DETAILED ACTION
Claim Status
Claims 1-12, 15-17, 20, 26, 29, and 32-33 are rejected.
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 .
Information Disclosure Statement
The Information Disclosure Statements filed on 01/18/2023 is in compliance with the provisions of 37 CFR 1.97 and have been considered in full. A signed copy of list of references cited from the IDS is included with this Office Action.
Priority
This application claims Domestic Benefit to application #63291552, filed 12/20/2021. Domestic Benefit is acknowledged. Therefore, the effective filing date of claim(s) 1-12, 15-17, 20, 26, 29, 32-33 is 12/20/2021.
Drawings
Drawings 3B, 5, and 7 are executed in color. Color photographs and color drawings are not accepted in utility applications unless a petition filed under 37 CFR 1.84(a)(2) is granted. Any such petition must be accompanied by the appropriate fee set forth in 37 CFR 1.17(h), one set of color drawings or color photographs, as appropriate, if submitted via the USPTO patent electronic filing system or three sets of color drawings or color photographs, as appropriate, if not submitted via the via USPTO patent electronic filing system, and, unless already present, an amendment to include the following language as the first paragraph of the brief description of the drawings section of the specification:
The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.
Color photographs will be accepted if the conditions for accepting color drawings and black and white photographs have been satisfied. See 37 CFR 1.84(b)(2).
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-12, 15-17, 20, 26, 29, and 32-33 rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claims 1, 2, 20, and 33 recite wherein clauses regarding the training of the machine learning model and generation of the training examples used in training the machine learning model are stated such that it is unclear if those steps are required to be performed in the metes and bounds of the invention or if they merely recite product-by-process limitations describing how the trained machine learning model was previously trained. Claims 3-12, 15-17, 26, 29, and 32 depend from claim 1 and are therefore rejected for the same indefiniteness issue.
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 1-12, 15-17, 20, 26, 29, and 32-33 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
In accordance with MPEP § 2106, claims found to recite statutory subject matter ( Step 1 : YES) are then analyzed to determine if the claims recite any concepts that equate to an abstract idea, law of nature or natural phenomenon (Step 2A, Prong 1). In the instant application, the claims recite the following limitations that equate to an abstract idea:
1. Iteratively… generate the small molecule compound
1. analyzing the graph adjacency tensor to generate probabilities
1. Selecting one of the available actions based on the probabilities
1. Updating the graph adjacency tensor with a value indicative of the selected action to generate an updated graph adjacency tensor
15. determining a subgraph of the graph adjacency tensor
2 and 15. analyzing a subgraph of the graph adjacency tensor to generate probabilities
2. decomposing the small molecule compound into a sequence of actions
7. updating a diagonal of the graph adjacency tensor with a value indicative of the added atom
8.updating an upper portion of the graph adjacency tensor with a value indicative of the added bond
9. updating a diagonal of the graph adjacency tensor with a value indicative of the assigned charge
The limitations for “generating,” “analyzing,” “determining,” “decomposing,” and “updating” are all verbal equivalents for calculations that could be performed by a human being with a pen and paper. The limitation to iteratively generate the small molecule compound includes the remaining limitations as its steps. In order to generate the small molecule compound, a series of steps are set forth for altering the numbers in a tensor, which is a mathematical object. Therefore, these limitations are mathematical relationships and mental processes. The limitation for “selecting” is a mental process because a human being can choose an available action based on a probability.
Claims 3, 5, 6, 10, 16, 17, 26, and 29 provide supportive details about the steps of the mathematical relationships/mental processes, such as the “available actions,” “threshold number,” “selecting one of the available actions,” the dimensions of the graph adjacency tensor, the dimensions of the subgraph, and the percentage thresholds of the molecules. These additional details do not alter the nature of the mathematical concepts/mental processes they depend on and are therefore part of the mathematical relationships/mental processes above.
While claims 1-2, 4, 15, 20, and 32-33 recite performing some aspects of the analysis with a “trained machine learning model” stipulated to be a “neural network”, or a “non-transitory computer readable medium,” there are no additional limitations that indicate that this neural network requires anything other than carrying out the recited mental process or mathematical concept in a generic computer environment. Merely reciting that a mental process is being performed in a generic computer environment does not preclude the steps from being performed practically in the human mind or with pen and paper as claimed. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then if falls within the “Mental processes” grouping of abstract ideas. As such, claim(s) 1-12, 15-17, 20, 26, 29, 32-33 recite(s) an abstract idea/law of nature/natural phenomenon ( Step 2A, Prong 1 : YES).
Claims found to recite a judicial exception under Step 2A, Prong 1 are then further analyzed to determine if the claims as a whole integrate the recited judicial exception into a practical application or not (Step 2A, Prong 2). This judicial exception is not integrated into a practical application because the claims do not recite an additional element that reflects an improvement to technology or applies or uses the recited judicial exception to effect a particular treatment for a condition. Rather, the instant claims recite additional elements that amount to mere instructions to implement the abstract idea in a generic computing environment or mere instructions to apply the recited judicial exception via a generic treatment. Specifically, the claims recite the following additional elements:
1. iteratively applying a trained machine learning model
32. the machine learning model is a trained neural network
1. obtaining a graph adjacency tensor
2. obtaining a plurality of (training) small molecule compounds
33. A non-transitory computer readable medium
There are no limitations that indicate that the claimed neural network, “non-transitory computer readable medium”, or the formats of the provided data require anything other than generic computing systems. As such, these limitations equate to mere instructions to implement the abstract idea on a generic computer that the courts have stated does not render an abstract idea eligible in Alice Corp., 573 U.S. at 223, 110 USPQ2d at 1983. See also 573 U.S. at 224, 110 USPQ2d at 1984.
The additional elements for “obtaining” input data do not add a meaningful limitation to the abstract idea because they amount to mere data gathering steps that would be required for the claimed mental processes. These limitations serve to gather data that is used as input for the abstract idea and there is no indication that the abstract idea has any impact on those data gathering steps. The courts have indicated that mere data gathering activity is insignificant extra-solution activity that does not provide a practical application (see MPEP 2106.05(g)).
There are limitations in claims 2 and 20 about the training of the machine learning model, but as the claims state that the neural network is already trained, they do not make up a part of the steps of the mathematical relationships/mental processes and are therefore additional elements.
As such, claims 1-12, 15-17, 20, 26, 29, 32-33 is/are directed to an abstract idea/law of nature/natural phenomenon ( Step 2A, Prong 2 : NO).
Claims found to be directed to a judicial exception are then further evaluated to determine if the claims recite an inventive concept that provides significantly more than the judicial exception itself (Step 2B). The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the claims recite additional elements that equate to mere instructions to apply the recited exception in a generic way or in a generic computing environment. The instant claims recite the following additional elements:
1. iteratively applying a trained machine learning model
32. the machine learning model is a trained neural network
1. obtaining a graph adjacency tensor
2. obtaining a plurality of (training) small molecule compounds
33. A non-transitory computer readable medium
As discussed above, there are no additional limitations to indicate that the claimed neural network or “non-transitory computer readable medium” requires anything other than generic computer components in order to carry out the recited abstract idea in the claims. Claims that amount to nothing more than an instruction to apply the abstract idea using a generic computer do not render an abstract idea eligible. Alice Corp., 573 U.S. at 223, 110 USPQ2d at 1983. See also 573 U.S. at 224, 110 USPQ2d at 1984.
Storing and retrieving information in memory, as occurs in the “obtaining” limitations, is a form of well-understood, routine, conventional activity, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015); OIP Techs., 788 F.3d at 1363, 115 USPQ2d at 1092-93.
Iterative application of an algorithm, as occurs in the first limitation of claim 1, is a type of performing repetitive calculations, Flook, 437 U.S. at 594, 198 USPQ2d at 199, which the courts have found to be well understood, routine, and conventional activity.
The additional elements do not comprise an inventive concept when considered individually or as an ordered combination that transforms the claimed judicial exception into a patent-eligible application of the judicial exception. Therefore, the claims do not amount to significantly more than the judicial exception itself ( Step 2B : No). As such, claims 1-12, 15-17, 20, 26, 29, and 32-33 are not patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-12, 15-17, 20, 26, 29, 32-33 are rejected under 35 U.S.C. 103 as being unpatentable over De Cao et al. (arXiv:1805.11973, IDS Reference) (hereafter “De Cao”), and Li et al. (ICML, PMLR 80, 2018, IDS reference) (hereafter “Li”), as evidenced by Chehreghani et al. (WIREs Data Mining Knowl Discov. 2021;11:e1393) and Encyclopedia Britannica (Britannica Editors. Encyclopedia Britannica, 2016),.
Regarding claim 1, fig. 2 of De Cao, reproduced below, illustrates a method, called MolGAN, for generating small molecule compounds. In this method, a machine learning model is applied to a graph adjacency tensor to generate probabilities of valid substructures of molecules based on actions taken. The probabilities of validity are calculated as part of their optimized objectives of “druglikeness”, “synthesizability”, and “solubility” in De Cao, and are shown in Table 2, also reproduced below.
PNG
media_image1.png
491
1025
media_image1.png
Greyscale
PNG
media_image2.png
591
1064
media_image2.png
Greyscale
De Cao does not teach the iterative updating of the graph adjacency tensor action-by-action as occurs in claim 1. However, De Cao does provide a suggestion about iterative updating: “The generative model of Mol-GAN predicts discrete graph structure at once (i.e., nonsequentially) for computational efficiency, although sequential variants are possible in general.”
Li provides for generating probabilities and selecting available actions based on probabilities, shown in Fig. 1 below (Page 3 left col ¶ 3: The graph generation process can be seen as a sequence of decisions, i.e., (1) add a new node or not (with probabilities provided by an faddnode module), (2) add a new edge or not (probabilities provided by faddedge), and (3) pick one node to connect to the new node (probabilities provided by fnodes).)”.
Li’s method is based on iteratively updating a vector: “We use a 3-D conditioning vector c which includes the number of atoms (nodes), the number of bonds (edges) and the number of aromatic rings in a molecule” (pg 8 section 5.3 left col ¶ 1).
A dictionary definition of “tensor” states that it is a “special case of a vector.” ("Tensor". Webster's II New Riverside University Dictionary. 1984, pg 1193). Therefore, Li’s method suggests updating of the graph adjacency tensor.
PNG
media_image3.png
265
805
media_image3.png
Greyscale
Claim 2 is restates the iterative updating algorithm of claim 1, differentiating itself by setting forth limitations related to subgraphs of the graph adjacency tensor, which are created and analyzed by MolGAN (De Cao Fig. 2). The subgraph, representing the bonds, is associated with an annotation matrix containing atom type and bond type (De Cao pg 3 right col ¶ 3), which are then transformed by the machine learning models into candidate transformed molecules. This is a non-sequential version of the steps of claim 2.
Claim 2 also includes language about a sequential method to create training data for the sequential prediction. De Cao gives suggestions about training models directly on graph data, and mentions sequential methods that were trained on graph data, contrasting it with their method of training on SMILES string data: “Several other works have explored training generative models on SMILES representations of molecules… A related line of research considers training deep generative models to output graph-structured data directly. Several works explored auto-encoder architectures utilizing graph convolutions for link prediction within graphs (Kipf & Welling, 2016; Grover et al., 2019; Davidson et al., 2018). Johnson (2017); Li et al. (2018b); You et al. (2018); Li et al. (2018a) on the other hand developed likelihood-based methods to directly output graphs of arbitrary size in a sequential manner.” (De Cao pg 5 left col ¶ 4-5) The term “link prediction” suggests the type of iterative graph-based approach to training and generation suggested by the instant application.
Regarding claim 3, the generator architecture of MolGAN has the space of variation, or “available actions,” limited as such: “We use N = 9 as the maximum number of nodes, T = 5 as the number of atom types (C, O, N, F, and one padding symbol), and Y = 4 as the number of bond types (single, double, triple and no bond).” (De Cao pg 5 right col ¶ 3). MolGAN generates bonds, charges and atoms based on reference molecules. In a sequential implementation, as in Li, this would be represented as selecting a sequence of actions corresponding to these available changes within the variation space (see Li fig. 1 above) .
Regarding claim 4, De Cao writes “We restrict the domain to graphs of a limited number of
nodes” (De Cao pg 4 left col ¶ 3). The nodes of the graph are atoms in De Cao. Therefore, the machine learning model terminates after a threshold number of atoms.
Regarding claim 5, De Cao writes “We use N = 9 as the maximum number of nodes” (De Cao pg 5 right col ¶ 3). The nodes are atoms in De Cao, and the authors suggest that the threshold number can vary. The maximum number of atoms is 9, which is more than 5.
Regarding claim 6, Li teaches that the decision making process seen in fig. 1 above is guided by graph nets (Li pg 3 left col ¶ 3-4) that are trained to select highest probabilities (Li pg 4 left col ¶ 3-4).
Regarding claims 7-9, Li teaches the iterative updating algorithm with adding an atom, bond, or charge (Li fig. 1 pg 8 section 5.3 left col ¶ 1).Chehreghani’s evidentiary figure, provided below, provides an example of updating the diagonal of a graph adjacency matrix, of which a graph adjacency tensor is a special case, shown by the stacked-matrix representation of the graph adjacency tensor in fig. 1 of De Cao.
PNG
media_image4.png
496
850
media_image4.png
Greyscale
Adding an atom, or adding a charge to an atom, in the schema of the instant application, would require updating the diagonal of the graph adjacency tensor, as shown by this image. In the example, updating an upper portion would update a bond edge if one cell is changed from a 1 to a 2 or vice versa. Therefore, these limitations are already part of claims 1 and 3, and are taught by De Cao and Li.
Regarding claims 10 and 11, three dimensions are used for the graph adjacency tensor in De Cao: “We consider that each molecule can be represented by an undirected graph G with a set of edges E and nodes V. Each atom corresponds to a node vi 2 V that is associated with a T-dimensional one-hot vector xi, indicating the type of the atom. We further represent each atomic bond as an edge (vi; vj) 2 E associated with a bond type y 2 f1; :::; Y g. For a molecular graph with N nodes, we can summarize this representation in a node feature matrix X = [x1; :::; xN]T 2 RN_T and an adjacency tensor A 2 RN_N_Y where Aij 2 RY is a one-hot vector indicating the type of the edge between i and j.”(pg 2 left col last ¶) De Cao continues: “We use N = 9 as the maximum number of nodes, T = 5 as the number of atom types (C, O, N, F, and one padding symbol), and Y = 4 as the number of bond types (single, double, triple and no bond). Thesedimensionalities are enough to cover all molecules inQM9.” (De Cao pg 5 right col ¶ 3).
Regarding claim 12, De Cao writes: “Lastly, it will be promising to explore alternative generative architectures within the Mol-GAN framework, such as recurrent graph-based generative models (Johnson, 2017; Li et al., 2018b; You et al., 2018), as our current one-shot prediction of the adjacency tensor is most likely feasible only for graphs of small size.” (De Cao pg 8 right col ¶ 4) De Cao is suggesting that if MolGAN were modified into a sequential algorithm, it would be sensible to increase the space of the adjacency tensor to be higher than the single digit values currently given, in order to test the feasibility of the algorithm on larger molecule sizes. Li, which uses a sequential algorithm, sets the atom threshold at 20 (page 5 right col section 5.1 ¶ 2).
Regarding claim 15, MolGAN determines a subgraph of the graph adjacency tensor and analyzes the subgraph to predict probabilities corresponding to available actions as part of using the graph adjacency tensor to determine probabilities (figure 2).
Regarding claims 16 and 17, the subgraph of MolGAN inherits the 3d structure of its parent, and its dimensions are necessarily smaller or equal than the graph adjacency tensor because the subgraph must be less than or equal to its parent graph. In the course of dividing a graph of size 9 into smaller subgraphs, subgraphs of size 1 to 5 would be generated as part of routine optimization by the ordinary artisan (figure 2). Additionally, figure 1 of Li shows an iteratively updating subgraph with 1-3 nodes.
Regarding claim 20, De Cao gives suggestions about training models directly on graph data, and mentions sequential methods that were trained on graph data, contrasting it with their method of training on SMILES string data: “Several other works have explored training generative models
on SMILES representations of molecules… A related line of research considers training deep generative models to output graph-structured data directly. Several works explored auto-encoder architectures utilizing graph convolutions for link prediction within graphs (Kipf & Welling, 2016; Grover et al., 2019; Davidson et al., 2018). Johnson (2017); Li et al. (2018b); You et al. (2018); Li et al. (2018a) on the other hand developed likelihood-based methods to directly output graphs of arbitrary size in a sequential manner.” (De Cao pg 5 left col ¶ 4-5) The term “link prediction” suggests the type of iterative graph-based approach used in the instant application.
Regarding claims 26 and 29, MolGAN generates “novelty” and “uniqueness” probabilities for its generated structures, at levels greater than 90% (table 2, 3).
Regarding claim 32, De Cao writes “We compare MolGAN against recent neural network-based
drug generation models in a range of experiments”. Evidentiary reference Goodfellow et al. introduced the GAN, explaining that it is a set of neural networks (arXiv:1406.2661, 2014) (page 1 ¶ 1-2). The authors are directly comparing the GAN to the neural network, providing a suggestion to use a single neural network.
Claim 33 is a restatement of claim 1, but directed towards “a non-transitory computer readable medium.” A compact disc (CD) is an example of a non-transitory computer readable medium. Encyclopedia Britannica teaches what a CD is (¶ 1). The arguments against claim 1 apply.
Regarding claims 1-12, 15-17, 20, 26, 29, and 32-33, An invention would have been prima facie obvious to one of ordinary skill in the art at the time of the effective filing date of the invention if some teaching, suggestion, or motivation in the prior art would have led that person to combine the prior art teachings to arrive at the claimed invention. There is a suggestion to use iterative updating in the text of De Cao. There is a suggestion to train the model with graph data, specifically when the model is iterative, the text of De Cao. There is a suggestion to use a neural network in the text of De Cao. Encyclopedia Britannica teaches a non-transitory computer readable medium (CD). There would be a reasonable expectation of success in making this combination to a person of ordinary skill in the art, as De Cao explains that it is “possible in general” to adapt their algorithm to iterative updating, they compare their method against single neural-network based methods, direct the reader to graph-based methods of training sequential molecule prediction models, which are detailed further in Li, teaching the selection of highest-probability available actions, and suggesting to update the graph adjacency tensor. Additionally, there is nothing about the format of the algorithms or data such that they could not be placed on a CD. De Cao also mentions a motivation to move past the one-shot prediction method to improve the model’s predictions about larger molecules, which would entail increasing the atom thresholds as in claim 12 of the instant application. Therefore, it would have been prima facie obvious to one of ordinary skill in the art at the time to modify the method of De Cao by applying the suggestion to use a sequential algorithm, graph data for training, and a single neural network, in order to improve the model’s predictions about larger molecules.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GRACELYN M HILL whose telephone number is (571)272-9871. The examiner can normally be reached Monday-Friday 8:30-5pm.
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, Olivia M. Wise, can be reached at 571-272-2249. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/G.M.H./Examiner, Art Unit 1685
/OLIVIA M. WISE/Supervisory Patent Examiner, Art Unit 1685
Prosecution Insights