Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in parent Application No. CN202111482319.9, filed on 12/07/2021.
Response to Arguments
Applicant’s argument filed 11/07/2025 have been fully considered but they are not persuasive.
Applicant’s Argument: On page 9-11 of Applicant’s response to rejections under 35 U.S.C. 101, applicant states that the amended claims no longer recite an abstract idea.
Examiner’s Response: Applicant’s argument is not persuasive. The claim limitations “learn physicochemical knowledge contained in compound structure in the unannotated compound data set” and “transfer knowledge from the at least two different types of compound properties to the pre-trained graph neural network” are considered as abstract idea of a mental process that can be performed in the human mind. A person having ordinary skills in the arts may be able to extract information from a dataset consisting of compound structure and properties and modifying a neural network to include the compound properties as additional input data.
Applicant’s Argument: On page 12-14 of Applicant’s response to rejections under 35 U.S.C. 101, applicant states “First, with reference to amended claim 1, an “optimized compound property prediction model” is a concrete set of trained parameters and architecture that represents a specialized tool for drug screening using specific ADMET criteria, and thus provides a tangible and practical application.
Second, the ordered combination of all of the elements in claim 1 provides various improvements in the technical field of computer functionality for drug discovery. The specification discloses that these improvements include enabling more accurate and earlier screening of compounds from a candidate compound library, and that these improvements are tied to claim 1 because the improvements are obtained by claim 1’s ordered combination of elements
Third, the same ordered combination of elements solves the technical problem of data scarcity in molecular property prediction (see Specification [0004] “deep learning model needs to acquire a large amount of annotated training data for supervised training”, [0036] “Moreover, as compared with the existing supervised training, the cost of acquiring the training data is reduced”) which is a meaningful limitation.”
Examiner’s Response: Applicant’s argument is not persuasive. An important consideration in determining whether a claim improves technology is the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome (see MPEP 2106.05(a)). The amended claims do not provide sufficient details to describe any technological improvement. If the specifications explicitly set forth an improvement but in a conclusory manner (see MPEP 2106.04(d)(1): a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art), the examiner should not determine the claim improves technology.
Applicant’s Argument: On page 14-17 of Applicant’s response to rejections under 35 U.S.C. 102 and 103, applicant states that Hao discloses a teacher and a student model that have different objectives and Hao does not disclose the three-stage process as claimed.
Examiner’s Response: Applicant’s argument is not persuasive. Claim 1 does not limit the interpretation to only training of one model. In claim 1, “a graph neural network”, and “a compound property prediction model” is recited. As recited, the graph neural network and the compound property prediction model may be different models. Hao (abstract) discloses a teacher-student framework that provides improvement to the field of molecular property prediction. The teacher model learns a general representation and transfer the knowledge into a smaller student model, where the student model is finetuned on a labeled dataset. The Applicant (Remarks, pg. 15) agrees that the teacher model goes through multi-task training of pre-training and fine-tuning through an iterative loop, but fails to disclose optimizing the model. In Hao (pg. 3, section 4.1, par. 1-2), the weights from the teacher are transfer to the student to further optimize the model to accurately predict the molecular properties.
Additionally, Sarshogh and Wenzel references are included to further teach the claim limitations.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 22, 26, and 30 rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claims contain subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention. The specification does not disclose any details on attaching and detaching task-specific output layers to and from the pre-trained graph neural network according to the flexible training schedule.
The Examiner interprets the claim limitation as modifying the architecture of the neural network based on the number of tasks. The neural network’s structure may consist of a number of output layers, where each outputs a prediction for each individual task.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-2, 5, 8-9, 12, 15-16, 19, 21, 23-25, 27-29, and 31 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding Claim 1:
Subject Matter Eligibility Analysis Step 1:
Claim 1 recites “A method of generating a trained computational model for use in screening candidate drug compounds, the method comprising” and is thus a process, one of the four statutory categories of patentable subject matter.
Subject Matter Eligibility Analysis Step 2A Prong 1:
“in compound structure in the unannotated compound data set” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. evaluation)
“” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
“then optimizing, ” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
Claim 1 therefore recites an abstract idea.
Subject Matter Eligibility Analysis Step 2A Prong 2:
"acquiring, by a computer, an unannotated compound data set” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
“pre-training, by a computer, a graph neural network using the unannotated compound data set, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
"acquiring, by a computer, a plurality of annotated compound data sets, each annotated compound data set being annotated with one kind of compound property, wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes” (This step is directed to data gathering, which is understood to be insignificant extra solution activity - see MPEP 2106.05(g))
“performing, by a computer, multi-task training on the pre-trained graph neural network using the plurality of annotated compound data sets, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“wherein said multi-task training comprises a flexible training schedule in which an entry and exit of each task is controlled based on a pre-defined number of training rounds specific to that task, thereby obtaining a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, wherein each task corresponds to a respective one of the annotated compound data sets” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“then optimizing, by the computer, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is directed to the abstract idea.
Subject Matter Eligibility Analysis Step 2B:
"acquiring, by a computer, an unannotated compound data set” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
“pre-training, by a computer, a graph neural network using the unannotated compound data set, trained graph neural network” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
"acquiring, by a computer, a plurality of annotated compound data sets, each annotated compound data set being annotated with one kind of compound property, wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes” (This step is directed to transmitting or receiving information, which is understood to be insignificant extra solution activity and well understood, routine and conventional activity of transmitting and receiving data as identified by the court - see MPEP 2106.05(d))
“performing, by a computer, multi-task training on the pre-trained graph neural network using the plurality of annotated compound data sets, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“wherein said multi-task training comprises a flexible training schedule in which an entry and exit of each task is controlled based on a pre-defined number of training rounds specific to that task, thereby obtaining a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, wherein each task corresponds to a respective one of the annotated compound data sets ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“then optimizing, by the computer, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
The additional elements as disclosed above alone or in combination do not recite significantly more than the abstract idea itself as they are mere insignificant extra solution activity in combination of generic computer functions being implemented with generic computer elements in a high level of generality to perform the disclosed abstract idea above. Therefore, Claim 1 is subject-matter ineligible.
Regarding Claim 8:
The claim recites a system that performs the method as described in claim 1. Therefore, claim 8 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 8 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“An electronic device, comprising: at least one processor; and a storage device, in communication with the at least one processor, wherein the storage device stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform operations, the operations for generating a trained computational model for use in screening candidate drug compounds, the operations comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claim 15:
The claim recites a system that performs the method as described in claim 1. Therefore, claim 15 is rejected for the same reasons as disclosed for claim 1. The limitations for additional elements of claim 15 are analyzed below.
Subject Matter Eligibility Analysis Step 2A Prong 1:
Please see Step 2A Prong 1 analysis of claim 1
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“A non-transitory computer readable storage medium, storing computer instructions, wherein the computer instructions, when executed by a computer, cause the computer to perform operations, the operations for generating a trained computational model for use in screening candidate drug compounds, the operations comprising” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 2, 9, and 16:
Subject Matter Eligibility Analysis Step 2A Prong 1:
“performing, ” (a mental process that can be performed in the human mind with the aid of pen and paper, i.e. judgement)
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
"acquiring, by the computer, a target annotated compound data set, the target annotated compound data set being annotated with a target kind of compound property” (This step is directed to data gathering, which is understood to be insignificant extra solution activity (2106.05(g) in step 2A prong 2) and well understood, routine and conventional activity of transmitting and receiving data as identified by the court (2106.05(d) in step 2B))
“performing, by the computer, ” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 5, 12, and 19:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
"obtaining, by the computer, self-supervised information based on unannotated compound data in the unannotated compound data set” (This step is directed to data gathering, which is understood to be insignificant extra solution activity (2106.05(g) in step 2A prong 2) and well understood, routine and conventional activity of transmitting and receiving data as identified by the court (2106.05(d) in step 2B))
“using, by the computer, the unannotated compound data as an input and the self-supervised information as an output, to pre-train the graph neural network to obtain the pre- trained graph neural network” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
“wherein the self-supervised information comprises at least one of: a compound local structure, a compound bond length, a compound bond angle, or a molecular fingerprint” (merely specifies a particular technological environment in which the abstract idea is to take place, ie. a field of use, and thus does not integrate the abstract idea into a practical application nor cannot provide significantly more than the abstract idea itself - see MPEP 2106.05(h))
Regarding Claims 21, 25, and 29:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein the pre-trained graph neural network, the compound property prediction model, and the optimized compound property prediction model all share and progressively update a single, continuous set of trainable parameters of a graph neural network backbone” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Regarding Claims 23, 27, and 31:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein during the multi-task training, the pre-trained graph neural network's parameters are updated continuously, such that when a first task is exited and a second task is introduced according to the flexible training schedule, the parameters are not reset” (merely specifies a particular technological environment in which the abstract idea is to take place, ie. a field of use, and thus does not integrate the abstract idea into a practical application nor cannot provide significantly more than the abstract idea itself - see MPEP 2106.05(h))
Regarding Claims 24 and 28:
Subject Matter Eligibility Analysis Step 2A Prong 1: None
Subject Matter Eligibility Analysis Step 2A Prong 2 & 2B:
“wherein the flexible training schedule is performed such that a first task of the multi-task training functions as pre-training for a subsequently introduced second task of the multi-task training” (mere instructions to apply the exception using a generic computer component - see MPEP 2106.05(f))
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2, 5, 8-9, 12, 15-16, 19, and 21-31 are rejected under 35 U.S.C. 103 as being unpatentable over Hao, “ASGN: An Active Semi-supervised Graph Neural Network for Molecular Property Prediction” in view of Sarshogh (US20220180201A1) and Wenzel, “Predictive Multitask Deep Neural Network Models for ADME-Tox Properties: Learning from Large Data Sets”.
Regarding claim 1, Hao teaches:
“A method of generating a trained computational model for use in screening candidate drug compounds, the method comprising” (abstract; pg. 1, Section 1, par. 1; Figure 2, A teacher-student framework is proposed to generate a trained student model for molecular property prediction. Prediction of the properties of molecules can help improve the process of drug discovery for developing specific medicines (candidate drug compounds).)
“acquiring, by a computer, an unannotated compound data set” ([pg. 6, section 5.1, par. 1-2; pg. 7, section 5.4, par. 1], The QM9 dataset contains the properties of compounds. The experiments are conducted with 5000 labeled molecules and the rest is a set of unlabeled molecules.)
“pre-training, by a computer, a graph neural network using the unannotated compound data set, to learn physicochemical knowledge contained in compound structure in the unannotated compound data set, to obtain a pre-trained graph neural network” ([pg. 3, section 4.1, par. 1-2; pg. 3, section 4.2, par. 1-2; pg. 4, section 4.2.2, par. 1], A teacher model is pre-trained using semi-supervised learning with an unlabeled dataset to learn the general representation for molecular graphs. The teacher model is a graph neural network. Training of the teacher model consists of node level representation learning, which captures domain knowledge from geometry information of a molecular graph.)
“acquiring, by a computer, a plurality of annotated compound data sets, each annotated compound data set being annotated with one kind of compound property, wherein the plurality of annotated compound data sets comprise at least two of: ” ([pg. 6, section 5.1, par. 1-2; pg. 7, section 5.4, par. 1], The experiments are conducted using QM9 and OPV datasets. The QM9 dataset contains the quantum mechanical properties of compounds. The experiments are conducted with 5000 labeled molecules and the rest is a set of unlabeled molecules.)
“performing, by a computer, multi-task training on the pre-trained graph neural network using the plurality of annotated compound data sets, to transfer knowledge from the at least two different types of compound properties to the pre-trained graph neural network” ([pg. 3, section 4.1, par. 1-2; pg. 3, section 4.2, par. 1-2; pg. 6, section 5.1, par. 1-2; Equation 10; Table 1], The task of the teacher model is to learn the general representation for molecular graphs from both labeled set and unlabeled set. The teacher model is required to learn several tasks simultaneously and Table 1 shows the results of the experiments for a plurality of different properties. The teacher model is optimize using 3 different loss terms as shown in Equation 10. The teacher model is optimized based on a combined loss that consists of a loss on labeled molecules and two unsupervised losses on all molecules.)
“wherein said multi-task training comprises a flexible training schedule in which an entry and exit of each task is controlled based on a pre-defined number of training rounds specific to that task, thereby obtaining a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, ” ([pg. 3, section 4.1, par. 1-2; pg. 3, section 4.2, par. 1-2; pg. 6, section 5.1, par. 1-2; pg. 6, section 5.2, par. 3], A teacher model is pre-trained using semi-supervised learning with an labeled dataset to learn the general representation for molecular graphs. The student model (compound property prediction model) is trained for molecular property prediction. In the specification of the claimed invention (par. 35), a first task is defined as a pre-training and the second task is defined as downstream fine-tuning. A specific number of training rounds may be defined for each task. Hao discloses that the experiments are conducted for 20 epochs by the teacher model and 20 epochs by the student model. The teacher model performs pre-training to learn the general representations and the student model is fine-tuned for molecular property predictions.)
“then optimizing, by the computer, the compound property prediction model by optimizing its parameters ” ([pg. 3, section 4.1, par. 1-2; pg. 5, section 4.3, par. 1-3; pg. 5, section 4.4, par. 1], The teacher model and student model are finetuned iteratively until accuracy budget is reached by implementing an active learning method for data selection. The student model learns to predict molecular property prediction.)
Hao does not explicitly disclose an implementation of “wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes”, “a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, wherein each task corresponds to a respective one of the annotated compound data sets”, and “then optimizing, by the computer, the compound property prediction model by optimizing its parameters in a single task mode using only one annotated compound data set in the plurality of annotated compound data sets, to obtain an optimized compound property prediction model which is configured to, in response to input data representing a candidate drug compound, computationally generate at least a prediction of an ADMET property of the candidate drug compound, wherein the only one annotated compound data set represents an ADMET endpoint”.
However, Sarshogh discloses in the same field of endeavor:
“a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, wherein each task corresponds to a respective one of the annotated compound data sets” ([0101-0103, Figure 5], A embedding model may be trained using multi-task training and the embedding model may output features to a plurality of property predictors. Each property predictor may perform a task that is different and each task is associated with its own training data. The training data consists of labeled training examples.)
“then optimizing, by the computer, the compound property prediction model by optimizing its parameters in a single task mode using only one annotated compound data set in the plurality of annotated compound data sets, to obtain an optimized compound property prediction model which is configured to, in response to input data representing a candidate drug compound, computationally generate at least a prediction of an ADMET property of the candidate drug compound, ” ([0101-0103, 0107, 0123-0126], The embedding model is finetuned by back propagation of the loss associated with the tasks and updating the weights of the model. Each training dataset is associated with a single task. For example, the second task may be predicting whether the molecule is toxic, which is an ADMET property.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “a compound property prediction model that is configured to computationally generate a prediction of a plurality of kinds of properties of a compound, wherein each task corresponds to a respective one of the annotated compound data sets”, and “then optimizing, by the computer, the compound property prediction model by optimizing its parameters in a single task mode using only one annotated compound data set in the plurality of annotated compound data sets, to obtain an optimized compound property prediction model which is configured to, in response to input data representing a candidate drug compound, computationally generate at least a prediction of an ADMET property of the candidate drug compound” from Sarshogh into the teaching of Hao. Doing so can improve the property predictions of molecules where limited data exists for a specific task by implementing multi-task training to learn generic representations, followed by specific downstream task learning (Sarshogh, abstract, par. 0107).
Hao in view of Sarshogh does not explicitly disclose an implementation of “wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes”, “wherein the only one annotated compound data set represents an ADMET endpoint”. However, Wenzel discloses in the same field of endeavor:
“wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes” ([abstract; pg. 3, col. 2, par. 1; pg. 4, col. 2, par. 1-4], A DNN model is used to model ADME-Tox data. Multiple datasets are implemented in the experiments such as ChEMBL and Sanofi in-house laboratory data. These datasets contain biological activities and physiochemical properties.)
“wherein the only one annotated compound data set represents an ADMET endpoint” ([abstract; pg. 3, col. 2, par. 1; Figure 1], The embedding model is finetuned by back propagation of the loss associated with the tasks and updating the weights of the model. Each training dataset is associated with a single task. For example, the second task may be predicting whether the molecule is toxic, which is an ADMET property.)
It would be obvious to one of the ordinary skill in the art before the effective filing date of the claimed invention to incorporate the teaching of “wherein the plurality of annotated compound data sets comprise at least two of: a compound data set annotated with absorption, distribution, metabolism, excretion and toxicity (ADMET) properties, a compound data set annotated with biological activities, and a compound data set annotated with compound physicochemical attributes”, “wherein the only one annotated compound data set represents an ADMET endpoint” from Wenzel into the teaching of Hao in view of Sarshogh. Doing so can improve the property predictions of molecules by implementing heterogeneous data to train the model in predicting ADME-Tox properties (Wenzel, abstract).
Regarding claim 8:
Claim 8 recites a system that performs the same process as described in Claim 1. Therefore claim 8 is rejected under the same reasons mention for claim 1. The additional elements of claim 8 is addressed below by Hao:
“An electronic device, comprising: at least one processor; and a storage device, in communication with the at least one processor, wherein the storage device stores instructions executable by the at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform operations, the operations for generating a trained computational model for use in screening candidate drug compounds, the operations comprising” ([abstract; pg. 6, section 5.2, par. 1-4; Algorithm 1], The experiments are conducted using CPUs and GPU. It is implied that the algorithm of the framework runs on a computer system that consists of processors and a memory that stores the instructions to perform the experiments.)
Regarding claim 15:
Claim 15 recites a system that performs the same process as described in Claim 1. Therefore claim 15 is rejected under the same reasons mention for claim 1. The additional elements of claim 15 is addressed below by Hao:
“A non-transitory computer readable storage medium, storing computer instructions, wherein the computer instructions, when executed by a computer, cause the computer to perform operations, the operations for generating a trained computational model for use in screening candidate drug compounds, the operations comprising” ([pg. 6, section 5.2, par. 1-4; Algorithm 1], The experiments are conducted using CPUs and GPU. It is implied that the algorithm of the framework runs on a computer system that consists of processors and a memory that stores the instructions to perform the experiments.)
Regarding claims 2, 9, and 16 Hao teaches:
“acquiring, by a computer, a target annotated compound data set, the target annotated compound data set being annotated with a target kind of compound property” ([pg. 5-6, section 4.4, par. 1], In each iteration, a subset of molecules is selected and the ground truth property labels are computed.)
“performing, by a computer, fine-tuning on the compound property prediction model using the target annotated compound data set, to obtain a fine-tuned compound property prediction model for computationally generating a prediction of the target kind of compound property of the compound” ([pg. 3, section 4.1, par. 1-2; pg. 5-6, section 4.4, par. 1], The data is used as a labeled dataset to finetune the teacher and the student model. The most diverse samples are selected for fine-tuning based on a maximum distance between the labeled and unlabeled set. The models are finetuned on the new labelled data set to improve the performance to accurately predict properties of those sampled molecules.)
Regarding claims 5, 12, and 19 Hao teaches:
“obtaining, by the computer, self-supervised information based on unannotated compound data in the unannotated compound data set” ([pg. 3, section 3, par. 2-4; pg. 5, section 4.3, par. 1-3, Figure 2], The student model is used to infer the whole unlabeled dataset and assign each unlabeled data a pseudo label consisting of the student’s prediction of its properties.)
“using, by the computer, the unannotated compound data as an input and the self-supervised information as an output, to pre-train the graph neural network to obtain the pre- trained graph neural network” ([pg. 5, section 4.3, par. 1-3, Figure 2], The pseudo labels are passed to the teacher model and the teacher model also learns from the pseudo labels. The teacher model uses semi-supervised learning in which it learns from an unlabeled dataset and the pseudo label provided by the student model’s output.)
“wherein the self-supervised information comprises at least one of: a compound local structure, a compound bond length, a compound bond angle, or a molecular fingerprint” ([pg. 1-2, section 1, par. 6; pg. 4, section 4.2.2, par. 1], The molecular features on a local scale may consist of atoms and bonds. The node embeddings may be used to reconstruct node types and distances between nodes.)
Regarding Claims 21, 25, and 29, Hao teaches:
“wherein the pre-trained graph neural network, the compound property prediction model, and the optimized compound property prediction model all share and progressively update a single, continuous set of trainable parameters of a graph neural network backbone” ([pg. 5, section 4.3, par. 1-3; pg. 7, section 5.3.1, par. 1], The teacher model’s weights are transfer to the student model and the student model is finetuned to learn the target properties. An effectiveness experiment is conducted where all baselines are conducted on the same network backbone.)
Regarding Claims 22, 26, and 30, Hao in view of Sarshogh and Wenzel:
“dynamically attaching and detaching a plurality of task-specific output layers to and from the pre-trained graph neural network according to the flexible training schedule, wherein each of the plurality of task-specific output layers corresponds to one of the different types of compound properties” ([Wenzel, pg. 3, col. 2, par. 1; Figure 1b], Figure 1b depicts a multitask deep neural network for predicting compound properties. The deep neural network comprises of X output units that corresponds to X datasets and targets. The structure of the deep neural network may change depending on the number of training datasets. It is implied that output units may be added or removed depending on the number of training datasets during the training process.)
Regarding Claims 23, 27, and 31, Hao in view of Sarshogh and Wenzel:
“wherein during the multi-task training, the pre-trained graph neural network's parameters are updated continuously, such that when a first task is exited and a second task is introduced according to the flexible training schedule, the parameters are not reset” ([Wenzel, pg. 4, col. 1, par. 1-2], The multitask deep neural network is trained over multiple iterations for all tasks. Prior to switching to the next task, the weights of the previous model is loaded for training.)
Regarding Claims 24 and 28, Hao in view of Sarshogh and Wenzel:
“wherein the flexible training schedule is performed such that a first task of the multi-task training functions as pre-training for a subsequently introduced second task of the multi-task training” ([Wenzel, pg. 4, col. 1, par. 1-2], The multitask deep neural network is trained over multiple iterations for all tasks. Prior to switching to the next task, the weights of the previous model are loaded for training. The training of a new task involves the parameters of the previous models. Thus, the earlier tasks functions as pre-training for subsequential tasks.)
Conclusion
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 GARY MAC whose telephone number is (703)756-1517. The examiner can normally be reached Monday - Friday 8:00 AM - 5:00 PM.
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, Abdullah Kawsar can be reached at (571) 270-3169. 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.
/GARY MAC/ Examiner, Art Unit 2127
/ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127