DETAILED ACTION
The applicant’s response, from 17 July 2026, has been fully considered. Amendments to the claims, from 17 July 2026, were received and entered. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
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 .
Claim Status
Claims 1-20 are currently pending and under examination herein.
Claims 1-20 are rejected.
Priority
The instant application claims priority as a continuation of PCT/CN2021/131652 filed 19 November 2021 and foreign priority to CN202011379315.3 filed 30 November 2020. Receipt is acknowledged of certified copies of the foreign priority document required by 37 CFR 1.55. In this action, claims 1-20 are examined as though they had an effective filing date of 30 November 2020. In future actions, the effective filing date of one or more claims may change, due to amendments to the claims, or further analysis of the disclosure(s) of the priority application(s).
Information Disclosure Statement
The information disclosure statements (IDSs) submitted on 09/29/2022 and 03/04/2024 are in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Drawings
The figures submitted on 29 September 2022 are accepted.
Claim Rejections - 35 USC § 112
The previously issued 35 USC 112(b) rejection is withdrawn in response to the amended claims. Particularly, the terms “case omics feature” and “image omics feature” have been clarified in amended claims 7 and 14.
Claim Rejections - 35 USC § 103
Arguments associated with the previously issued 35 USC 103 rejection are considered persuasive in regard to the amended independent claims (see response to arguments below the rejection). The previously issued rejection is therefore reiterated and modified as has been necessitated by amendment.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-5, 7-12, and 14-19 are rejected under 35 U.S.C. 103 as being unpatentable over Xie et al. (2019, Research Square: 1-25, cited in previous action), in view of Parisot et al. (2018, Medical Image Analysis, Vol. 48: 117-130, cited in previous action), and in further view of Choi et al. (2017, KDD '17: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining: 787-795, cited in previous action) and Saris et al. (2009, BMC Genomics, Vol. 10: 1-16). Italicized text from reference art. Underlined text correspond to amendment.
Applicable claims include:
Claim 1. A method for processing clinical omics data based on a graph neural network, the method comprising: i. acquiring, by a device comprising a memory storing instructions and a processor in communication with the memory, first omics data of a target object; ii. extracting, by the device, at least two first omics features from the first omics data; iii. determining, by the device, a first correlation between different omics features of the at least two first omics features by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing on the correlation matrix by setting a threshold to obtain an edge matrix; iv. constructing, by the device based on the at least two first omics features and the first correlation, a first graph structure corresponding to the first omics data, the first graph structure comprising at least two nodes and at least one connecting edge, each node representing one of the first omics features in the first omics data, the at least one connecting edge connecting the at least two nodes and representing a first correlation corresponding to two connected nodes; v. obtaining, by the device, a node feature of each node in the first graph structure through a first graph neural network based on the first graph structure, the node feature having at least one dimension; and vi. performing, by the device, medical analysis on the target object based on the node feature of each node to obtain a medical analysis result corresponding to each dimension in the at least one dimension, wherein: the medical analysis comprises performing disease diagnosis, disease typing, and survival prediction on the target object, and the medical analysis result comprises a probability of the target object suffering from a disease corresponding to each dimension, a probability that the disease of the target object corresponding to each dimension is a certain disease category, and a survival probability of the target object corresponding to each dimension.
Claim 2. The method according to claim 1, wherein the constructing, based on the at least two first omics features and the first correlation, the first graph structure corresponding to the first omics data comprises: for any two of the at least two first omics features, in response to a first correlation between the two first omics features being greater than or equal to a set value, establishing, a connecting edge between two nodes corresponding to the two first omics features to construct the first graph structure.
Claim 3. The method according to claim 1, wherein: the method further comprises: i. extracting, for each node in the first graph structure, a first feature of the first omics features, the first feature being a feature of each node in the first graph structure that comprises only a single first omics feature; and the obtaining the node feature of each node in the first graph structure through the first graph neural network based on the first graph structure comprises: ii. obtaining, for each node in the first graph structure, a second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having a connecting edge relationship with the node, each hierarchy corresponding to a feature extraction layer of the first graph neural network, and iii. fusing, for each node, a first feature corresponding to the node and each second feature to obtain a node feature of the node.
Claim 4. The method according to claim 3, wherein the obtaining, for each node in the first graph structure, the second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having the connecting edge relationship with the node comprises: i. acquiring an initial feature of each node of the first graph structure; ii. determining, for each node, a weight of each associated feature through the first graph neural network based on each associated feature of the node, wherein each associated feature of the node comprises an initial feature of the node, and an initial feature of each target node having a connecting edge relationship with the node; and iii. performing, for each node, weighted fusion on each associated feature of the node through the first graph neural network based on the weight of each associated feature of the node to obtain a second feature of a hierarchy of the node, wherein: iv. in response to a node corresponding to second features of at least two hierarchies, a second feature of any hierarchy other than a first hierarchy is obtained based on a second feature of a previous hierarchy of the hierarchy.
Claim 5. The method according to claim 4, wherein the acquiring the initial feature of each node of the first graph structure comprises: i. using, in response to determining a second feature of a first hierarchy of each node, the first feature corresponding to each node as the initial feature of each node; and ii using, in response to determining a second feature of any hierarchy other than the first hierarchy, the second feature of the previous hierarchy of the hierarchies the initial feature of each node.
Claim 7. The method according to claim 1, wherein the acquiring the first omics data of the target object comprises: i. acquiring initial omics data of the target object, the initial omics data comprising at least two initial omics features; ii. acquiring an associated omics feature of the initial omics data, the associated omics feature and the initial omics data belonging to the same target object, and the associated omics feature comprising at least one of a case omics feature or an image omics feature, wherein the case omics feature comprises one or more omics features from samples for a subject's disease diagnosis, disease typing, or survival conditions, and the image omics feature comprises one or more omics features from samples for a subject's disease images; and iii. fusing each initial omics feature and the associated omics feature, respectively, to obtain a fusion omics feature corresponding to each initial omics feature, and using the fusion omics feature as a first omics feature.
Claim 8. An apparatus for processing clinical omics data based on a graph neural network, the apparatus comprising: a memory storing instructions; and a processor in communication with the memory, wherein, when the processor executes the instructions, the processor is configured to cause the apparatus to perform: i. acquiring first omics data of a target object, ii. extracting at least two first omics features from the first omics data, iii. determining a first correlation between different omics features of the at least two first omics features by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing on the correlation matrix by setting a threshold to obtain an edge matrix, iv constructing, based on the at least two first omics features and the first correlation, a first graph structure corresponding to the first omics data, the first graph structure comprising at least two nodes and at least one connecting edge, each node representing one of the first omics features in the first omics data, the at least one connecting edge connecting the at least two nodes and representing a first correlation corresponding to two connected nodes, v. obtaining a node feature of each node in the first graph structure through a first graph neural network based on the first graph structure, the node feature having at least one dimension, and vi. performing medical analysis on the target object based on the node feature of each node to obtain a medical analysis result corresponding to each dimension in the at least one dimension, wherein: the medical analysis comprises performing disease diagnosis, disease typing, and survival prediction on the target object, and the medical analysis result comprises a probability of the target object suffering from a disease corresponding to each dimension, a probability that the disease of the target object corresponding to each dimension is a certain disease category, and a survival probability of the target object corresponding to each dimension.
Claim 9. The apparatus according to claim 8, wherein, when the processor is configured to cause the apparatus to perform constructing, based on the at least two first omics features and the first correlation, the first graph structure corresponding to the first omics data, the processor is configured to cause the apparatus to perform: for any two of the at least two first omics features, in response to a first correlation between the two first omics features being greater than or equal to a set value, establishing, a connecting edge between two nodes corresponding to the two first omics features to construct the first graph structure.
Claim 10. The apparatus according to claim 8, wherein: when the processor executes the instructions, the processor is configured to further cause the apparatus to perform: i. extracting, for each node in the first graph structure, a first feature of the first omics features, the first feature being a feature of each node in the first graph structure that comprises only a single first omics feature; and when the processor is configured to cause the apparatus to perform obtaining the node feature of each node in the first graph structure through the first graph neural network based on the first graph structure, the processor is configured to cause the apparatus to perform: ii. obtaining, for each node in the first graph structure, a second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having a connecting edge relationship with the node, each hierarchy corresponding to a feature extraction layer of the first graph neural network, and iii. fusing, for each node, a first feature corresponding to the node and each second feature to obtain a node feature of the node.
Claim 11. The apparatus according to claim 10, wherein, when the processor is configured to cause the apparatus to perform obtaining, for each node in the first graph structure, the second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having the connecting edge relationship with the node, the processor is configured to cause the apparatus to perform: i. acquiring an initial feature of each node of the first graph structure; ii. determining, for each node, a weight of each associated feature through the first graph neural network based on each associated feature of the node, wherein each associated feature of the node comprises an initial feature of the node, and an initial feature of each target node having a connecting edge relationship with the node; and iii. performing, for each node, weighted fusion on each associated feature of the node through the first graph neural network based on the weight of each associated feature of the node to obtain a second feature of a hierarchy of the node, wherein: iv. in response to a node corresponding to second features of at least two hierarchies, a second feature of any hierarchy other than a first hierarchy is obtained based on a second feature of a previous hierarchy of the hierarchy.
Claim 12. The apparatus according to claim 11, wherein, when the processor is configured to cause the apparatus to perform acquiring the initial feature of each node of the first graph structure, the processor is configured to cause the apparatus to perform: i. using, in response to determining a second feature of a first hierarchy of each node, the first feature corresponding to each node as the initial feature of each node; and ii. using, in response to determining a second feature of any hierarchy other than the first hierarchy, the second feature of the previous hierarchy of the hierarchies the initial feature of each node.
Claim 14. The apparatus according to claim 8, wherein, when the processor is configured to cause the apparatus to perform acquiring the first omics data of the target object, the processor is configured to cause the apparatus to perform: i. acquiring initial omics data of the target object, the initial omics data comprising at least two initial omics features; ii. acquiring an associated omics feature of the initial omics data, the associated omics feature and the initial omics data belonging to the same target object, and the associated omics feature comprising at least one of a case omics feature or an image omics feature, wherein the case omics feature comprises one or more omics features from samples for a subject's disease diagnosis, disease typing, or survival conditions, and the image omics feature comprises one or more omics features from samples for a subject's disease images; and iii. fusing each initial omics feature and the associated omics feature, respectively, to obtain a fusion omics feature corresponding to each initial omics feature, and using the fusion omics feature as a first omics feature.
Claim 15. A non-transitory computer-readable storage medium, storing computer-readable instructions, wherein, the computer-readable instructions, when executed by a processor, are configured to cause the processor to perform: i. acquiring first omics data of a target object, ii. extracting at least two first omics features from the first omics data, iii. determining a first correlation between different omics features of the at least two first omics features by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing on the correlation matrix by setting a threshold to obtain an edge matrix, iv. constructing, based on the at least two first omics features and the first correlation, a first graph structure corresponding to the first omics data, the first graph structure comprising at least two nodes and at least one connecting edge, each node representing one of the first omics features in the first omics data, the at least one connecting edge connecting the at least two nodes and representing a first correlation corresponding to two connected nodes, v. obtaining a node feature of each node in the first graph structure through a first graph neural network based on the first graph structure, the node feature having at least one dimension, and vi performing medical analysis on the target object based on the node feature of each node to obtain a medical analysis result corresponding to each dimension in the at least one dimension, wherein: the medical analysis comprises performing disease diagnosis, disease typing, and survival prediction on the target object, and the medical analysis result comprises a probability of the target object suffering from a disease corresponding to each dimension, a probability that the disease of the target object corresponding to each dimension is a certain disease category, and a survival probability of the target object corresponding to each dimension.
Claim 16. The non-transitory computer-readable storage medium according to claim 15, wherein, when the computer-readable instructions are configured to cause the processor to perform constructing, based on the at least two first omics features and the first correlation, the first graph structure corresponding to the first omics data, the computer-readable instructions are configured to cause the processor to perform: for any two of the at least two first omics features, in response to a first correlation between the two first omics features being greater than or equal to a set value, establishing, a connecting edge between two nodes corresponding to the two first omics features to construct the first graph structure.
Claim 17. The non-transitory computer-readable storage medium according to claim 15, wherein: when the computer-readable instructions are executed by the processor, the computer-readable instructions are configured to further cause the processor to perform: i. extracting, for each node in the first graph structure, a first feature of the first omics features, the first feature being a feature of each node in the first graph structure that comprises only a single first omics feature; and when the computer-readable instructions are configured to cause the processor to perform obtaining the node feature of each node in the first graph structure through the first graph neural network based on the first graph structure, the computer-readable instructions are configured to cause the processor to perform: ii. obtaining, for each node in the first graph structure, a second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having a connecting edge relationship with the node, each hierarchy corresponding to a feature extraction layer of the first graph neural network, and iii. fusing, for each node, a first feature corresponding to the node and each second feature to obtain a node feature of the node.
Claim 18. The non-transitory computer-readable storage medium according to claim 17, wherein, when the computer-readable instructions are configured to cause the processor to perform obtaining, for each node in the first graph structure, the second feature of at least one hierarchy of the node through the first graph neural network based on the node in the first graph structure and each target node having the connecting edge relationship with the node, the computer-readable instructions are configured to cause the processor to perform: i. acquiring an initial feature of each node of the first graph structure; ii. determining, for each node, a weight of each associated feature through the first graph neural network based on each associated feature of the node, wherein each associated feature of the node comprises an initial feature of the node, and an initial feature of each target node having a connecting edge relationship with the node; and iii. performing, for each node, weighted fusion on each associated feature of the node through the first graph neural network based on the weight of each associated feature of the node to obtain a second feature of a hierarchy of the node, wherein: iv. in response to a node corresponding to second features of at least two hierarchies, a second feature of any hierarchy other than a first hierarchy is obtained based on a second feature of a previous hierarchy of the hierarchy.
Claim 19. The non-transitory computer-readable storage medium according to claim 18, wherein, when the computer-readable instructions are configured to cause the processor to perform acquiring the initial feature of each node of the first graph structure, the computer-readable instructions are configured to cause the processor to perform: i. using, in response to determining a second feature of a first hierarchy of each node, the first feature corresponding to each node as the initial feature of each node; and ii. using, in response to determining a second feature of any hierarchy other than the first hierarchy, the second feature of the previous hierarchy of the hierarchies the initial feature of each node.
Regarding Claims 1, 8, and 15, Xie et al. teach (Claim 1.i) acquiring, first omics data of a target object (Page 3, Paragraph 1: In this study, we propose a graph neural network based approach, GNNDR, for predicting drug responses of cell lines and xenografts based on their genomic features and protein-protein interaction (PPI) information). Genomic features and PPI information represent omics data. Xie et al. also teach (Claim 1.ii) extracting at least two first omics features from the first omics data (Page 10, Paragraph 2: We preprocess the raw data to generate four different molecular profiles (i.e., features) for each Patient Derived Xenografts: Gene Expression, Copy Number, Detected Somatic Mutation, Copy Number Alteration). Xie et al. also teach (Claim 1.v) obtaining a node feature of each node through a graph neural network, the node feature having at least one dimension (Page 8, Paragraph 1: SAGpool, which defines a GNN module to learn attention scores for all the nodes using both node features and topological features). Figure 3 (Page 9) demonstrates how their graph neural network creates nodes (output graph) from an input graph. Claim 8 recites the limitations of Claim 1 directed to an apparatus and Claim 15 recites the limitations of Claim 1 directed to a NTCRM.
Regarding Claim 3, 10, and 17, Xie et al. teach (Claim 3.i) extracting, for each node, a first feature of the first omics features, the first feature of each node in the graph structure that comprises only a single omics feature (Page 5, Paragraph 2: First, each node collects learned embeddings in the previous step of nodes in its immediate neighborhood and applies aggregate function to produce a single vector). Xie et al. also teach (Claim 3.ii) obtaining, for each node, a second feature of a hierarchy of the node through the first graph neural network, each hierarchy corresponding to a feature extraction layer of the graph neural network (Page 6, Paragraph 3: Ying et al. proposed an end-to-end differentiable graph pooling strategy, Diffpool, that learns hierarchical representations of graphs by defining a procedure to gradually coarsen input graphs, which is illustrated in Figure 2. Diffpool defines a separate GNN module to learn node assignment at each pooling layer). Xie et al. utilize the Diffpool procedure within their methods. Xie et al. also teach (Claim 3.iii) fusing, for each node, a feature corresponding to the node and each second feature to obtain a node feature (Page 5, Paragraph 2: The aggregated neighborhood embedding is then concatenated with the node’s current embedding and this concatenated vector is fed through a fully connected layer with nonlinear activation unit, which produces the representation that will be used in the next step). The term concatenation is synonymous with fusion. Claim 10 recites the limitations of Claim 3 directed to an apparatus and Claim 17 recites the limitations of Claim 3 directed to a NTCRM.
Regarding Claim 4, 11, and 18, Xie et al. teach (Claim 4.i) acquiring an initial feature of each node of the first graph structure (Page 5, Paragraph 2: First, each node collects learned embeddings in the previous step of nodes in its immediate neighborhood and applies aggregate function to produce a single vector). Xie et al. also teach (Claim 4.ii) determining, for each node, a weight of each associated feature through the first graph neural network based on each associated feature of the node, wherein each associated feature of the node comprises an initial feature of the node, and an initial feature of each target node having a connecting edge relationship with the node (Page 5, Paragraph 2: In general, to generate new embedding of a node given the current learned embeddings in the network, graph neural network (GNN) collects embeddings of neighboring nodes, merge them using some aggregate function and then multiply the aggregated embedding with trainable weights (i.e. the weights are generated by the GNN) and this procedure could be repeated for K hops which is usually termed the depth of the graph convolution). Xie et al. also teach (Claim 4.iii) performing, for each node, weighted fusion on each associated feature of the node through the first graph neural network based on the weight of each associated feature of the node to obtain a second feature of a hierarchy of the node (Page 5, Paragraph 2: In general, to generate new embedding of a node given the current learned embeddings in the network, GNN collects embeddings of neighboring nodes, merge (i.e., fuse) them using some aggregate function and then multiply the aggregated embedding with trainable weights (i.e., the fusion involves weights) and this procedure could be repeated for K hops which is usually termed the depth of the graph convolution). Xie et al. also teach (Claim 4.iv) in response to a node corresponding to second features of at least two hierarchies, a second feature of any hierarchy other than a first hierarchy is obtained based on a second feature of a previous hierarchy of the hierarchy (Page 6, Paragraph 3: Diffpool defines a separate GNN module to learn node assignment at each pooling layer). A new hierarchy is assigned at each layer of multiple layers indicating that there are multiple hierarchies that are utilized (see Page 8, Figure 2). Claims 11 recites the limitations of claim 4 directed to an apparatus and claim 18 recites the limitations of claim 4 directed to a NTCRM.
Regarding Claim 5, 12, and 19, Xie et al. teach (Claim 5.i) using, in response to determining a second feature of a first hierarchy of each node, the first feature corresponding to each node as the initial feature of each node (Page 5, Paragraph 3: Hierarchical Differentiable Graph Pooling - Apart from defining a GNN module that is specialized in learning node embeddings, Diffpool defines a separate GNN module to learn node assignment at each pooling layer; Page 8, Paragraph 1: SAGpool, which defines a GNN module to learn attention scores for all the nodes using both node features and topological features and the learned attention scores are used to only choose a subset of the original node sets that have the maximum attention scores during pooling). The hierarchical structure used to define the pooling will determine initial nodes based on the previous layer. Xie et al. also teach (Claim 5.ii) using, in response to determining a second feature of any hierarchy other than the first hierarchy, the second feature of the previous hierarchy of the hierarchies the initial feature of each node (Page 5, Paragraph 3: Ying et al. proposed an end-to-end differentiable graph pooling strategy, Diffpool, that learns hierarchical representations of graphs by defining a procedure to gradually coarsen input graphs, which is illustrated in Figure 2). Because Diffpool learns hierarchical representations as part of the GNN, it will be separate for each of the implemented iterations (see Page 8, Figure 2 for an illustration of its implementation). Claim 12 recites the limitations of Claim 5 directed to an apparatus and Claim 19 recites the limitations of Claim 5 directed to a NTCRM.
Regarding Claims 7 and 14, Xie et al. teach (Claim 7.iii) fusing each initial omics feature and the associated omics feature, respectively, to obtain a fusion omics feature corresponding to each initial omics feature, and using the fusion omics feature as a first omics feature (Page 5, Paragraph 2: The aggregated neighborhood embedding is then concatenated with the node’s current embedding and this concatenated vector is fed through a fully connected layer with nonlinear activation unit, which produces the representation that will be used in the next step). The term concatenation is synonymous with fusion. Claim 14 recites the limitations of Claim 7 directed to an apparatus.
Xie et al. do not explicitly teach determining a correlation between omics features with a Weighted Gene Co-Expression Network Analysis (Claims 1.iii, 8.iii, 15.iii). Xie et al. also do not explicitly teach constructing, based on the omics features and the correlation, a graph structure, comprising at least two nodes and at least one connecting edge, each node representing one omics features, the edge connecting the two nodes and representing a correlation (Claims 1.iv, 8.iv, 15.iv). Xie et al. also do not explicitly teach the medical analysis comprises performing disease diagnosis, disease typing, and survival prediction, and the medical analysis result comprises a probability of the target object suffering from a disease corresponding to each dimension, a probability that the disease of the target object corresponding to each dimension is a certain disease category, and a survival probability of the target object corresponding to each dimension (Claims 1.vi, 8.vi. 15.vi). Xie et al. also do not explicitly teach the constructing comprises for any two of the omics features, in response to a correlation between the omics features being greater than or equal to a set value, establishing, a connecting edge between the two nodes (Claims 2, 9, and 16). Xie et al. also do not explicitly teach acquiring initial omics data of the target object, the initial omics data comprising at least two initial omics features (Claims 7.i, 14.i). Xie et al. also do not explicitly teach acquiring an associated omics feature of the initial omics data, the associated omics feature and the initial omics data belonging to the same target object, and the associated omics feature comprising at least one of a case omics feature or an image omics feature, wherein the case omics feature comprises one or more omics features from samples for a subject's disease diagnosis, disease typing, or survival conditions, and the image omics feature comprises one or more omics features from samples for a subject's disease images (Claims 7.ii, 14.ii).
Regarding Claims 1, 8, and 15, Parisot et al. suggest (Claim 1.iii) determining a correlation between different omics features (i.e. nodes) (Page 120, Column 1, Paragraph 3: The two main decisions required to build the population model are: 1) the definition of the feature vector x ( v ) describing each graph node/acquisition, and 2) the connectivity of the graph, i.e. its edges Eand their weights W , which models the similarity be- tween nodes/subjects/scans and their corresponding features; Page 120, Column 2, Paragraph 1: For the ADNI dataset, we use the volumes of all C = 138 seg- mented brain structures, a type of feature which has been highly effective for prediction of Alzheimer’s disease; Page 120, Column 2, Paragraph 3: graph edges - where, Sim(Sv, Sw) is a measure of similarity between subjects, increasing the edge weights between the most similar graph nodes; Page 121, Column 1, Paragraph 2: Finally, we define the similarity measure as equation 3 and one of the variables (ρ) is the correlation distance). Parisot et al. teach (Claim 1.iv) constructing, based on the omics features and the correlation, a graph structure, comprising at least two nodes and at least one connecting edge, each node representing one omics features, the edge connecting the two nodes and representing a correlation (Page 120, Column 1, Paragraph 3: We provide an illustration of the construction of the population graph in Fig. 1). Figure 1 is an overview the pipeline used for classification of population graphs using Graph Convolutional Networks. It includes omics data (Phenotypic data showing DNA) from subjects as inputs to create a graph structure with multiple nodes and edges. The edges are based on correlations (see Claim 1.iii above). Parisot et al. teach (Claim 1.vi) performing medical analysis on the target object based on the node feature of each node corresponding to each dimension and comprises performing, including the probability of each dimension, disease typing prediction on the target object (Page 123, Column 1, Paragraph 3: Results are shown in Fig. 5, reporting classification accuracy). The classification accuracy, reported as a percentage, is equivalent to a probability of their prediction being correct. They are predicting classifications of disease corresponding to Autism Spectrum Disorder and Alzheimer’s disease. Claim 8 recites the limitations of Claim 1 directed to an apparatus and Claim 15 recites the limitations of claim 1 directed to a NTCRM.
Regarding Claim 2, 9, and 16, Parisot et al. teach for any two of the omics features, in response to a correlation between the omics features being greater than or equal to a set value, establishing, a connecting edge between the two nodes (Page 121, Column 1, Paragraph 1: Constructing edge weights from quantitative measures is slightly less straightforward. In such cases, we define γ as a unit-step function with respect to a threshold θ). The threshold indicates edges are established based on comparison to a set value. Parisot et al. teach these edges are defined through correlations (See Claim 1.iii above). Claim 9 recites the limitations of Claim 2 directed to an apparatus and claim 16 recites the limitations of Claim 2 directed to a NTCRM.
Regarding Claims 7 and 14, Parisot et al. teach (Claim 7.i) acquiring initial omics data of the target object, the initial omics data comprising at least two initial omics features (Page 120, Column 1, Paragraph 3: We provide an illustration of the construction of the population graph in Fig. 1). Figure 1 is an overview the pipeline used for classification of population graphs using Graph Convolutional Networks. It includes omics data (Phenotypic data showing DNA) from subjects as inputs to create a graph structure with multiple nodes and edges. Parisot et al. teach (Claim 7.ii) acquiring an associated omics feature of the initial omics data, the associated omics feature and the initial omics data belonging to the same target object, and the associated omics feature comprising at least one of a case omics feature or an image omics feature, wherein the case omics feature comprises omics features from samples for a subject's disease diagnosis, disease typing, or survival conditions, and the image omics feature comprises omics features from samples for a subject's disease images (Page 119, Column 2, Paragraph 3: We demonstrate the potential and versatility of the model using two large and challenging databases, namely the ABIDE and ADNI databases; Page 119, Column 2, Paragraph 4: The ABIDE database aggregates data from different international acquisition sites and openly shares neuroimaging and phenotypic data of 1112 subjects; Page 119, Column 2, Paragraph 5: To date, ADNI in its three studies has recruited over 1700 adults, aged between 55 and 90 years, from over 50 sites from the U.S. and Canada. In this work, a subset of 540 early/late MCI subjects that contained longitudinal T1 MR images and their respective anatomical segmentations was used). Claim 14 recites the limitations of Claim 1 directed to an apparatus. Claims 14 recites the limitations of claim 7 directed to an apparatus.
Parisot et al. do not explicitly teach determining a correlation between omics features with a Weighted Gene Co-Expression Network Analysis (Claims 1.iii, 8.iii, 15.iii). Parisot et al. do not explicitly teach performing medical analysis on the target object based on the node feature of each node corresponding to each dimension and comprises performing, including the probability of each dimension, disease diagnosis and survival prediction on the target object (Claim 1.vi, 8.vi, 15.vi).
Regarding Claim 1, 8, and 15, Choi et al. teach (Claim 1.vi) performing medical analysis on the target object based on the node feature of each node corresponding to each dimension and comprises performing, including the probability of each dimension, disease diagnosis (Page 790, Column 2, Paragraph 3: We conducted two sequential diagnoses prediction tasks using two different datasets) and survival (Page 790, Column 2, Paragraph 4: We also conducted a heart failure prediction task) prediction on the target object. The predictions involve probabilities as indicated by the accuracy estimates (Page 788, Column 1, Paragraph 1: We compare predictive performance (i.e. accuracy, data needs, interpretability)). Claim 8 recites the limitations of Claim 1 directed to an apparatus and Claim 15 recites the limitations of claim 1 directed to a NTCRM. Additionally, Choi et al. teach their methods can be implemented through an apparatus (computer) and non-transitory computer-readable storage medium (Page 790, Column 2, Paragraph 2: The source code of GRAM is publicly available at https://github.com/mp2893/gram). A computer would inherently contain non-transitory computer-readable storage medium, memory, and at least one processor.
Choi et al. do not explicitly teach determining a correlation between omics features with a Weighted Gene Co-Expression Network Analysis (Claims 1.iii, 8.iii, 15.iii).
Regarding Claims 1, 8, and 15, Saris et al. teach (Claim 1.iii) determining a first correlation between different omics features of the at least two first omics features by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing on the correlation matrix by setting a threshold to obtain an edge matrix (Page 12, Column 1, Paragraph 4: We constructed weighted gene co-expression networks. The determination of weighted co-expression starts by calculating a correlation matrix. By raising the absolute value of the Pearson correlation to a power β ≥ 1 (soft thresholding), the weighted gene coexpression network construction emphasizes large correlations at the expense of low correlations. We used the scale free topology criterion to choose the soft threshold β = 6). The soft thresholding approach was considered to be equivalent to binary processing in light of Paragraph 0059 of the published specification. Saris et al. further suggests performing binary processing on the correlation matrix by setting a threshold (Page 12, Column 1, Paragraph 5: use a hard threshold to dichotomize the correlation matrix).
It would have been obvious to one of ordinary skill in the art at the time of the effective filing date to modify Xie et al. with Parisot et al. and Choi et al. because each provides a motivation for incorporating some element that was not originally part of Xie et al. Parisot et al. teaches specific implementations of an applications of the methods and includes novel incorporation of image data in graph based networks (Page 119, Column 1, Paragraph 2: Our evaluation on two different datasets aims to demonstrate the framework’s versatility as it facilitates the incorporation of domain-specific knowledge in two different clinical settings, while at the same time showing consistent improvement with respect to baselines for both challenging problems; A novel formulation of subject classification as a graph labelling problem, integrating imaging and non imaging data). Choi et al. teach specific advanced methods and novel implementations of graph based networks for medical predictions (Page 788, Column 1, Paragraph 1: We demonstrate that GRAM is up to 10% more accurate than the basic RNN for predicting diseases less observed in the training data; Page 794, Column 2, Paragraph 1: no one has designed attention model based on knowledge ontology, which is the focus of this work). Therefore, it would have been obvious to someone of ordinary skill in the art at the time of the effective filling date to combine the methods from the references indicated above. Furthermore, one of ordinary skill in the art would predict that the method taught by Parisot et al. and Choi et al. could be readily added to the methods of Xie et al. with a reasonable expectation of success because they were shown to function in the same technical field and objective – use graph based networks to make predictions from medical data.
Additionally, It would have been obvious to one of ordinary skill in the art at the time of the effective filing date to combine Saris et al. with Xie et al., Parisot et al., and Choi et al. Saris et al. teach their analysis techniques were more effective than other modeling approaches utilizing omics data (Page 11, Column 1, Paragraph 4: We compare the findings of WGCNA with those of a standard analysis. As can be seen from our functional enrichment analysis results of different gene lists, keeping track of module membership leads to statistically more significant enrichment results. WGCNA's systems biologic, module-centric approach hones in on disease related pathways and their key drives). Saris et al. also suggest utilizing their analysis methods to diagnosis disease (Page 11, Column 2, Paragraph 2: Weighted gene co-expression network analysis applied to blood gene expression data from ALS patients is combined with Ingenuity Pathway Analysis to implicate disease pathways, molecular mechanisms, and connections to other disorders. Our results suggest that development of an ALS biomarker based on gene expression in peripheral blood may be possible), which are related to the objectives of the cited art and the instant application. Furthermore, one of ordinary skill in the art would predict that the methods could be readily combined with a reasonable expectation of success because both are within the same technical field – the methods are based on graph theory/modeling which are consistent with graph based approaches and modeling of the cited art and the instant application.
Claims 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Xie et al., in view of Parisot et al., and in further view of Choi et al. and Saris et al., as applied to claims 1-5, 7-12, and 14-19 above, and in further view of Barbiero et al. (2020, arXiv: 1-21, cited in previous action). Italicized text from reference art. Underlined text correspond to amendment.
Applicable claims include:
Claims 1-5, 7-12, and 14-19 presented above.
Claim 6. The method according to claim 1, wherein: the method further comprises: i. acquiring at least one piece of second omics data, different pieces of omics data in the first omics data and the at least one piece of second omics data belonging to different omics, and the at least one piece of second omics data and the first omics data belonging to the same target object, and ii. extracting a data feature corresponding to each second omics data; and iii. the performing medical analysis on the target object based on the node feature of each node to obtain the medical analysis result comprises: determining the medical analysis result on the target object based on the node feature of each node and the data feature corresponding to each second omics data.
Claim 13. The apparatus according to claim 8, wherein: when the processor executes the instructions, the processor is configured to further cause the apparatus to perform: i. acquiring at least one piece of second omics data, different pieces of omics data in the first omics data and the at least one piece of second omics data belonging to different omics, and the at least one piece of second omics data and the first omics data belonging to the same target object, and ii. extracting a data feature corresponding to each second omics data; and iii. when the processor is configured to cause the apparatus to perform performing medical analysis on the target object based on the node feature of each node to obtain the medical analysis result, the processor is configured to cause the apparatus to perform: determining the medical analysis result on the target object based on the node feature of each node and the data feature corresponding to each second omics data.
Claim 20. The non-transitory computer-readable storage medium according to claim 15, wherein: when the computer-readable instructions are executed by the processor, the computer-readable instructions are configured to further cause the processor to perform: i. acquiring at least one piece of second omics data, different pieces of omics data in the first omics data and the at least one piece of second omics data belonging to different omics, and the at least one piece of second omics data and the first omics data belonging to the same target object, and ii. extracting a data feature corresponding to each second omics data; and iii. when the computer-readable instructions are configured to cause the processor to perform performing medical analysis on the target object based on the node feature of each node to obtain the medical analysis result, the computer-readable instructions are configured to cause the processor to perform: determining the medical analysis result on the target object based on the node feature of each node and the data feature corresponding to each second omics data.
Regarding Claims 1-5, 7-12, and 14-19, are taught by Xie et al., Parisot et al., Choi et al., and Saris et al. as applied to Claims 1-5, 7-12, and 14-19 above.
Regarding Claim 6, 13, and 20, Barbiero et al. teach (Claim 6.i) acquiring at least one piece of second omics data, different pieces of omics data in the first omics data and the at least one piece of second omics data belonging to different omics, and the at least one piece of second omics data and the first omics data belonging to the same target object (Page 6, Paragraph 1: The transcriptomic layer operates on the set of RNA transcripts produced by the genome at a particular time; Page 6, Paragraph 6: The exposome refers to the totality of exposure individuals experience from conception until death and its impact on chronic and acute diseases). Barbiero et al. utilize these multiple omics datasets for their graph networks. Barbiero et al. also teach (Claim 6.ii) extracting a data feature corresponding to each second omics data (Page 6, Paragraph 3: Specifically, we analyse to what extent the expression of genes involved in the renin-angiotensin system can be explained by genes from signaling and receptor pathways, including the chemokine, TNF, and TGF- pathways; Page 6, Paragraph 6: In this work, we consider four types of exposures: dietary habits, physical activity, therapeutic treatments, and viral infections). Features are extracted from each omics dataset. Barbiero et al. also teach (Claim 6.iii) performing medical analysis on the target object based on the node feature of each node to obtain the medical analysis result on the target object based on the node feature of each node and the data feature corresponding to each second omics data (Page 4, Paragraph 3: Each subsystem can be represented as a different node or set of nodes in a GNN, while inter-process signals can be reframed as message passing operations supporting multiscale ripple effects. Page 7, Paragraph 1: The resulting GNN model will combine a simple and modular design with a versatile structure accommodating for complex multiscale systems where clinical endpoints (i.e., medical analysis) can be easily monitored and forecast in real time). Claim 13 recites the limitations of Claim 6 directed to an apparatus. Claim 20 recites the limitations of Claim 6 directed to a NTCRM.
It would have been obvious to one of ordinary skill in the art at the time of the effective filing date to modify the combination of Xie et al., Parisot et al., Choi et al., and Saris et al. with Barbiero et al. because Barbiero et al. provide a motivation for incorporating some element that was not originally part of Xie et al., Parisot et al., Choi et al. or Saris et al. Parisot et al. teach the use and integration of multiple omics types incorporated into graphed based networks and highlight 8 beneficial features of their methods - non-linearity, interpretability, non-euclidean geometry, modularity, cross-modality, generative, multiscale, and spectral density (Page 3, Paragraph 3: Several properties of graph and generative adversarial neural networks make them suitable for medical data analysis). Therefore, it would have been obvious to someone of ordinary skill in the art at the time of the effective filling date to combine the methods from the references indicated above. Furthermore, one of ordinary skill in the art would predict that the method taught by Barbiero et al. could be readily added to the methods of Xie et al., Parisot et al., and Choi et al. with a reasonable expectation of success because they were shown to function within the same technical field and had the same objective – use graph based networks to make predictions from medical data.
Response to Arguments
Applicant asserts the previously cited references to not teach the amended limitation of independent claims, “determining, by the device, a first correlation between different omics features of the at least two first omics features by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing on the correlation matrix by setting a threshold to obtain an edge matrix” (Page 15, Paragraph 4 of the remarks). The examiner agrees the cited references do not explicitly teach the amended limitation. The previously issued 35 USC 103 Rejection has therefore been updated to address amended limitation. As indicated by the updated rejection above, Saris et al. teach the entirety the amended claim, determining correlations between multiple omics features (gene expression) by calculating a correlation matrix between the different omics features using Weighted Gene Co-Expression Network Analysis (WGCNA), and performing binary processing using a threshold. Furthermore, the methods of Saris et al. are related to the methods of the previously cited references and teach and suggest pertinent motivations why one of ordinary skill in the art would be motivated to combine them (see updated rejection above). The claims therefore stand rejected under the updated 35 USC 103 as necessitated by amendment.
Applicant also asserts “Parisot constructs a population graph in which each node represents a patient or subject, not an individual omics feature. The edges in Parisot's graph represent similarity between subjects based on phenotypic measures such as age and sex. See Parisot, page 120. Parisot does not teach or suggest calculating a correlation matrix between omics features using WGCNA or performing binary processing on such a matrix to obtain an edge matrix. Moreover, Parisot's population graph is a subject-level graph in which each node represents an entire patient or subject and edges encode inter-subject similarity derived from phenotypic measures” (Page 15, last paragraph or remarks). As indicated by the rejection above, Parisot et al. describes a graph structure with nodes and edges (see Page 120, Figure 1). Parisot et al. describes the nodes and edges as “The two main decisions required to build the population model are : 1) the definition of the feature vector x ( v ) describing each graph node/acquisition, and 2) the connectivity of the graph, i.e. its edges Eand their weights W , which models the similarity be- tween nodes/subjects/scans and their corresponding features” (Page 120, Column 1, Paragraph 3). A feature vector (i.e. nodes) set utilized is described as “For the ADNI dataset, we use the volumes of all C = 138 segmented brain structures, a type of feature which has been highly effective for prediction of Alzheimer’s disease” (Page 120, Column 2, Paragraph 1). This is consistent with case and image omics type data described by specification (Paragraph 0138 of the published specification: the case omics feature may include one or more omics feature from samples and/or examples for a subject's disease diagnosis, disease typing, and survival conditions. In some implementations, the image omics feature may include one or more omics feature from samples and/or examples for a subject's disease images (e.g., X-ray images, CT images, MRI images, and the like)). The MR scans used from ADNI dataset are taken from individuals. The similarities between nodes (i.e. edges) are calculated based on correlations, “we define the similarity measure as equation 3, where ρ is the correlation distance”. Therefore, Parisot et al. suggest at least the claim limitation - determining a first correlation between different omics features of the at least two first omics features. The application to individual based modeling is especially clear in light of the combination of Parisot et al. with Barbiero et al. (See 103 Rejection of claims 1-20), who teaches individual based omics modeling (digital twin). As this is a 35 USC 103 Rejection, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references (MPEP 2145). Therefore, this argument is deemed unpersuasive.
Double Patenting
No double patenting issues were found.
Conclusion
As indicated by the pervious office action, no 35 USC 101 rejection is issued due to additional elements that were not routine or conventional at the time of the effective filling date (2020) within the independent claims. These include the use of a graph neural network to structure omics data. The additional elements in combination with the judicial exceptions (extracting data and performing correlations and analysis) therefore comprise an inventive concept that transforms the claimed judicial exception into a patent-eligible application of the judicial exception.
No Claims are allowed.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to BLAKE H ELKINS whose telephone number is (571)272-2649. The examiner can normally be reached Monday-Thursday 8-5PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, 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.
/B.H.E./Examiner, Art Unit 1687
/Karlheinz R. Skowronek/Supervisory Patent Examiner, Art Unit 1687
Prosecution Insights