DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Applicant’s amendments and arguments filed 26 May 2026 are acknowledged and entered.
Withdrawn Rejections/Objections
The rejection of claims 5, 12, and 19 under 35 U.S.C. §102 over Wang in the Office action mailed 25 February 2026 is withdrawn in view of the amendments filed 26 May 2026.
Rejections and/or objections not reiterated from previous office actions are hereby withdrawn. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
Claim Status
Claims 5-6, 12-13, and 19-20 were amended by Applicant’s paper filed 26 May 2026 (hereinafter “Amendment”).
Claims 1-25 are currently pending and under exam herein.
Claims 1-25 are rejected.
Claim Rejections - 35 USC § 102
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 the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 4, 8, 11, 15, and 18 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Wang (ACM Trans. Graph, Vol. 37, (4 December 2018)), as evidenced by TechnoLynx (The Foundation of Generative AI: Neural Networks Explained, (28 April 2025)), C3.ai. (Infrastructure: Machine Learning Hardware Requirements, (15 May 2021)), and AWS (What is a Neural Network?, (29 April 2022)).
Regarding claim 1, Wang discloses a generative neural network to generate a 3D model of a 3D object. At 2 col.2 paras.2-3. Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN); and using the GNN to form a generative model of the 3D domain based at least in part on the input representations; wherein the input representations comprise a global-shape input representation of the 3D domain). While Wang does not explicitly teach computer-implementation, generative neural networks are inherently computer-implemented, as evidenced by TechnoLynx. § Introduction (a computer-implemented method comprising).
Regarding claim 4, Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (the computer-implemented method of claim 1, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Regarding claim 8, Wang discloses a generative neural network to generate a 3D model of a 3D object. At 2 col.2 paras.2-3. Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN); and using the GNN to form a generative model of the 3D domain based at least in part on the input representations; wherein the input representations comprise a global-shape input representation of the 3D domain). Wang does not explicitly state that the generative neural network is operated on a computer system comprising a memory and a processor communicatively coupled to the memory. However, these elements are inherent in the disclosure of Wang because a computer system comprising a memory and a processor are necessary for the functionality of a generative neural network, as evidenced by C3.ai. § Processors: CPUs, GPUs, TPUs, and FPGAs; § Memory and Storage.
Regarding claim 11, Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (the computer system of claim 8, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Regarding claim 15, Wang discloses a generative neural network to generate a 3D model of a 3D object. At 2 col.2 paras.2-3. Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN); and using the GNN to form a generative model of the 3D domain based at least in part on the input representations; wherein the input representations comprise a global-shape input representation of the 3D domain). Wang does not explicitly state that the generative neural network is a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor system to cause the processor system to perform operations. However, a neural network is inherently a computer program product because neural networks are software programs or algorithms that use computing systems to solve mathematical calculations, as evidenced by AWS. § How do neural networks work? para.1. Additionally, Wang inherently discloses the program instructions being executable by a processor system because execution by a processor is necessary for the functionality of a neural network, as evidenced by C3.ai. § Processors: CPUs, GPUs, TPUs, and FPGAs.
Regarding claim 18, Wang teaches applying global and local input representations of the 3D object into the generative neural network to form the 3D model. At 5 col.1 para.4; Fig.2 (the computer program product of claim 15, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Response to Arguments
Applicant first argues that the 102 rejection of claims 1, 4-5, 8, 11-12, and 18-19 is improper because the rejection relies on the teachings of more than one reference. Under MPEP § 2131.01, a 35 U.S.C. 102 rejection over multiple references has been held to be proper when the extra references are cited to: (A) prove the primary reference contains an "enabled disclosure;"(B) explain the meaning of a term used in the primary reference; or (C) show that a characteristic not disclosed in the reference is inherent. “To serve as an anticipation when the reference is silent about the asserted inherent characteristic, such gap in the reference may be filled with recourse to extrinsic evidence. Such evidence must make clear that the missing descriptive matter is necessarily present in the thing described in the reference, and that it would be so recognized by persons of ordinary skill.” Continental Can Co. USA v. Monsanto Co., 948 F.2d 1264, 1268, 20 USPQ2d 1746, 1749-50 (Fed. Cir. 1991); and MPEP § 2131.01(III).
In the instant application, Wang describes a generative neural network to generate a 3D model of a 3D object. Wang is silent about computer-implementation, a computer system comprising a memory and a processor communicatively coupled to the memory, and a computer program product comprising a computer readable storage medium having processor executable program instructions embodied therewith. TechnoLynx, C3.ai., and AWS make clear that generative neural networks such as the one described by Wang are inherently computer-implemented, and one of ordinary skill in the art would recognize that a generative neural network is stored as program instructions to be executed by a processor. While Applicant asserts that the claims do not merely recite a generic neural network, the extra references were not used to equate the claimed invention to a generic neural network but were instead relied upon to show that a person having ordinary skill in the art would recognize that neural networks are inherently computer implemented and stored as software for execution by a processor. Therefore, Applicant’s argument is not persuasive because the multiple reference 102 rejection is proper when the extra references are cited to show that a characteristic not explicitly disclosed in the reference is inherent.
Applicant also argues that Wang’s global stage in a 3D shape generation pipeline is not equivalent to the claimed global-shape input representation because Wang does not disclose the claimed data flow relationship. Applicant asserts that a generated global shape in Wang occupies a different position in Wang’s data flow from the claimed global-shape input representation. In the instant application, claim 1 recites applying global-shape input representations of a 3D domain to a generative neural network to model the 3D domain. The instant specification limits “global-shape input representation” to input that provides information that is valid for the entire domain. See para. [0054]. Wang discloses that the global discriminator takes a whole image as input to assess whether it is coherent as a whole. At 3 col.1 para.1. Wang uses whole images of an object as input to train a generative adversarial network (GAN), which constructs the overall structure of the object. At abstract; 3 col.1 para.1; Fig.1. While the input of Wang is encoded as vectors, the vectors represent the structure of a whole 3D object. At 3 col.2 para.2; 5 col.2 para.3. Therefore, Applicant’s argument is not persuasive because the GAN input of Wang constitutes a global-shape input representation when it represents the overall structure of a 3D object and provides information that is valid for the entire object.
Applicant similarly argues that Wang’s local refinement does not disclose the claimed local point level input representation because a process that refines local points of a generated 3D shape is not equivalent to applying a local point level input representation of the 3D domain to the GNN. The input of Wang is a representation of the entire object with labels for each semantic part. At 5 col.2 para.3. The GAN generates a global structure of the object, with local part labels for further refinement. At 2 col.1 para.3. While the input of Wang is a vector representation of the overall object with local parts labeled, the language of instant claims 1 and 4 (and similarly claims 8, 11, 15, and 18) does not preclude the global-shape input representation and the local point-level input representation from being part of the same input. Therefore, Applicant’s argument is not persuasive because the GAN input of Wang constitutes a local point-level input representation when it contains labels for semantic parts.
Applicant also argues that the reconstruction of Wang is inadequate to constitute a reconstructed version of the input representations. Wang does not generate any 3D shape, but instead uses 3D models of objects to generate a refined version of that object. At 5 col.2 para.3; Fig.1. Wang trains the GAN on four separate objects and generates a reconstructed version of each object. At 5 col.2 para.3. Therefore, Applicant’s argument is not persuasive because Wang discloses reconstructing four separate objects based on four separate input representations.
Applicant finally argues that the reconstruction loss of Wang does not constitute the reconstruction loss of claims 5, 12, and 19 because the loss is not tied to the reconstructed version of the input representations. Upon further consideration, Wang does not teach the encoding limitation of claims 5, 12, and 19 because the input representation of Wang is already encoded as a vector when applied to the GAN. Therefore, the 102 rejection of claims 5, 12, and 19 has been withdrawn, and the argument on the reconstruction loss is moot.
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.
Claims 1-5, 7-12, 14-19, and 21-25 are rejected under 35 U.S.C. 103 as being unpatentable over Oono (US 2017/0161635 A1), in view of Keller (ChemRxiv. (2 October 2018)) and Krishnapriyan (Sci Rep 11, no. 8888 (26 April 2021)), as evidenced by Nicolas (Geometric Deep Learning (15 December 2025)) and Xie (Front Pharmacol. (18 December 2020)).
Regarding claim 1, Oono discloses a computer implemented method where chemical compound representations in the form of fingerprints are input into a generative framework comprising a neural network. Para. [0008] (a computer-implemented method comprising: applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN)). Oono defines chemical compound fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure. Para. [0088]. Oono teaches that the generative framework can generate chemical compounds that have desired characteristics based on the input representations. Abstract; para. [0098] (using the GNN to form a generative model of the 3D domain based at least in part on the input representations).
Oono fails to teach wherein the input representation comprise a global-shape input representation of the 3D domain. However, Keller discloses a molecular representation scheme based on persistent homology. At 2 paras.5-6. Persistent homology offers a means to capture the global geometric and topological structure of complex data, as evidenced by Nicolas. § Persistent Homology. Keller notes that the persistent homology representation is intended to alleviate the significant difficulty in molecular design related to the possibility of multiple confirmations of compounds that have the same molecular formula but important structural differences (i.e. two molecules can share local descriptors, but have distinct global descriptors). At 2 para.3.
To solve a similar issue, Krishnapriyan discloses a machine learning model for metal-organic frameworks (MOFs) that automatically generates feature descriptors based on persistent homology representations and the elemental composition of the molecule. At 2 paras.4-5. Krishnapriyan notes that to comprehensively understand MOFs, it is necessary to recognize geometric and chemical features responsible for their performance in particular applications. At 1 para.3. While input features can be related to a MOF’s performance in a particular application, standard structural descriptors are not able to capture some relevant information, such as the pressure during absorption or local strong absorption sites. At 2 para.2. To overcome these challenges, Krishnapriyan discloses using a topological descriptor called persistent homology and the elemental composition of a molecule as input into a machine learning model. At 2 para.4. The elemental compositions are embedded and the persistent homology representations are translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework. Keller discloses a persistent homology representation aiming to alleviate major difficulty in molecular design regarding multiple confirmations of compounds that have the same molecular formula but important structural differences. Krishnapriyan discloses a solution to a similar issue in the context of MOFs where persistence images (i.e. global representation) and elemental compositions (i.e. local representation) are used as inputs for machine learning algorithms. One of ordinary skill in the art could apply Krishnapriyan’s technique of inputting global and local representations to the method of Oono by additionally inputting the persistent homology representation of Keller. This would predictably result in an improved method of molecular design because, similar to the results in Krishnapriyan, the incorporation of Keller’s persistent homology representation as input into Oono’s generative framework will remedy the issue of molecules consisting of the same chemical formula but different structures. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 2, Oono discloses that the generative framework may be implemented as part of a cloud computer system. Para. [0146] (the computer-implemented method of claim 1, wherein the GNN is part of a cloud computing system).
Regarding claim 3, Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning. At 3 para.1. Specifically, Krishnapriyan translates the persistent homology representations into persistence images. Id.
Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 1 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be translated into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 4, Oono discloses inputting chemical compound fingerprints into the generative framework. Para. [0009]. As evidenced by Xie, molecular fingerprints describe the local aspect of chemical structures. At 3 col.1 para.1 (the computer-implemented method of claim 1, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Regarding claim 5, Oono discloses that the generative framework can be trained to form a generative model, para. [0040], by encoding the input representations into latent representations and decoding the latent representations into reconstructions of the input representations, para. [0041] (the computer-implemented method of claim 4, wherein using the GNN to form the generative model of the 3D domain comprises: encoding, using the GNN, the input representations to generate latent code; decoding, using the GNN, the latent code to generate a reconstructed version of the input representations). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the input representations). Oono discloses inputting chemical compound fingerprints into the generative framework, and defines chemical compound fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure. Para. [0088] (wherein the local point-level input representation comprises a string input representation). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1 (the global-shape input representation comprises a persistence image input representation).
Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 1 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 7, Keller discloses a multi-parameter persistent homology representation that captures the important properties of the shapes of molecules and incorporates non-shape information in a coherent and effective manner. At 4 para.2. Keller teaches that the first parameter of the representation captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, while the second parameter may capture partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4. One of ordinary skill in the art would know to include Keller’s multi-parameter persistent homology representation as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 1 above). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
The method resulting from the combination of Oono, Keller, and Krishnapriyan would input chemical composition fingerprints and multi-parameter persistence images capturing shape and non-shape molecular information into a generative framework (the input representations further comprise an input representation of a characteristic of the 3D domain). The first parameter of the persistence image captures a set of points in Euclidean space representing atom centers in a molecule (the global-shape input representation comprises a first parameter of a multi-parameter persistence image), while the second parameter may capture partial charge, atomic mass, hydrogen donor/acceptor status, or some other non-shape molecular characteristic (the input representation of the characteristic of the 3D domain comprises a second parameter of the multi-parameter persistence image).
Regarding claim 8, Oono discloses a computer system, para. [0041], comprising a memory coupled to a bus for communicating information to a processor, which executes instructions, para. [0145], to input chemical compound fingerprints into a generative framework comprising a neural network, para. [0008] (a computer system comprising a memory and a processor communicatively coupled to the memory, wherein the processor is operable to perform operations comprising: applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN)). Oono teaches that the generative framework can generate chemical compounds that have desired characteristics based on the input representations. Abstract; para. [0098] (using the GNN to form a generative model of the 3D domain based at least in part on the input representations).
Oono fails to teach wherein the input representation comprise a global-shape input representation of the 3D domain. However, Keller discloses a molecular representation scheme based on persistent homology. At 2 paras.5-6. Keller notes that the persistent homology representation is intended to alleviate the significant difficulty in molecular design related to the possibility of multiple confirmations of compounds that have the same molecular formula but important structural differences (i.e. two molecules can share local descriptors, but have distinct global descriptors). At 2 para.3.
To solve a similar issue, Krishnapriyan discloses a machine learning model for metal-organic frameworks (MOFs) that automatically generates feature descriptors based on persistent homology representations and the elemental composition of the molecule. At 2 paras.4-5. Krishnapriyan notes that to comprehensively understand MOFs, it is necessary to recognize geometric and chemical features responsible for their performance in particular applications. At 1 para.3. While input features can be related to a MOF’s performance in a particular application, standard structural descriptors are not able to capture some relevant information, such as the pressure during absorption or local strong absorption sites. At 2 para.2. To overcome these challenges, Krishnapriyan discloses using a topological descriptor called persistent homology and the elemental composition of a molecule as input into a machine learning model. At 2 para.4. The elemental compositions are embedded and the persistent homology representations are translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework. Keller discloses a persistent homology representation aiming to alleviate major difficulty in molecular design regarding multiple confirmations of compounds that have the same molecular formula but important structural differences. Krishnapriyan discloses a solution to a similar issue in the context of MOFs where persistence images (i.e. global representation) and elemental compositions (i.e. local representation) are used as inputs for machine learning algorithms. One of ordinary skill in the art could apply Krishnapriyan’s technique of inputting global and local representations to the method of Oono by additionally inputting the persistent homology representation of Keller. This would predictably result in an improved method of molecular design because, similar to the results in Krishnapriyan, the incorporation of Keller’s persistent homology representation as input into Oono’s generative framework will remedy the issue of molecules consisting of the same chemical formula but different structures. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 9, Oono discloses that the generative framework may be implemented as part of a cloud computer system. Para. [0146] (the computer system of claim 8, wherein the GNN is part of a cloud computing system).
Regarding claim 10, Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 8 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be translated into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 11, Oono discloses inputting chemical compound fingerprints into the generative framework. Para. [0009]. As evidenced by Xie, molecular fingerprints describe the local aspect of chemical structures. At 3 col.1 para.1 (the computer system of claim 8, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Regarding claim 12, Oono discloses that the generative framework can be trained to form a generative model, para. [0040], by encoding the input representations into latent representations and decoding the latent representations into reconstructions of the input representations, para. [0041] (the computer system of claim 11, wherein using the GNN to form the generative model of the 3D domain comprises: encoding, using the GNN, the input representations to generate latent code; decoding, using the GNN, the latent code to generate a reconstructed version of the input representations). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the input representations). Oono discloses inputting chemical compound fingerprints into a generative framework, para. [0008], and defines chemical compound fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure, para. [0088] (wherein the local point-level input representation comprises a string input representation). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1 (wherein the global-shape input representation comprises a persistence image input representation).
Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 8 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 14, Keller discloses a multi-parameter persistent homology representation that captures the important properties of the shapes of molecules and incorporates non-shape information in a coherent and effective manner. At 4 para.2. Keller teaches that the first parameter of the representation captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, while the second parameter may capture partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4. One of ordinary skill in the art would know to include Keller’s multi-parameter persistent homology representation as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 8 above). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
The method resulting from the combination of Oono, Keller, and Krishnapriyan would input chemical composition fingerprints and multi-parameter persistence images capturing shape and non-shape molecular information into a generative framework (the computer system of claim 12, wherein: the input representations further comprise an input representation of a characteristic of the 3D domain). The first parameter of the persistence image captures a set of points in Euclidean space representing atom centers in a molecule (the global-shape input representation comprises a first parameter of a multi-parameter persistence image), while the second parameter may capture partial charge, atomic mass, hydrogen donor/acceptor status, or some other non-shape molecular characteristic (the input representation of the characteristic of the 3D domain comprises a second parameter of the multi-parameter persistence image).
Regarding claim 15, Oono discloses a computer program stored in a computer readable storage medium with instructions, para. [0143], that are executable by a communicably coupled processor, which executes instructions, para. [0145], to input chemical compound fingerprints into a generative framework comprising a neural network, para. [0008] (a computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor system to cause the processor system to perform operations comprising: applying input representations of a three-dimensional (3D) domain to a generative neural network (GNN)). Oono teaches that the generative framework can generate chemical compounds that have desired characteristics based on the input representations. Abstract; para. [0098] (using the GNN to form a generative model of the 3D domain based at least in part on the input representations).
Oono fails to teach wherein the input representation comprise a global-shape input representation of the 3D domain. However, Keller discloses a molecular representation scheme based on persistent homology. At 2 paras.5-6. Keller notes that the persistent homology representation is intended to alleviate the significant difficulty in molecular design related to the possibility of multiple confirmations of compounds that have the same molecular formula but important structural differences (i.e. two molecules can share local descriptors, but have distinct global descriptors). At 2 para.3.
To solve a similar issue, Krishnapriyan discloses a machine learning model for metal-organic frameworks (MOFs) that automatically generates feature descriptors based on persistent homology representations and the elemental composition of the molecule. At 2 paras.4-5. Krishnapriyan notes that to comprehensively understand MOFs, it is necessary to recognize geometric and chemical features responsible for their performance in particular applications. At 1 para.3. While input features can be related to a MOF’s performance in a particular application, standard structural descriptors are not able to capture some relevant information, such as the pressure during absorption or local strong absorption sites. At 2 para.2. To overcome these challenges, Krishnapriyan discloses using a topological descriptor called persistent homology and the elemental composition of a molecule as input into a machine learning model. At 2 para.4. The elemental compositions are embedded and the persistent homology representations are translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework. Keller discloses a persistent homology representation aiming to alleviate major difficulty in molecular design regarding multiple confirmations of compounds that have the same molecular formula but important structural differences. Krishnapriyan discloses a solution to a similar issue in the context of MOFs where persistence images (i.e. global representation) and elemental compositions (i.e. local representation) are used as inputs for machine learning algorithms. One of ordinary skill in the art could apply Krishnapriyan’s technique of inputting global and local representations to the method of Oono by additionally inputting the persistent homology representation of Keller. This would predictably result in an improved method of molecular design because, similar to the results in Krishnapriyan, the incorporation of Keller’s persistent homology representation as input into Oono’s generative framework will remedy the issue of molecules consisting of the same chemical formula but different structures. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 16, Oono discloses that the generative framework may be implemented as part of a cloud computer system. Para. [0146] (the computer program product of claim 15, wherein the GNN is part of a cloud computing system).
Regarding claim 17, Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 15 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be translated into persistence images, which is proper format for machine learning input (the computer program product of claim 15, wherein the global-shape input representation of the 3D domain comprises a persistence image). Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 18, Oono discloses inputting chemical compound fingerprints into the generative framework. Para. [0009]. As evidenced by Xie, molecular fingerprints describe the local aspect of chemical structures. At 3 col.1 para.1 (the computer program product of claim 15, wherein the input representations further comprise a local point-level input representation of the 3D domain).
Regarding claim 19, Oono discloses that the generative framework can be trained to form a generative model, para. [0040], by encoding the input representations into latent representations and decoding the latent representations into reconstructions of the input representations, para. [0041] (the computer program product of claim 18, wherein using the GNN to form the generative model of the 3D domain comprises: encoding, using the GNN, the input representations to generate latent code; decoding, using the GNN, the latent code to generate a reconstructed version of the input representations). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the input representations). Oono discloses inputting chemical compound fingerprints into a generative framework, para. [0008], and defines chemical compound fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure, para. [0088] (wherein the local point-level input representation comprises a string input representation). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1 (wherein the global-shape input representation comprises a persistence image input representation).
Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 15 above). Krishnapriyan discloses vectorizing persistent homology representations via persistence images to be usable as input for machine learning. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 21, Keller discloses a multi-parameter persistent homology representation that captures the important properties of the shapes of molecules and incorporates non-shape information in a coherent and effective manner. At 4 para.2. Keller teaches that the first parameter of the representation captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, while the second parameter may capture partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4. One of ordinary skill in the art would know to include Keller’s multi-parameter persistent homology representation as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 15 above). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
The method resulting from the combination of Oono, Keller, and Krishnapriyan would input chemical composition fingerprints and multi-parameter persistence images capturing shape and non-shape molecular information into a generative framework (the computer program product of claim 19, wherein: the input representations further comprise an input representation of a characteristic of the 3D domain). The first parameter of the persistence image captures a set of points in Euclidean space representing atom centers in a molecule (the global-shape input representation comprises a first parameter of a multi-parameter persistence image), while the second parameter may capture partial charge, atomic mass, hydrogen donor/acceptor status, or some other non-shape molecular characteristic (the input representation of the characteristic of the 3D domain comprises a second parameter of the multi-parameter persistence image).
Regarding claim 22, Oono discloses a computer system, para. [0041], comprising a memory coupled to a bus for communicating information to a processor, which executes instructions, para. [0145], that generate chemical compound models having desired characteristics based on the input representations. Abstract; para. [0098] (a computer system comprising a memory and a processor communicatively coupled to the memory, wherein the processor is operable to form a generative model of a three-dimensional (3D) domain by performing operations comprising). Oono teaches that the generative framework encodes input representations into latent representations. Para. [0041] (encoding, using a generative neural network (GNN), input representations to generate latent code). Oono discloses that the latent representations are decoded in the generative framework to reconstructions of the input representations. Id (decoding, using the GNN, the latent code to generate a reconstructed version of the input representations). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the input representations). Oono discloses the input representations comprise chemical compound fingerprints, para. [0008], and defines fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure, para. [0088] (wherein the input representations comprise: a string input representation of the 3D domain).
Oono fails to teach the input representations comprising a 3D coordinate input representation of the 3D domain. However, Keller discloses a multi-parameter persistent homology representation that captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, and some other non-shape molecular characteristic, at 12 para.4. Keller notes that the persistent homology representation is intended to alleviate the significant difficulty in molecular design related to the possibility of multiple confirmations of compounds that have the same molecular formula but important structural differences (i.e. two molecules can share local descriptors, but have distinct global descriptors). At 2 para.3.
To solve a similar issue, Krishnapriyan discloses a machine learning model for metal-organic frameworks (MOFs) that automatically generates feature descriptors based on persistent homology representations and the elemental composition of the molecule. At 2 paras.4-5. Krishnapriyan notes that to comprehensively understand MOFs, it is necessary to recognize geometric and chemical features responsible for their performance in particular applications. At 1 para.3. While input features can be related to a MOF’s performance in a particular application, standard structural descriptors are not able to capture some relevant information, such as the pressure during absorption or local strong absorption sites. At 2 para.2. To overcome these challenges, Krishnapriyan discloses using a topological descriptor called persistent homology and the elemental composition of a molecule as input into a machine learning model. At 2 para.4. The elemental compositions are embedded and the persistent homology representations are translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework to form a generative model of the molecule. Keller discloses a persistent homology representation that captures a set of points in Euclidean space and aims to alleviate major difficulty in molecular design regarding multiple confirmations of compounds that have the same molecular formula but important structural differences. Krishnapriyan discloses a solution to a similar issue in the context of MOFs where persistence images (i.e. global representation) and elemental compositions (i.e. local representation) are used as inputs for machine learning algorithms. One of ordinary skill in the art could apply Krishnapriyan’s technique of inputting global and local representations to the method of Oono by additionally inputting the persistent homology representation of Keller. This would predictably result in an improved method of molecular design because, similar to the results in Krishnapriyan, the incorporation of Keller’s persistent homology representation as input into Oono’s generative framework will remedy the issue of molecules consisting of the same chemical formula but different structures. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 23, Keller discloses a multi-parameter persistent homology representation that captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, and some other non-shape molecular characteristic, at 12 para.4 (the computer system of claim 22, wherein: the input representations further comprise an input representation of a characteristic of the 3D domain). Keller teaches that the first parameter of the representation captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, while the second parameter may capture partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4. One of ordinary skill in the art would know to include Keller’s multi-parameter persistent homology representation as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 22 above). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1. A person having ordinary skill in the art would know to translate the persistent homology representations of Keller into persistence images because the resulting vectorized persistent homology representation would be suitable as an additional input for the generative framework of Oono. This modification would yield predicable results, in that the persistent homology representations of Keller would be vectorized into persistence images, which is proper format for machine learning input. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller in the form of persistence images as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
The system resulting from the combination of Oono, Keller, and Krishnapriyan would input chemical composition fingerprints and multi-parameter persistence images capturing shape and non-shape molecular information into a generative framework. The first parameter of the persistence image captures a set of points in Euclidean space representing atom centers in a molecule (the 3D coordinates input representation of the 3D domain is represented as a first parameter of a multi-parameter persistence image), while the second parameter may capture partial charge, atomic mass, hydrogen donor/acceptor status, or some other non-shape molecular characteristic (the input representation of the characteristic of the 3D domain is represented as a second parameter of the multi-parameter persistence image).
Regarding claim 24, Oono discloses a computer system, para. [0041], comprising a memory coupled to a bus for communicating information to a processor, which executes instructions, para. [0145], that generate chemical compound models having desired characteristics based on the input representations. Abstract; para. [0098] (a computer system comprising a memory and a processor communicatively coupled to the memory, wherein the processor is operable to form a generative model of a three-dimensional (3D) domain by performing operations comprising). Oono teaches that the generative framework encodes input representations into latent representations. Para. [0041] (encoding, using a generative neural network (GNN), input representations to generate latent code). Oono discloses that the latent representations are decoded in the generative framework to reconstructions of the input representations. Id (decoding, using the GNN, the latent code to generate a reconstructed version of the input representations). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the input representations). Oono discloses the input representations comprise chemical compound fingerprints, para. [0008], and defines fingerprints as a string of values of molecular descriptors that contain the information of a compound's chemical structure, para. [0088] (wherein the input representations comprise: a string input representation of the 3D domain).
Oono fails to teach the input representations comprising a 3D coordinates input representation of the 3D domain and an input representation of a characteristic of the 3D domain. However, Keller discloses a multi-parameter persistent homology representation that captures a set of points in Euclidean space representing atom centers in a molecule, at 5 para.4, and partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4. Keller notes that the persistent homology representation is intended to alleviate the significant difficulty in molecular design related to the possibility of multiple confirmations of compounds that have the same molecular formula but important structural differences (i.e. two molecules can share local descriptors, but have distinct global descriptors). At 2 para.3.
To solve a similar issue, Krishnapriyan discloses a machine learning model for metal-organic frameworks (MOFs) that automatically generates feature descriptors based on persistent homology representations and the elemental composition of the molecule. At 2 paras.4-5. Krishnapriyan notes that to comprehensively understand MOFs, it is necessary to recognize geometric and chemical features responsible for their performance in particular applications. At 1 para.3. While input features can be related to a MOF’s performance in a particular application, standard structural descriptors are not able to capture some relevant information, such as the pressure during absorption or local strong absorption sites. At 2 para.2. To overcome these challenges, Krishnapriyan discloses using a topological descriptor called persistent homology and the elemental composition of a molecule as input into a machine learning model. At 2 para.4. The elemental compositions are embedded and the persistent homology representations are translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework to form a generative model of the molecule. Keller discloses a persistent homology representation that captures a set of points in Euclidean space and some non-shape molecular characteristic with the goal of alleviating major difficulty in molecular design regarding multiple confirmations of compounds that have the same molecular formula but important structural differences. Krishnapriyan discloses a solution to a similar issue in the context of MOFs where persistence images (i.e. global representation) and elemental compositions (i.e. local representation) are used as inputs for machine learning algorithms. One of ordinary skill in the art could apply Krishnapriyan’s technique of inputting global and local representations to the method of Oono by additionally inputting the persistent homology representation of Keller. This would predictably result in an improved method of molecular design because, similar to the results in Krishnapriyan, the incorporation of Keller’s persistent homology representation as input into Oono’s generative framework will remedy the issue of molecules consisting of the same chemical formula but different structures. Therefore, it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to improve the method of Oono by additionally inputting the persistent homology representations of Keller as taught by Krishnapriyan. Use of known technique to improve similar devices (methods, or products) in the same way is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, C).
Regarding claim 25, Oono discloses that the object being analyzed and modeled is a molecule or chemical compound. Para. [0090] (the computer system of claim 24, wherein: the 3D domain comprises a molecule). Keller discloses that the non-shape characteristic captured in the persistent homology representation can be partial charge, at 9 para.1; Figure 3 caption, atomic mass, at 13 para.1; Figure 5 caption, hydrogen donor/acceptor status, or some other non-shape molecular characteristic, at 12 para.4 (the characteristic of the molecule is selected from the group consisting of an atomic charge and an atomic weight).
Claims 6, 13, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Oono, Keller, and Krishnapriyan as applied to claims 1-5, 7-12, 14-19, and 21-25 above, and further in view of Youpeng (Artificial Neural Networks and Machine Learning – ICANN 2021 (7 September 2021).
Regarding claims 6, 13, and 20, the combination of Oono, Keller, and Krishnapriyan discloses the limitations of claims 1, 8, and 19 (see 103 rejections above). Oono discloses that the generative framework can be trained to form a generative model, para. [0040], by encoding the input representations into latent representations and decoding the latent representations into reconstructions of the input representations, para. [0041] (wherein using the GNN to form the generative model of the 3D domain comprises: encoding, using the GNN, the … input representation to generate latent code; decoding, using the GNN, the latent code to generate a reconstructed version of the … input representation). Oono further discloses that the generative framework calculates a reconstruction loss function based on the input representations and the reconstructed input representations. Para. [0046]; Fig.2A (generating, using the GNN, a reconstruction loss based at least in part on the reconstructed version of the … input representations). Keller discloses a persistent homology representation that one of ordinary skill in the art would know to include as an additional input into Oono’s base method of inputting molecular fingerprints into a generative framework (see 103 rejection of claim 1 above). Krishnapriyan discloses that persistent homology representations must be translated into vectors suitable as input for machine learning via persistence images. At 3 para.1.
Oono does not disclose a global-shape input representation and thus does not explicitly teach encoding, decoding, and generating a reconstruction loss for global-shape input representations. However, a person having ordinary skill in the art of generative neural networks would understand that when incorporating the persistence image input into Oono’s framework, the additional input would be subjected to the encoding/decoding/reconstruction loss pathway disclosed by Oono. Youpeng demonstrates this by proposing a variational autoencoder for multi-view representation. At abstract. Real-world data are typically described using multiple types of descriptors that are considered as multiple views, but the heterogeneity gap between views can hinder machine learning models from comprehensively utilizing multimodal data. At 391 para.1. To remedy this, Youpeng encodes multi-view input representations into latent code and fuses the latent code before decoding it into reconstructions of the input views. At 395 para.3; Fig.1. Youpeng generates a reconstruction loss for each input view in order to train the variational autoencoder. At 396 para.2.
Oono discloses a base method where molecular fingerprints are input into a generative framework where they are encoded into latent code, decoded into a reconstruction of the input, and a reconstruction loss is determined. When combined with the teachings of Keller and Krishnapriyan, the base method involves inputting both molecular fingerprints and persistence images into the generative framework. Youpeng demonstrates a generative framework for multi-view representations where each input view is encoded then decoded into a reconstruction of the view and a reconstruction loss is calculated for each view. A person having ordinary skill in the art would recognize that the incorporation of the persistence image input should be done in a manner similar to Youpeng’s technique for multi-view representations. One of ordinary skill in the art would reasonably predict that using Youpeng’s technique would yield an improved method because incorporating persistence images remedies the issue of molecules consisting of the same chemical formula but different structures, and Youpeng’s technique reduces the heterogeneity gap to allow comprehensive utilization of multi-view data by machine learning models. Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007) (see MPEP § 2143, D).
Response to Arguments
Applicant first argues that the combination of references fails to satisfy the claimed global-shape input representations because a descriptor that captures topological information about a molecule is not necessarily a global-shape input representation. In the instant application, claim 1 recites applying global-shape input representations of a 3D domain to a generative neural network to model the 3D domain. The instant specification limits “global-shape input representation” to input that provides information that is valid for the entire domain. See paras. [0054] & [0056]. Nicolas teaches that persistent homology offers a means to capture the global geometric and topological structure of complex data. § Persistent Homology. Additionally, Krishnapriyan discloses using persistence images for the persistent homology descriptor, which is the exact global-shape input representation recited in claim 3. Therefore, Applicant’s argument is not persuasive because the persistence image input representation of the combination of teachings constitutes a global-shape input representation.
Applicant similarly argues that the molecular fingerprint is not equivalent to the claimed local point-level input representation. The instant claims and specification do not provide a limitation on what constitutes a local point-level input representation, and the term does not appear to have a well-established definition in the art beyond a representation that encodes specific features, geometry, or topological information for individual points. Xie teaches that molecular fingerprints describe the local aspect of chemical structures. At 3 col.1 para.1. Therefore, Applicant’s argument is not persuasive because the molecular fingerprint input representation of the combination of teachings constitutes a local point-level input representation.
Applicant also argues that the combination fails to teach encoding and decoding the claimed set of input representations and generating a reconstruction loss for each input. Applicant asserts that the combination teaches applying persistence image input and molecular fingerprint input to the generative framework, where only the molecular fingerprint input is encoded and decoded into a reconstructed version of the molecular fingerprints. However, a person having ordinary skill in the art would understand that the persistence image input must similarly be encoded and decoded into a reconstructed version of the persistence image to avoid the input being treated as irrelevant side information. Simply adding a persistence image input and only reconstructing the fingerprint input would allow the topological features to be effectively discarded. A person having ordinary skill in the art would understand that in combining the teachings of the references, the additional persistence image input would be encoded and decoded into a reconstructed version of the additional input. One of ordinary skill in the art would understand that the reconstructed versions of the inputs would be used to compute the reconstruction loss, and the loss would not be computed based on the reconstructed fingerprints alone. Therefore, Applicant’s argument is not persuasive because a person having ordinary skill in the art would understand that the global-shape input representation is encoded, decoded, and a reconstruction loss computed in a similar manner to the fingerprints.
Applicant finally argues that the combination fails to teach the claimed reconstruction architecture of claims 22 and 24 because a persistent homology parameter that captures atom centers is not automatically the claimed 3D coordinates input representation, and reconstruction of the fingerprints is not automatically reconstruction of the full set of input representations. The instant specification recites that the “3D coordinates representation of the 3D domain is represented as a first parameter of a multi-parameter persistence image; and the input representation of the characteristic of the 3D domain is represented as a second parameter of the multi-parameter persistence image.” See para. [0019]. Keller discloses a multi-parameter persistent homology representation that captures a set of points in Euclidean space representing atom centers in a molecule, along with partial charge, atomic mass, hydrogen donor/acceptor status, or some other non-shape molecular characteristic. At 5 para.4; 9 para.1; Figure 3 caption; 13 para.1; Figure 5 caption; 12 para.4. Krishnapriyan discloses that persistent homology representations should be translated into persistence images that are suitable as input for machine learning algorithms to generate feature descriptors. At 2 para.11- 3 para.2. Thus, in combining the references, the multi-parameter persistent homology representation that captures a set of points in Euclidean space representing atom centers in a molecule is equivalent to the claimed 3D coordinates input representation. Additionally, as discussed above, a person having ordinary skill in the art would understand that both the string input representation and the 3D coordinates input representation would be encoded and decoded into reconstructed versions of the inputs, and a reconstruction loss would be computed for both inputs/reconstructions. Therefore, Applicant’s argument is not persuasive because the multi-parameter persistent homology representation is equivalent to the claimed 3D coordinates input representation, and a person having ordinary skill in the art would understand that the 3D coordinates input representation is encoded, decoded, and a reconstruction loss computed in a similar manner to the string input representation.
Conclusion
No claims are allowed.
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Emily A Darrigrand whose telephone number is (571) 272-1098. The examiner can normally be reached Monday-Thursday 7:00AM-4:00PM.
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, Larry Riggs, can be reached at (571) 270-3062. 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.
/E.A.D./Examiner, Art Unit 1686
/LARRY D RIGGS II/Supervisory Patent Examiner, Art Unit 1686
Prosecution Insights