DETAILED ACTION
Notice of 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 .
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 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.
Withdrawal of Objections and Rejections
Applicant's response, filed 03/24/2026, has been fully considered.
In view of the amendment and remarks from 03/24/2026, the objection to the claims, and the rejection of the following claims are withdrawn:
claims 1-20 under 35 USC § 112(b);
claims 1-20 under 35 U.S.C. § 101 – reasons for eligibility: referring to the 101 analysis as organized in MPEP 2106, the 101 rejections are withdrawn at least in view of the analysis at Step 2A, 2nd prong, 1st consideration relating to an improvement over the previous state of the technology field integrating possible judicial exceptions into a practical application (MPEP 2106.04(d) and (d)(1)). The instant improvement is to the integrated technological contribution using computer technology and machine learning with structured data to make improved predictions (i.e. an inference at inference time) about chemical properties. The claims integrate the judicial exception into a practical application by improving a computer function in "the self-attention laver being configured to modify a scaled query-key compatibility value by adding the learnable scalar bias terms prior to softmax." In this regard, Applicant's 03/24/26 remarks at pg. 10 para. 6 further support withdrawal of the rejection.
The following rejections and/or objections are either maintained or newly applied for claims 1-7, 9-16 and 18-20. They constitute the complete set applied to the instant application. Herein, "the previous Office action" refers to the Non-Final Rejection of 01/07/2026.
Status of the Claims
Claims 8 and 17 are canceled.
Claims 1-7, 9-16 and 18-20 are pending.
Claims 1-7, 9-16 and 18-20 are rejected.
Priority
This US Application 17/806,075 (06/08/2022) claims no priority herein as reflected in the filing receipt mailed on 06/17/2022. The claims to the benefit of priority are acknowledged; and the effective filing date of claims 1-7, 9-16 and 18-20 is 06/08/2022.
Claim Rejections - 35 USC § 103
The following is a quotation of pre-AIA 35 U.S.C. 103(a) which forms the basis for all obviousness rejections set forth in this Office action:
(a) A patent may not be obtained though the invention is not identically disclosed or described as set forth in section 102, if the differences between the subject matter sought to be patented and the prior art are such that the subject matter as a whole would have been obvious at the time the invention was made to a person having ordinary skill in the art to which said subject matter 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 pre-AIA 35 U.S.C. 103(a) 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.
A. Claims 1-6, 9-15, and 18-20 are rejected under 35 U.S.C. 103(a) as being unpatentable over Mansimov ("Molecular geometry prediction using a deep generative graph neural network." Scientific reports 9(1):20381 (2019)) in view of Gasteiger ("Universal directional graph neural networks for molecules." Advances in Neural Information Processing Systems 34:6790-6802 (2021)) in view of Baingana ("Embedding Graphs under Centrality Constraints." Thesis Dissertation. University of Minnesota (2013)) - as cited on the 01/07/2026 Form PTO-892 - in view Guo ("Normalized and geometry-aware self-attention network for image captioning." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020), as cited on the attached Form PTO-892). Any newly recited portions are necessitated by claim amendment.
Claim 1 recites a computer system, comprising: a processor and memory storing instructions that, during execution, cause the processor, during a training phase, comprising steps. Claim 10 recites a computerized method comprising during a training phase, said steps. Claim 19 recites a computing system, comprising: a processor and memory storing instructions that, during execution, cause the processor, during a training phase, comprising said steps.
The prior art to Mansimov discloses a method and a system related to a deep generative graph neural network that learns an energy function by directly learning to generate energetically favorable molecular conformations (pg. 1 para. 1); wherein energy was calculated using GAMESS software (pg. 7 Table 2). The steps performed by the system of claim 1, a method of claim 10, and a system of claim 19 recite:
provide a training data set including a plurality of training data pairs, each of the training data pairs including a pre-transformation molecular graph and post-transformation energy parameter value representing an energy change in a molecular system following an energy transformation, wherein the pre-transformation molecular graph includes a plurality of normal nodes connected by edges, each normal node representing an atom in the molecular system
• Mansimov teaches a deep generative graph neural network that learns an energy function by directly learning to generate energetically favorable molecular conformations (pg. 1 para. 1); wherein each molecule is paired with a reference conformation obtained by optimizing the molecular geometry with density functional theory (pg. 5 para. 5); which was used to train the model with a batch size of 20 molecules during a learning phase (i.e. a plurality of training data pairs) (pg. 6 para. 2); wherein the model uses matrices representing the linear transformations for the nodes and edges respectively (pg. 4 para. 2) with each atom represented as a vector of node features (i.e. a plurality of normal nodes connected by edges, each node representing an atom in the molecular system) (pg. 2 para. 3); wherein the weights of the message passing function are shared across the network layers (i.e. reading on normal nodes – nodes that applies weights and functions to learn) (pg. 3 para. 4). wherein molecules can transition between conformations (i.e. energy change due to energy transformations) and end up in different local minima based on the stability of the respective conformations and environmental conditions (i.e. post-transformation energy parameter value representing an energy change in a molecular system following an energy transformation) (pg. 2 para. 4); wherein energy was calculated using GAMESS software (i.e. post- transformation energy parameter value) (pg. 7 Table 2); wherein given the initial graph of a molecule (i.e. a pre-transformation molecular graph), the task of molecular geometry prediction is the generation of a set of plausible conformations using the 3-D coordinates (i.e. inference at inference time – trained neural network making a prediction) (pg. 2 para. 3).
encode structural information in each pre-transformation molecular graph as learnable embeddings …
an edge encoding representing a type of bond between a pair of the normal nodes in each pre-transformation molecular graph …
input the training data set to a transformer-based graph neural network to thereby train the transformer-based graph neural network to perform an inference at inference time
• Mansimov teaches a deep generative graph neural network that learns an energy function by directly learning to generate energetically favorable molecular conformations (pg. 1 para. 1); wherein given the initial graph of a molecule the task of molecular geometry prediction (i.e. encode structural information in each pre-transformation molecular graph) is the generation of a set of plausible conformations using the 3-D coordinates (i.e. inference at inference time – trained neural network making a prediction) (pg. 2 para. 3); wherein the model uses matrices representing the linear transformations for the nodes and edges respectively (pg. 4 para. 2) with each atom represented as a vector of node features (i.e. a plurality of normal nodes connected by edges, each node representing an atom in the molecular system) (pg. 2 para. 3); wherein the weights of the message passing function are shared across the network layers (i.e. reading on normal nodes – nodes that applies weights and functions to learn) (pg. 3 para. 4) and shared weight embeddings during prior and posterior parameterization (i.e. learnable embeddings) (pg. 4 para. 3); wherein the inference approach samples from a trained conditional variational graph autoencoder by first sampling from the prior distribution and taking the mean vectors from the likelihood distribution (pg. 6 para. 3); using structural information the atoms, bonds, molecular mass and symmetry of molecules in each dataset (pg. 6 Fig. 1); wherein the number of rotatable bonds is identified per molecule (i.e. an edge encoding representing a type of bond between a pair of the normal nodes in each pre-transformation molecular graph) (pg. 6 Fig. 1).
the structural information describing relative positions of the atoms represented by the normal nodes
• Mansimov does not teach the recitation above. However, Gasteiger teaches a graph neural networks with directed edge embeddings and two-hop message passing as universal approximators for predictions (pg. 1 para. 1); wherein the model incorporates relative directional information (pg. 5 para. 4) by representing molecules by a point cloud of n atoms represented by points and denotes vector's norm, its direction and relative position (pg. 3 para. 3).
a spatial encoding representing a shortest path distance along the edges between the pair of the normal nodes in each pre-transformation molecular graph
• Mansimov does not teach the recitation above. However, Baingana teaches a graph embedding algorithms with node hierarchy being captured by betweenness centrality in describing the extent to which information is routed through a specific node by measuring the fraction of all shortest paths traversing it (pg. 4 para. 2); wherein the graph network is represented by edges, nodes and encoded network structure characteristics such as centrality and cardinality (pg. 14 para. 1).
wherein the encoded structural information is represented as one or more learnable scalar bias terms that are injected into a self-attention layer of an encoder of the transformer-based graph neural network, the self-attention layer being configured to modify a scaled query-key compatibility value by adding the learnable scalar bias terms prior to softmax
• Mansimov does not teach the recitation above. However, Guo teaches a class of Geometry-aware Self-Attention that extends self-attention network to explicitly and efficiently consider the relative geometry relations between the objects (pg. 10327 Abstract); wherein a geometric bias is applied to all the query-key pairs in an self-attention layer (i.e. wherein the encoded structural information is represented as one or more learnable scalar bias terms that are injected into a self-attention layer) (pg. 10331 col. 1 para. 3); wherein query-dependent geometric bias and key-dependent geometric bias can be added (pg. 10331 col. 1para. 3) prior to normalization by a softmax function (i.e. the self-attention layer being configured to modify a scaled query-key compatibility value by adding the learnable scalar bias terms prior to softmax) (pg. 10333 col. 1 para. 1); wherein the model applying the query-dependent geometric bias and key-dependent geometric bias adopts a Transformer-Base architecture, using 6 self-attention layers for both the encoder and the decoder (i.e. one or more learnable scalar bias terms that are injected into a self-attention layer of an encoder of the transformer-based graph neural network) (pg. 10333 col. 1 para. 3).
Claims 2 and 11 recite:
wherein, to perform the inference at inference time, the instructions are further configured to cause the processor to
receive inference-time input of an inference-time pre-transformation molecular graph at the transformer-based graph neural network; and output the inference-time post-transformation energy parameter value based on the inference-time pre-transformation molecular graph
• Mansimov teaches a conditional deep generative graph neural network that learns an energy function pg. 1 para. 1) for the inference of predicting molecular geometry; wherein the data input is represented by SMILES representations (pg. 5 para. 3) and contains information regarding the atoms, bonds, molecular mass and symmetry of molecules in each dataset (pg. 6 Fig. 1); wherein the task of molecular geometry prediction (i.e. output) is the generation of a set of plausible conformations using the 3-D coordinates (i.e. inference at inference time – trained neural network making a prediction) (pg. 2 para. 3).
Claims 3 and 12 recite:
wherein the encoded structural information includes a centrality encoding embedding for at least one of the normal nodes of each pre-transformation molecular graph
• Mansimov does not teach the recitation above. However, Baingana teaches graphs being used for encoding different networks (pg. 1 para. 1); wherein betweenness centrality is used to investigate the extent to which nodes are located between other pairs of nodes (pg. 9 para. 4).
Claims 4 and 13 recite:
wherein the centrality encoding is a degree of the at least one of the normal nodes of each pre-transformation molecular graph
• Mansimov does not teach the recitation above. However, Baingana teaches degree centrality which accounts for the number of edges adjacent to a particular node in directed graphs, specifying the in-degree and out-degree corresponding to the direction of the edge (pg. 9 para. 2); wherein the graph network is represented by edges, nodes and the real-valued pairwise dissimilarity between two nodes i and j computed via a well-defined criterion (pg. 14 para. 1).
Claims 5 and 14 recite:
wherein the centrality encoding assigns the at least one of the normal nodes two real-valued embedding vectors according to an indegree and an outdegree of a respective normal node
• Mansimov does not teach the recitation above. However, Baingana teaches degree centrality which accounts for the number of edges adjacent to a particular node in directed graphs, specifying the in-degree and out-degree corresponding to the direction of the edge (pg. 9 para. 2); wherein the graph network is represented by edges, nodes and the real-valued pairwise dissimilarity between two nodes i and j computed via a well-defined criterion (pg. 14 para. 1).
Claims 6 and 15 recite:
wherein the shortest path distance represented by the spatial encoding is a weighted shortest path distance
• Mansimov does not teach the recitation above. However, Baingana teaches node hierarchy being captured by betweenness centrality in describing the extent to which information is routed through a specific node by measuring the fraction of all shortest paths traversing it (pg. 4 para. 2); wherein the graph network is represented by edges, nodes and encoded network structure characteristics such as centrality and cardinality (pg. 14 para. 1); wherein the centrality constrained local linear embedding approach uses reconstruction weights determined under centrality constraints in closed form, with the weight preservation step, which turns out to decouple across nodes being solved by coordinate descent iterations (i.e. spatial encoding) (pg. 28 para. 1).
Claims 9 and 18 recite:
wherein the energy transformation is due to molecular relaxation of the molecular system
• Mansimov teaches as a deep generative graph neural network that learns an energy function by directly learning to generate energetically favorable molecular conformations (pg. 1 para. 1); wherein molecules can transition between conformations (i.e. energy change due to energy transformations) and end up in different local minima based on the stability of the respective conformations and environmental conditions (i.e. representing an energy relaxation in a molecular system being allowed to transition conformations in the search for molecular stability) (pg. 2 para. 4).
Claim 20 recites:
wherein the pre-transformation graph is a social graph that models a social network of friends, a map that models a network of locations connected by roads or railways, or a knowledge graph that models knowledge sources connected by references
• Mansimov does not teach the recitation above. However, Baingana teaches applications for centrality focused algorithm being used to visualize two real-world networks such as an underground train transit network, wherein nodes represent stations whereas the edges represent the routes connecting them; with the objective being to generate an embedding in which stations traversed by most routes are placed closer to the center, thus highlighting their relative significance in metro transit (pg. 22 para. 2).
Rationale for combining (MPEP §2142-2143)
Regarding claims 1-6, 9-15, and 18-20, it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine, in the course of routine experimentation and with a reasonable expectation of success, the methods of Mansimov in view of Gasteiger, Baingana and Guo because all references disclose methods for applying graph neural networks to perform predictions. The motivation would have been to:
• incorporate directional information to the proposed model (pg. 3 para. 2 Gasteiger);
• develop algorithms for embedding graphs with centrality structure (pg. 4 para. 3 Baingana); and
• to explicitly and efficiently consider the relative geometry relations between the objects (pg. 10327 Abstract Guo).
Therefore it would have been obvious to one of ordinary skill in the art to substitute the predictive graph neural network of Mansimov to the methods by Gasteiger, Baingana and Guo because such a substitution is no more than the simple substitution of one known element for another. One of ordinary skill in the art would be able to motivated to combine the teachings in these references with a reasonable expectation of success since the described teachings pertain to methods for applying graph neural networks to perform predictions.
Claims 7 and 16 are rejected under 35 U.S.C. 103(a) as being unpatentable over Mansimov, Gasteiger, Baingana and Guo as applied above to claims 1 and 10 further in view of Rkhami ("On the use of graph neural networks for virtual network embedding." 2020 International Symposium on Networks, Computers and Communications (ISNCC) IEEE (2020)), as cited on the 01/07/2026 Form PTO-892.
Claims 7 and 16 recite:
wherein each pre- transformation molecular graph further includes one virtual node fully connected by virtual edges to all normal nodes of a respective pre-transformation molecular graph
• Neither Mansimov or Gasteiger or Baingana or Guo teach the recited limitation above. However, Rkhami teaches the modeling of a virtual network embedding problem using a graph convolutional network based neural architecture (pg. 1 col. 1 para. 1) including a virtual node mapping step (pg. 2 col. 1 para. 5); wherein the generated graph is defined by the number of nodes n and the probability p of creating an edge between the nodes (pg. 4 col. 2 para. 4).
Rationale for combining (MPEP §2142-2143)
Regarding claims 7 and 16, it would have been prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to combine, in the course of routine experimentation and with a reasonable expectation of success, the methods of Mansimov,Gasteiger, Baingana and Guo in view of Rkhami because all references disclose methods for applying graph neural networks to perform predictions. The motivation would have been to incorporate graph neural networks for virtual network embedding for a more efficient exploration of the solutions space (pg. 1 col. 1 para. 1 Rkhami).
Therefore it would have been obvious to one of ordinary skill in the art to substitute the predictive graph neural network of Mansimov,Gasteiger, Baingana and Guo to the methods by Rkhami because such a substitution is no more than the simple substitution of one known element for another. One of ordinary skill in the art would be able to motivated to combine the teachings in these references with a reasonable expectation of success since the described teachings pertain to methods for applying graph neural networks to perform predictions.
Response to applicant's remarks in regard to Claim Rejection 35 U.S.C. ~ 103
The Remarks of 03/24/2026 have been fully considered but are not persuasive for the reasons below:
Applicant asserts in pg. 13 para. 5:
The newly added language further specifies the feature of original claim 8 by making explicit where the improvement occurs within the transformer's attention mechanism, and thus distinguishes over the cited reference. This amendment is supported by the specification's attention equations (e.g., Equations (4), (6), (7)) and figures (e.g., FIG. 3), which show the scalar bias terms being added to the scaled dot-product of the query and key vectors before the softmax operation. This provides a practical technical benefit because adding these learned bias terms before the softmax helps the attention mechanism focus more reliably on the most relevant parts of the molecular graph, making the model's predictions more accurate and stable In contrast, Mansimov, Gasteiger, and Baingana do not disclose structural information represented as learnable scalar bias terms (as also acknowledged by the Examiner on page 18 of the Office action) nor do they disclose a transformer encoder that adds learned scalar biases to the scaled query-key values prior to softmax. Li describes a signed-graph attention model in which attention weights are generated by applying a tan/MLP function to concatenated node embeddings (see Li, Eq. (1), col. 2, page 4774). Li, however, focuses on this MLP-based embedding-pair scoring and does not disclose transformer-style attention or any mechanism for modifying a scaled query-key value with learned scalar bias terms. Thus, Li also fails to disclose or suggest a transformer encoder that adds learned scalar biases to the scaled query-key values prior to softmax in combination with the remaining features of claim 1, and thus does not cure the deficiencies of Mansimov, Gasteiger, and Baingana
It is respectfully submitted that this is not persuasive because the new amendment as recited are taught by newly cited prior art to Guo. The Applicant states the disagreement with the rejections in the previous office action but no arguments have been provided in relation to said rejections. The motivation to incorporate the teachings by each cited reference is described in the claim rejection above. The argued claim 1 is not in condition for allowance because the prima facie case of obviousness has been established. MPEP 2141.III for "RATIONALES TO SUPPORT REJECTIONS UNDER 35 U.S.C. 103"; wherein "(G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference or to combine prior art reference teachings to arrive at the claimed invention." Furthermore, in this instant application, the amendments support existing claim rejections, in which the recited limitations are all addressed, see Claim Rejections above.
Conclusion
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 FRANCINI A FONSECA LOPEZ whose telephone number is (571)270-0899. The examiner can normally be reached Monday - Friday 8AM - 5PM ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Olivia Wise can be reached at (571) 272-2249. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/F.F.L./Examiner, Art Unit 1685
/OLIVIA M. WISE/Supervisory Patent Examiner, Art Unit 1685
Prosecution Insights