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 .
Priority
This application claims benefit of U.S. Provisional Application 63/446,242 filed 2/16/2023. As such the effective filing date of claims 1-19 is 2/16/2023.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 10/30/2023 and 9/20/2026 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statements are being considered by the examiner.
Claim Status
Claims 1-19 are pending.
Claims 1-19 are rejected.
Drawings
The drawings are objected to as failing to comply with 37 CFR 1.84(p)(5) because they include the following reference character(s) not mentioned in the description: Figure 12, Item 1115. Corrected drawing sheets in compliance with 37 CFR 1.121(d), or amendment to the specification to add the reference character(s) in the description in compliance with 37 CFR 1.121(b) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
The drawings are objected to because as described on page 51 of the specification, drawing number 6A is missing (it looks like it is supposed to be figure 7A as many of the item numbers correspond). Corrected drawing sheets in compliance with 37 CFR 1.121(d) are required in reply to the Office action to avoid abandonment of the application. Any amended replacement drawing sheet should include all of the figures appearing on the immediate prior version of the sheet, even if only one figure is being amended. The figure or figure number of an amended drawing should not be labeled as “amended.” If a drawing figure is to be canceled, the appropriate figure must be removed from the replacement sheet, and where necessary, the remaining figures must be renumbered and appropriate changes made to the brief description of the several views of the drawings for consistency. Additional replacement sheets may be necessary to show the renumbering of the remaining figures. Each drawing sheet submitted after the filing date of an application must be labeled in the top margin as either “Replacement Sheet” or “New Sheet” pursuant to 37 CFR 1.121(d). If the changes are not accepted by the examiner, the applicant will be notified and informed of any required corrective action in the next Office action. The objection to the drawings will not be held in abeyance.
Specification
The use of the term Bluetooth, which is a trade name or a mark used in commerce, has been noted in this application. The term should be accompanied by the generic terminology; furthermore the term should be capitalized wherever it appears or, where appropriate, include a proper symbol indicating use in commerce such as ™, SM , or ® following the term.
Although the use of trade names and marks used in commerce (i.e., trademarks, service marks, certification marks, and collective marks) are permissible in patent applications, the proprietary nature of the marks should be respected and every effort made to prevent their use in any manner which might adversely affect their validity as commercial marks.
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.
Claim 10 is 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.
The term “some” in claim 10 is a relative term which renders the claim indefinite. The term “some” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. Here the term “some” modifies the number of parameters being shared between the models being used to predict the binding site of a protein, and as such renders indefinite the number of parameters needing to be shared to achieve the proper accuracy of the prediction model.
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 18-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to non-statutory subject matter. The claim(s) does/do not fall within at least one of the four categories of patent eligible subject matter because the claimed subject matter is “computer program” which is stored in a computer-readable storage medium, which is directed to software per se.
Claims 1-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract ideas without significantly more. The claims recite methods, and a CRM for predicting the binding site of a protein. This judicial exception is not integrated into a practical application because while claims 1-19 attempt to integrate the exception into a practical application, said practical application is a generically recited computer element that does not add meaningful limitations to the abstract idea as it is simply implementing the abstract idea on a computer. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception because the computer elements only store and retrieve information in memory as well as perform basic calculations that are known to be well-understood, routine and conventional computer functions as recognized by the decisions listed in MPEP § 2106.05(d).
Framework with which to Analyze Subject Matter Eligibility:
Step 1: Are the claims directed to a category of statutory subject matter (a process, machine manufacture, or composition of matter)? [see MPEP § 2106.03]
Claims are directed to statutory subject matter, specifically a method (1, and 6-9).
Step 2A Prong One: Do the claims recite a judicially recognized exception, i.e., an abstract idea, a law of nature, or a natural phenomenon? [see MPEP § 2106.04(a)]
With respect to the Step 2A Prong One evaluation, the instant claims are found herein to recite abstract ideas that fall into the grouping of mental processes and mathematical concepts.
The following claims recite abstract ideas (mental processes and mathematical concepts):
Claims 1 and 18: Filtering the one or more candidate data, and predicting a binding residue are processes of comparing/contrasting, identifying, selecting, and calculating information that can be done via pen and paper or within the human mind and are therefore abstract ideas, specifically mental processes. The neural network model sharing some parameters with a second neural network is merely further limiting the data itself which is an abstract idea, specifically a mental process.
Claim 2: Candidate data including the specified data is merely further limiting the data itself which is an abstract idea, specifically a mental process. Using an algorithm to predict the binding site in a protein structure is process of calculating information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 3: The model including the specified sub neural networks, and the second network sharing at least one of the sub networks with the first network is merely further limiting the data itself which is an abstract idea, specifically a mental process.
Claim 4: The first neural network including a 3D CNN, the second sub neural network including a geometric attention layer, and the first neural network including a third sub network is merely further limiting the data itself which is an abstract idea, specifically a mental process.
Claim 5: Applying a grid alignment with the first sub network is a process of calculating information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 6: Randomly transforming an orientation of a residue via the second sub network is a process of selecting and calculating information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claims 7 and 19: Calculating a score for each of the candidate data is a process of calculating information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 8: Predicting a training binding site based on the first training data, and predicting a training binding site based on the second training data are processes of calculating information that can be done via pen and paper or within the human mind and are therefore an abstract ideas, specifically mental processes.
Claim 9: The first neural network model being the type of model specified is merely further limiting the data itself which is an abstract idea, specifically a mental process.
Claim 10: Setting some parameters shared with the trained second neural network as an initial condition is a process of identifying and selecting information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 11: Aligning the training data and external data, and assigning first sub data are processes of calculating and selecting information that can be done via pen and paper or within the human mind and are therefore abstract ideas, specifically mental processes.
Claim 13: The ground truth data including the specified data, and assigning the first sub data based on the ground truth data are merely further limiting the data itself which is an abstract idea, specifically a mental process. Assigning the center of the ground truth binding site is a process of calculating and selecting information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 14: Measuring the distance between the obtained center of the candidate binding site and the center of the binding site corresponding to the training data, and assigning eh center of the ground truther to the center of the binding site are processes of calculating, comparing/contrasting, and selecting information that can be done via pen and paper or within the human mind and are therefore abstract ideas, specifically mental processes.
Claim 15: the ground truth data including the specified data is merely further limiting the data itself which is an abstract idea, specifically a mental process. Assigning the center of the ground truth site to the center of the binding site, calculating a ratio, and assigning the center of the ground truth based on the calculated ratio, are processes of identifying, selecting, comparing/contrasting, and calculating information that can be done via pen and paper or within the human mind and are therefore abstract ideas, specifically mental processes.
Claim 16: The ground truth data including the specified data is merely further limiting the data itself which is an abstract idea, specifically a mental process. Assigning the first ground truth binding residue to the specified data is a process of selecting information that can be done via pen and paper or within the human mind and is therefore an abstract idea, specifically a mental process.
Claim 17: The ground truth data including the specified data is merely further limiting the data itself which is an abstract idea, specifically a mental process. Assigning the first ground truth to the specified data, and assigning the second ground truth to the specified data are process of selecting information that can be done via pen and paper or within the human mind and are therefore abstract ideas, specifically mental processes.
Step 2A Prong Two: If the claims recite a judicial exception under prong one, then is the judicial exception integrated into a practical application? [see MPEP § 2106.04(d)]
Because the claims do recite judicial exceptions, direction under Step 2A Prong Two provides that the claims must be examined further to determine whether they integrate the abstract ideas into a practical application.
The following claims recite the following additional elements in the form of non-abstract elements:
Claims 1 and 18: A computing device, computer program, computer-readable storage medium, and processors are generic and nonspecific elements of a computer that do not improve the functioning of any computer or technology described herein [See MPEP § 2106.04(d)(1) and MPEP § 2106.05(d)]. Obtaining one or more candidate data is an insignificant extra solution activity, specifically mere data gathering (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 2: Obtaining one or more candidate data is an insignificant extra solution activity, specifically mere data gathering (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 7: Obtaining the filtered candidate data based on the score for each is an insignificant extra solution activity, specifically necessary data outputting (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 8: Obtaining first training data, training the first neural network model, obtaining second training data, and training the second neural network model are insignificant extra solution activities, specifically mere data gathering (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 10: Performing training based on the initial condition is an insignificant extra solution activity, specifically mere data gathering (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 11: A computing device is a generic and nonspecific element of a computer that does not improve the functioning of any computer or technology described herein [See MPEP § 2106.04(d)(1) and MPEP § 2106.05(d)]. Obtaining training data, ground truth data, and external data, and obtaining aligned data are insignificant extra solution activities, specifically mere data gathering and necessary data outputting (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 12: Obtaining amino acid sequence data, and obtaining aligned external data are insignificant extra solution activities, specifically mere data gathering and necessary data outputting (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 15: Obtaining a center of a candidate binding site, and obtaining a center of the binding site are insignificant extra solution activities, specifically mere data gathering and necessary data outputting (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Claim 17: Obtaining the second augmented data is an insignificant extra solution activity, specifically necessary data outputting (See Performing clinical tests on individuals to obtain input for an equation, In re Grams, 888 F.2d 835, 839- 40; 12 USPQ2d 1824, 1827-28 (Fed. Cir. 1989) and Determining the level of a biomarker in blood, Mayo, 566 U.S. at 79, 101 USPQ2d at 1968. See also PerkinElmer, Inc. v. Intema Ltd., 496 Fed. App'x 65, 73, 105 USPQ2d 1960, 1966 (Fed. Cir. 2012) (assessing or measuring data derived from an ultrasound scan, to be used in a diagnosis)) [See MPEP § 2106.05(g)].
Step 2B: If the claims do not integrate the judicial exception, do the claims provide an inventive concept? [see MPEP § 2106.05]
Because the additional claim elements do not integrate the abstract ideas into a practical application, the claims are further examined under Step 2B, which evaluates whether the additional elements, individually and in combination, amount to significantly more than the judicial exception itself by providing an inventive concept.
The claims do not recite additional elements that are sufficient to amount to significantly more than the judicial exceptions because the claims recite additional elements that are generic, conventional, nonspecific, or insignificant extra solution activity. These additional elements include:
The additional elements of a computing device, computer program, computer-readable storage medium, and processors are all generic and nonspecific elements of a computer that are well-understood, routine and conventional within the art and therefore do not improve the functioning of any computer or technology described therein (Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information), Performing repetitive calculations, Flook, 437 U.S. at 594, 198 USPQ2d at 199 (recomputing or readjusting alarm limit values), and Storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015)) [See MPEP § 2106.05(d)(II)]. Therefore, taken both individually and as a whole, the additional elements do not amount to significantly more than the judicial exception by providing an inventive concept.
The additional elements of obtaining one or more candidate data, obtaining the filtered candidate data based on the score for each, obtaining first training data, training the first neural network model (Conventional: Basheer et al. Page 7, Section 2.6), obtaining second training data, training the second neural network model (Conventional: Basheer et al. Page 7, Section 2.6), performing training based on the initial condition (Conventional: Basheer et al. Page 7, Section 2.6), obtaining training data, ground truth data, and external data, obtaining aligned data, obtaining amino acid sequence data, obtaining aligned external data, obtaining a center of a candidate binding site, obtaining a center of the binding site, and obtaining the second augmented data are all insignificant extra solution activities, specifically mere data gathering and necessary data outputting (See Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information), TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission), OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network), buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network), and Storing and retrieving information in memory, Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015)) [See MPEP § 2106.05(g)]. Therefore, taken both individually and as a whole, the additional elements do not amount to significantly more than the judicial exception by providing an inventive concept.
Therefore, claims 1-19, when the limitations are considered individually and as a whole, are rejected under 35 U.S.C. § 101 as being directed to non-statutory subject matter.
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.
Claims 1-2, and 7-19 are rejected under 35 U.S.C. 103 as being unpatentable over Aggarwal et al. (Journal of Chemical Information and Modeling (2022) 5069-5079) and Jumper et al. (Nature (2021) 583-589).
Claim 1 is directed to a method for predicting a binding site of a protein using multiple neiral networks to filter and predict a binding residue.
Claim 18 is directed to a CRM for predicting a binding site of a protein using multiple neiral networks to filter and predict a binding residue.
Aggarwal et al. teaches in the abstract “Although several methods for detecting binding sites have been developed beforehand, a majority of them surprisingly fail in the identification and ranking of binding sites accurately. The rapid adoption and success of deep learning algorithms in various sections of structural biology beckons the usage of such algorithms for accurate binding site detection. As a combination of geometry based software and deep learning, we report a novel framework, DeepPocket that utilizes 3D convolutional neural networks for the rescoring of pockets identified by Fpocket and further segments these identified cavities on the protein surface”, reading on a method for predicting a binding site of a protein, the method performed by a computing device. Aggarwal et al. teaches on page 5071, column 1, paragraph 3 “In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs”, reading on obtaining one or more candidate data. Aggarwal et al. teaches on page 5071, column 1, paragraph 2 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure”, reading on filtering the one or more candidate data, and obtaining the filtered candidate data, by using a first neural network model for detecting a binding site and predicting a binding residue based on the filtered candidate data by using a second neural network model for identifying the binding residue.
Aggarwal et al. does not teach the sharing parameters between the first and second models.
Jumper et al. teaches in the abstract “we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known”, on page 590, column 1, paragraph 2 “The IPA module combines the pair representation, the single representation and the geometric representation to update the single representation (Supplementary Fig. 8). Each of these representations contributes affinities to the shared attention weights and then uses these weights to map its values to the output. The IPA operates in 3D space. Each residue produces query points, key points and value points in its local frame. These points are projected into the global frame using the backbone frame of the residue in which they interact with each other. The resulting points are then projected back into the local frame. The affinity computation in the 3D space uses squared distances and the coordinate transformations ensure the invariance of this module with respect to the global frame”, and on supplemental information page 51, paragraph 1 “As a dramatic example, removal of Invariant Point Attention (IPA) from the model has only a tiny effect on accuracy for recent PDB, but removing both recycling and IPA from the model has a much larger effect than removing recycling alone. This double-ablation must be interpreted with caution, however, as it is not clear which missing features of recycling (embedding of intermediate structures, much deeper network, intermediate losses, or weight tying) explain the severe ablation”, reading on wherein the first neural network model shares some parameters with the second neural network model.
It would have been obvious at the time of first filing to have modified the teachings o Aggarwal et al. for the method of claim 1 with the teachings of Jumper et al. for the use of weight tying between neural network models as the latter teaches in the abstract “We validated an entirely redesigned version of our neural network-based model, AlphaFold, in the challenging 14th Critical Assessment of protein Structure Prediction (CASP14)15, demonstrating accuracy competitive with experimental structures in a majority of cases and greatly outperforming other methods. Underpinning the latest version of AlphaFold is a novel machine learning approach that incorporates physical and biological knowledge about protein structure, leveraging multi-sequence alignments, into the design of the deep learning algorithm”. One would have had a reasonable expectation of success given that this would merely be a simple substitution of one known (the separate networks of Aggarwal et al.) method for another (the IPA weight tying model of Jumper et al.) each with expected outcomes (DeepPocket and AlphaFold respectively). 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 that the candidate data include the specified data and that the site is predicted using an algorithm.
Aggarwal et al. teaches on page 5071, column 2, paragraph 6 “We ran Fpocket across all data sets to generate data for the classification and segmentation models. Fpocket version 3.0 was used to extract candidate pocket centers for all protein structures”, reading on wherein the one or more candidate data includes: one or more candidate binding sites of a protein, or a center of at least one of the candidate binding sites, and wherein the obtaining the one or more candidate data includes: obtaining the one or more candidate binding sites of a protein, or the center of at least one of the candidate binding sites, by using an algorithm for predicting the binding site in a protein structure.
Claim 7 is directed to the method of claim 1 but further specifies the calculating of a score for the candidate data for the binding site and obtaining filtered data based on this.
Claim 19 is directed to the method of claim 1 but further specifies the calculating of a score for the candidate data for the binding site and obtaining filtered data based on this.
Aggarwal et al. teaches on page 5071, column 2, paragraph 6 “We ran Fpocket across all data sets to generate data for the classification and segmentation models. Fpocket version 3.0 was used to extract candidate pocket centers for all protein structures. We used Fpocket as it is a geometry-based software that can detect pocket curvatures in most protein structures with high accuracy and therefore has good recall. Specifically, Fpocket found pocket centers that were within 4 Å of any ligand atom for 96.4% of ligands in scPDB, 80.78% of ligands in COACH420, 87.62% of ligands in HOLO4k, 91.64% of ligands in SC6K, and 95.89% of ligands in the Refined set… Any center that was within 4 Å of any ligand atom was marked as a positive data point and the rest were marked as negative”, reading on wherein the filtering the one or more candidate data, and obtaining the filtered candidate data, by using the first neural network model for detecting a binding site includes: calculating a score for each of the one or more candidate data by using the first neural network model for detecting the binding site; and obtaining the filtered candidate data based on the score for each of the candidate data.
Claim 8 is directed to the method of claim 1 but further specifies the use of training data along with ground truth data for the training of each network.
Aggarwal et al. teaches on page 5071, column 1, paragraph 3 “In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, reading on wherein the first neural network model is a model trained based on:an operation of obtaining first training data and first ground truth data corresponding to the first training data; an operation of predicting a training binding site based on the first training data using the first neural network model; and an operation of training the first neural network model based on the predicted training binding site and the first ground truth data, and wherein the second neural network model is a model trained based on: an operation of obtaining second training data and second ground truth data corresponding to the second training data; an operation of predicting a training binding residue based on the second training data using the second neural network model; and an operation of training the second neural network model based on the predicted training binding residue and the second ground truth data.
Claim 9 is directed to the method of claim 8 and thus claim 1, but further specifies training the first network after the training of the second network.
Jumper et al. teaches on page 46, paragraph 1 of the Supplemental Information , “First, we train model 1 with a lower crop size and number of sequences for about a week. Then, we fine-tune this model with more sequences, enabled violation losses, and a lower learning rate. For historical reasons, we also added the “experimentally resolved” predictor (subsubsection 1.9.10) at this stage. We fine-tune two models: model 1.1 uses templates, while model 1.2 does not (the parameters pertaining to templates are dropped in this case). To obtain the final five models, we perform a second round of fine-tuning with larger sequence crop size and more extra MSA sequences from these two models”, here the models are trained separately, and it would be obvious to optimize which ever of the two models to train first to optimize the performance of the overall method, thereby reading on wherein the first neural network model is a model trained based on an operation of performing training for the first neural network model after performing a training process of the second neural network model.
Claim 10 is directed to the method of claim 9 and thus claim 1, but further specifies that the model shares some parameters between networks and that some of the parameters are set to an initial condition.
Jumper et al. teaches on page 4 “The trunk of the network is followed by the structure module that introduces an explicit 3D structure in the form of a rotation and translation for each residue of the protein (global rigid body frames). These representations are initialized in a trivial state with all rotations set to the identity and all positions set to the origin, but rapidly develop and refine a highly accurate protein structure with precise atomic details”, on page 590, column 1, paragraph 2 “The IPA module combines the pair representation, the single representation and the geometric representation to update the single representation (Supplementary Fig. 8). Each of these representations contributes affinities to the shared attention weights and then uses these weights to map its values to the output. The IPA operates in 3D space. Each residue produces query points, key points and value points in its local frame. These points are projected into the global frame using the backbone frame of the residue in which they interact with each other. The resulting points are then projected back into the local frame. The affinity computation in the 3D space uses squared distances and the coordinate transformations ensure the invariance of this module with respect to the global frame”, and on supplemental information page 51, paragraph 1 “As a dramatic example, removal of Invariant Point Attention (IPA) from the model has only a tiny effect on accuracy for recent PDB, but removing both recycling and IPA from the model has a much larger effect than removing recycling alone. This double-ablation must be interpreted with caution, however, as it is not clear which missing features of recycling (embedding of intermediate structures, much deeper network, intermediate losses, or weight tying) explain the severe ablation”, reading on wherein the first neural network model shares some parameters with the trained second neural network model, wherein the operation of performing training for the first neural network model after performing the training process of the second neural network model includes: an operation of setting some parameters shared with the trained second neural network model as an initial condition, and an operation of performing the training for the first neural network model based on the set initial condition.
Claim 11 is directed to a method for predicting a binding site of a protein using training data, ground truth data, and external data to align and produce augmented data for training and then predicting the binding residue.
Aggarwal et al. teaches on page 5071, column 1, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, on page 5071, column 2, paragraph 3 “We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark”, reading on a method for predicting a binding site of a protein, the method performed by a computing device, the method comprising: obtaining training data, ground truth data corresponding to the training data, and external data; aligning the training data and the external data, and obtaining aligned external data corresponding to the training data; and assigning first sub data of the training data as sub data of the aligned external data based on the ground truth data to obtain augmented data, wherein the augmented data is used in a process of training at least one of a first neural network model for detecting the binding site or a second neural network model for identifying the binding residue.
Claim 12 is directed to the method of claim 11 but further specifies that when the obtained amino acid data is conserved in the external data with the training data, then aligned external data corresponding to the training data is obtained.
Aggarwal et al. teaches on page 5071, column 2, paragraph 5 “We used stringent conditions on external data sets to avoid data leakage. Concretely, we removed all proteins from the training set that had either sequence identity greater than 50% or ligand similarity greater than 0.9 and sequence identity greater than 30% to any of the structures in the test set. This ensured that the proteins in the training and test sets neither had similar global structure nor similar binding sites”. This claim seems inherent to the idea of conservation and a tautology, i.e. a conserved sequence is identical to sequence being compared to, so if a conserved sequence of external data was obtained, that is conserved compared to the training data, then the external data corresponding to the training data is inherently aligned, thus reading on wherein the aligning the training data and the external data, and obtaining the aligned external data corresponding to the training data includes: obtaining an amino acid sequence associated with the training data; and when the obtained amino acid sequence is conserved in the aligned external data with the training data, obtaining aligned external data corresponding to the training data.
Claim 13 is directed to the method of claim 11 but further specifies that the ground truth data includes one or more ground truth binding sites, the training data includes training protein structure, and the augmented data includes assigning a center of the ground truth site to a center of a binding site in the external data.
Aggarwal et al. teaches on page 5071, column 1, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, on page 5071, column 2, paragraph 3 “We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark”, reading on wherein the training data includes a training protein structure, and wherein the ground truth data includes one or more ground truth binding sites for the training protein structure and a center of each of the ground truth binding sites, and wherein the assigning the first sub data of the training data as the sub data of the aligned external data based on the ground truth data to obtain the augmented data includes: assigning the center of the ground truth binding site to a center of a binding site included in the aligned external data to obtain first augmented data.
Claim 14 is directed to the method of claim 13 and thus claim 11, but further specifies the obtaining centers of binding sites using an algorithm for predicting binding site, measuring the distances between the obtained centers, and when that distance is within a predetermined threshold assigning the center as the ground truth.
Aggarwal et al. teaches on page 5071, column 1, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, on page 5071, column 2, paragraph 3 “We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark…For this task, we processed the data set using the pocket centers predicted by Fpocket. Any center that was within 4 Å of any ligand atom was marked as a positive data point and the rest were marked as negative. Due to the presence of redundant proteins in the scPDB database, we processed the data set to take account of all the binding sites for a redundant protein and labeled the candidate pocket centers from Fpocket according to the information from all of these binding sites. We had a total of 5 18 460 data points across the cross-validation set, out of which, 22 030 (4.25%) were positive and the rest negative”, reading on wherein the assigning the center of the ground truth binding site to the center of the binding site corresponding to the training data included in the aligned external data to obtain the first augmented data includes: obtaining a center of a candidate binding site by using an algorithm for predicting a binding site in a protein structure for the external data; obtaining a center of the binding site of the aligned external data corresponding to the training data; measuring a distance between the obtained center of the candidate binding site of the aligned external data and the center of the binding site corresponding to the training data; and when the measured distance is within a predetermined threshold, assigning the center of the ground truth binding site to the center of the binding site of the aligned external data corresponding to the training data to obtain the first augmented data.
Claim 15 is directed to the method of claim 13 and thus claim 11, but further specifies calculation of a ratio between the aligned external data corresponding to the training data corresponding to the ground truth, and assigning the ground truth based on said ratio.
Aggarwal et al. teaches on page 5071, column2, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark…For this task, we processed the data set using the pocket centers predicted by Fpocket. Any center that was within 4 Å of any ligand atom was marked as a positive data point and the rest were marked as negative. Due to the presence of redundant proteins in the scPDB database, we processed the data set to take account of all the binding sites for a redundant protein and labeled the candidate pocket centers from Fpocket according to the information from all of these binding sites. We had a total of 5 18 460 data points across the cross-validation set, out of which, 22 030 (4.25%) were positive and the rest negative”,and on page 5073, column 2, paragraph 2 “Known binding site residues can be defined as the set of all residues that are within a distance threshold from any ligand atom. The ratio can be represented mathematically as [Equation 3]”, reading on wherein the ground truth data further includes a ground truth binding residue for each of the ground truth binding sites, and the assigning the center of the ground truth binding site to the center of the binding site of the aligned external data corresponding to the training data to obtain the first augmented data includes: calculating a ratio at which the binding residue of the aligned external data corresponding to the training data corresponds to the ground truth binding residue; and when the calculated ratio is equal to or more than a predetermined threshold, assigning the center of the ground truth binding site to the center of the binding site of the aligned external data corresponding to the training data to obtain the first augmented data.
Claim 16 is directed to the method of claim 11 but further specifies assigning the first ground truth binding residue to the external data corresponding to the training data in order to obtain augmented data.
Aggarwal et al. teaches on page 5071, column 1, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, on page 5071, column 2, paragraph 3 “We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark…For this task, we processed the data set using the pocket centers predicted by Fpocket. Any center that was within 4 Å of any ligand atom was marked as a positive data point and the rest were marked as negative. Due to the presence of redundant proteins in the scPDB database, we processed the data set to take account of all the binding sites for a redundant protein and labeled the candidate pocket centers from Fpocket according to the information from all of these binding sites. We had a total of 5 18 460 data points across the cross-validation set, out of which, 22 030 (4.25%) were positive and the rest negative”, reading on wherein the training data includes a first training protein structure, wherein the ground truth data includes a first ground truth binding residue for the first training protein structure, and wherein the assigning the first sub data of the training data as the sub data of the aligned external data based on the ground truth data to obtain the augmented data includes: assigning the first ground truth binding residue to the binding residue of the aligned external data corresponding to the training data to obtain second augmented data.
Claim 17 is directed to the method of claim 16 and thus claim 11, but further specifies the use of a second protein structure with a second ground truth binding residue, assigning the corresponding ground truths to the corresponding binding residues, and obtaining augmented data.
Aggarwal et al. teaches on page 5071, column 1, paragraph 2-3 “We follow a multistep approach to get the final pocket location and 3D shape prediction from the input protein structure. We first clean the input structure by removing all heteroatoms and solvent molecules from the protein structure using the Biopython library. Then, we run Fpocket across the structure and calculate the barycenter of each predicted pocket. These become the candidate pocket centers that need to be ranked by the CNN scoring function. Therefore, constant-sized grids are placed at each barycenter followed by scoring using the CNN. The top-ranked centers are then sent through a CNN segmentation model to get the final pocket structure…In this work, we used the scPDB v.2017 database49 for training and cross-validation of the model. It contains 17 594 binding sites, which corresponds to 16 612 proteins and 5540 UniProt IDs. Different versions of the scPDB data set have been used previously by other binding site detection methods such as Kalasanty and DeepSite. The proteins, ligands, and 3D shapes of binding site structures in the data set were generated by Volsite. Volsite projects the protein on a 3D grid lattice of resolution 2 Å, after which, each accessible voxel is assigned a pharmacophoric property based on the nearest protein atom. Since each voxel has a fixed volume, cavities are estimated by counting these pharmacophore-annotated voxels. The data set was cleaned and split in a similar way as that used for Kalasanty, into 10 cross validation sets based on their Uniprot IDs to avoid data leakage”, on page 5071, column 2, paragraph 3 “We developed a new data set, SC6K by taking all the protein structures submitted to the Protein Data Bank from January 1st, 2018, until February 28th, 2020, that contained at least one ligand. These were run through the pdbconv tool from the IChem Toolkit51 to get a data set with the same filters and site selection algorithm as scPDB…The Refined set is constructed by taking the union of the Refined sets of v2007, v2013, v2015, and v2016 from PDBbind. The data set consists of 4414 protein−ligand pairs. On running Volsite across this data set, we obtain ground truth segmentation masks for 2793 (63.28%) protein−ligand complexes. These masks were used to obtain the best performing model checkpoint from training for the substructure benchmark…For this task, we processed the data set using the pocket centers predicted by Fpocket. Any center that was within 4 Å of any ligand atom was marked as a positive data point and the rest were marked as negative. Due to the presence of redundant proteins in the scPDB database, we processed the data set to take account of all the binding sites for a redundant protein and labeled the candidate pocket centers from Fpocket according to the information from all of these binding sites. We had a total of 5 18 460 data points across the cross-validation set, out of which, 22 030 (4.25%) were positive and the rest negative”, reading on wherein the training data further includes a second training protein structure, wherein the ground truth data includes a second ground truth biding residue for the second training protein structure, and wherein the assigning the first ground truth binding residue to the binding residue of the aligned external data corresponding to the training data to obtain the second augmented data includes: assigning the first ground truth binding residue to a first binding residue of the aligned external data corresponding to the training data to obtain a first assigned binding residue; assigning the second ground truth binding residue to a second binding residue of the aligned external data corresponding to the training data to obtain a second assigned binding residue; and obtaining the second augmented data based on the first assigned binding residue and the second assigned binding residue.
Claims 3-5 are rejected under 35 U.S.C. 103 as being unpatentable over Aggarwal et al. (Journal of Chemical Information and Modeling (2022) 5069-5079) and Jumper et al. (Nature (2021) 583-589) as applied to claim 1-2, and 7-19 above, and further in view of Zeng et al. (Bioinformatics (2020) 1114-1120).
Claim 3 is directed to the method of claim 1 but further specifies that the first neural network comprise at least a first and second sub network and that the sub network is shared with the second neural network.
Aggarwal et al. and Jumper et al. teach the method of claim 1.
Aggarwal et al. teaches on page 5070, in Figure 1 “DeepPocket: Ligand binding site detection using 3D CNNs: A protein structure taken from Protein Data Bank is passed to the geometry based software Fpocket, which detects candidate pockets. DeepPocket then uses 3D CNNs to learn the pocket structure and outputs a decreasing list of pockets probabilities with the most druggable pocket at the top”, and in Table S6 includes 8 sub networks as part of the first network.
Jumper et al. teaches in the abstract “we provide the first computational method that can regularly predict protein structures with atomic accuracy even in cases in which no similar structure is known”, on page 590, column 1, paragraph 2 “The IPA module combines the pair representation, the single representation and the geometric representation to update the single representation (Supplementary Fig. 8). Each of these representations contributes affinities to the shared attention weights and then uses these weights to map its values to the output. The IPA operates in 3D space. Each residue produces query points, key points and value points in its local frame. These points are projected into the global frame using the backbone frame of the residue in which they interact with each other. The resulting points are then projected back into the local frame. The affinity computation in the 3D space uses squared distances and the coordinate transformations ensure the invariance of this module with respect to the global frame”, and on supplemental information page 51, paragraph 1 “As a dramatic example, removal of Invariant Point Attention (IPA) from the model has only a tiny effect on accuracy for recent PDB, but removing both recycling and IPA from the model has a much larger effect than removing recycling alone. This double-ablation must be interpreted with caution, however, as it is not clear which missing features of recycling (embedding of intermediate structures, much deeper network, intermediate losses, or weight tying) explain the severe ablation”, reading on wherein the second neural network model shares at least one of the first sub neural network or the second sub neural network with the first neural network model.
Aggarwal et al. and Jumper et al. do not teach the extraction of local features by one network that are then reconstructed globally by another.
Zeng et al. teaches in the abstract “Existing computational methods usually use local contextual features to predict PPI sites. Actually, global features of protein sequences are critical for PPI site prediction. A new end-to-end deep learning framework, named DeepPPISP, through combining local contextual and global sequence features, is proposed for PPI site prediction. For local contextual features, we use a sliding window to capture features of neighbors of a target amino acid as in previous studies. For global sequence features, a text convolutional neural network is applied to extract features from the whole protein sequence. Then the local contextual and global sequence features are combined to predict PPI sites”, and in Figure 1 “The deep neural network structure of DeepPPISP. The input consists of two types of data: subsequence and whole protein sequence. For subsequence, a sliding window size of 7 is applied to extract features of neighbors of a target amino acid. Then the combination of three types of features is flattened a local feature vector. For the whole protein sequence, the length (L) is set to 500. First, through raw sequence feature embedding, three types of features are concatenated to a preprocessed vector, which is fed into a TextCNN with different kernels (13, 15 and 17) to obtain a global sequence feature vector. The local and global feature vectors are concatenated, and then the concatenated vector is fed into two fully connected layers for prediction”, reading on wherein the first neural network model includes: a first sub neural network for extracting a local feature of the binding site; and a second sub neural network for globally aggregating the local features.
It would have been obvious at the time of first filing to have modified the teachings of Aggarwal et al. and Jumper et al. for the method of claim 1 and the use of multiple sub networks with the teachings of Zeng et al. for the extraction of local features using a first network and the generation of global features using said local features via a second network as the latter teaches in the abstract “DeepPPISP achieves the state-of-the-art performance, which is better than the other competing methods”. One would have had a reasonable expectation of success given that this would merely be a simple substitution of one known (the shared sub networks of Aggarwal et al. and Jumper et al.) method for another (a local feature extraction network and a global feature network of Zeng et al.) each with expected outcomes (DeepPocket/AlphaFold and DeepPPISP respectively). Therefore, it would have been obvious at the time of first filing to have modified the teachings of each and to be successful.
Claim 4 is directed to the method of claim 3 and thus claim 1, but further specifies that the network include 3D convolutional network, geometric attention layer, and third scalar mapping network.
Aggarwal et al. teaches in the abstract “As a combination of geometry based software and deep learning, we report a novel framework, DeepPocket that utilizes 3D convolutional neural networks”, on page 5070, in Figure 1 “DeepPocket: Ligand binding site detection using 3D CNNs: A protein structure taken from Protein Data Bank is passed to the geometry based software Fpocket, which detects candidate pockets. DeepPocket then uses 3D CNNs to learn the pocket structure and outputs a decreasing list of pockets probabilities with the most druggable pocket at the top”, and in Table S6 includes 8 sub networks as part of the first network.
Jumper et al. teaches on page 584, column 2, paragraph 2 “we demonstrate a new architecture to jointly embed multiple sequence alignments (MSAs) and pairwise features, a new output representation and associated loss that enable accurate end-to-end structure prediction, a new equivariant attention architecture, use of intermediate losses to achieve iterative refinement of predictions, masked MSA loss to jointly train with the structure, learning from unlabelled protein sequences using self-distillation and self-estimates of accuracy…The Evoformer blocks contain a number of attention-based and non-attention-based components”, and on page 587, column 1, paragraph 1 “First, a geometry-aware attention operation that we term ‘invariant point attention’ (IPA) is used to update”.
Zeng et al. teaches in the abstract “Existing computational methods usually use local contextual features to predict PPI sites. Actually, global features of protein sequences are critical for PPI site prediction. A new end-to-end deep learning framework, named DeepPPISP, through combining local contextual and global sequence features, is proposed for PPI site prediction. For local contextual features, we use a sliding window to capture features of neighbors of a target amino acid as in previous studies. For global sequence features, a text convolutional neural network is applied to extract features from the whole protein sequence. Then the local contextual and global sequence features are combined to predict PPI sites”, and in Figure 1 “The deep neural network structure of DeepPPISP. The input consists of two types of data: subsequence and whole protein sequence. For subsequence, a sliding window size of 7 is applied to extract features of neighbors of a target amino acid. Then the combination of three types of features is flattened a local feature vector. For the whole protein sequence, the length (L) is set to 500. First, through raw sequence feature embedding, three types of features are concatenated to a preprocessed vector, which is fed into a TextCNN with different kernels (13, 15 and 17) to obtain a global sequence feature vector. The local and global feature vectors are concatenated, and then the concatenated vector is fed into two fully connected layers for prediction”, reading on wherein the first sub neural network for extracting the local feature of the binding site includes a 3D convolutional network, wherein the second sub neural network for globally aggregating the local features includes a geometric attention layer, and wherein the first neural network model further includes a third sub neural network for mapping the feature aggregated through the second sub neural network to a single scalar quantity.
Claim 5 is directed to the method of claim 3 and thus claim 1, but further specifies the use of a grid alignment.
Aggarwal et al. teaches in the abstract “Although several methods for detecting binding sites have been developed beforehand, a majority of them surprisingly fail in the identification and ranking of binding sites accurately. The rapid adoption and success of deep learning algorithms in various sections of structural biology beckons the usage of such algorithms for accurate binding site detection. As a combination of geometry based software and deep learning, we report a novel framework, DeepPocket that utilizes 3D convolutional neural networks for the rescoring of pockets identified by Fpocket and further segments these identified cavities on the protein surface”, reading on wherein the first sub neural network performs an operation of applying a grid alignment.
Subject Matter Potentially Free from the Prior Art
Claim 6 describes an operation of randomly transforming the orientation of a residue during the training process for predicting protein binding sites. The earliest described instance of this method is found in Lee et al. published 4/18/2023 which is not considered prior art as the effective filing date is 2/16/2023.
Conclusion
No claims are allowed.
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
