure-based key or fingerprints and that “the claimed fixed bit fingerprints are sets of bits indicating the presence or absence of chemical groups defined by the structure-based key”.
This argument is not persuasive because Zhang the use of bit vector representation (p. 735; p. 739: “represented using a topological fingerprint which generates a bit vector based on the molecular fragments.”). In addition, as taught by Marchi and Wigh, such molecular representations were well known in the art and one of ordinary skill in the art would have considered their use to provide an accurate numerical representation of the molecular structure that was computationally efficient.
Applicant argues that the fingerprints are generated by a different process than the cited art, that “the SMARTS sub-structure descriptors of the chemical groups defined by the key and the SMILES structure descriptors of the candidate molecules are converted into internal graph representations for sub-structure searching, which is parallelized in batches and carried out for each chemical group defined by the key for every candidate molecule”, and that “Maintaining the order of the chemical groups defined by the key in the output fingerprints preserves chemical information and is not addressed by the cited references.”
This argument is not persuasive because one of ordinary skill in the art would have known of the importance to preserve the structural information in the representation in a manner that could be effectively utilized in a computational technique. In addition, Wigh teaches the importance of ordering of the representation (p. 5-6) and Marchi teaches generation of a bit vector hash for featurization (p. 956).
Conclusion
No claims 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 ROBERT H HAVLIN whose telephone number is (571)272-9066. The examiner can normally be reached 9am - 6pm.
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, Kortney Klinkel can be reached at 571-270-5239. 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.
/ROBERT H HAVLIN/Primary Patent Examiner, Art Unit 1626
Prosecution Insights