DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Status of Claims
Claims 1-20 are currently pending and examined on the merits.
Priority
Applicant’s claim for the benefit of a prior-filed application 63/338,487 filed 05 May 2022 is acknowledged and accepted. The effective filing date is May 5th, 2022.
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.
Claim(s) 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea of mental steps, mathematic concepts, organizing human activity, or a natural law without significantly more.
Step 2A, Prong 1
Claims 1-20 are drawn to a process (For example, “A Method”, clms. 1-16. “A system”, clms. 17-20). In accordance with MPEP § 2106, claims found to recite statutory subject matter (Step 1: YES) are then analyzed to determine if the claims recite any concepts that equate to an abstract idea, law of nature or natural phenomenon (Step 2A, Prong 1).
In the instant application, the claims (italicized) recite the following limitations that equate to an abstract idea:
Claims 1 recites:
converting, by a generative network, an input into a latent vector representation of a sample chemical compound, wherein the input is one of an order-dependent representation of the sample chemical compound and a molecular graph representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
determining, by an output neural network, one or more properties of the sample chemical compound based on the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
performing, by the output neural network, an optimization routine to select a candidate latent vector representation from among a plurality of latent vector representations based on the latent vector representation of the sample chemical compound, wherein the plurality of latent vector representations includes the latent vector representation of the sample chemical compound; (which is a mathematical concept of a mathematical calculation)
identifying, by the output neural network, a candidate chemical compound based on the candidate latent vector representation, (which is a mathematical concept of a mathematical calculation)
Claim 2 recites
the optimization routine is one of a gradient descent routine, an iterative expansion routine, and a genetic algorithm routine, (which is a mathematical concept of a mathematical calculation)
Claim 3 recites
the optimization routine is the gradient descent routine, (which is a mathematical concept of a mathematical calculation) ,
and wherein performing the gradient descent routine to select the candidate latent vector representation further comprises: setting the latent vector representation of the sample chemical compound as an initial value of the gradient descent routine; (which is a mathematical concept of a mathematical calculation)
descending along a gradient model of the plurality of latent vector representations to determine a gradient value of a given latent vector representation from among a remaining set of the plurality of latent vector representations; (which is a mathematical concept of a mathematical calculation)
determining whether the gradient value satisfies a convergence condition; (which is a mathematical concept of a mathematical calculation)
designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition. (which is a mathematical concept of a mathematical calculation)
Claim 4 recites:
the optimization routine is the iterative expansion routine, (which is a mathematical concept of a mathematical calculation)
setting the latent vector representation of the sample chemical compound as an initial value of the iterative expansion routine, (which is a mathematical concept of a mathematical calculation)
selecting a given latent vector representation from among the plurality of latent vector representations that is proximate to the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
determining whether the gradient value satisfies a convergence condition; (which is a mathematical concept of a mathematical calculation)
designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition, (which is a mathematical concept of a mathematical calculation)
Claim 5 recites:
the optimization routine is the genetic algorithm routine, (which is a mathematical concept of a mathematical calculation)
determining a fitness score for each latent vector representation of at least one set of the plurality of latent vector representations; (which is a mathematical concept of a mathematical calculation)
selecting a given latent vector representation from among each of the at least one set based on the fitness score; (which is a mathematical concept of a mathematical calculation)
performing, for each selected given latent vector representation, a reproduction routine to generate an additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
determining an additional fitness score associated with the additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
designating the additional latent vector representation as the candidate latent vector representation in response to the additional fitness score satisfying a convergence condition, (which is a mathematical concept of a mathematical calculation).
Claim 6 recites:
the generative network further comprises a graph convolutional neural network and an input neural network, which further limits claim 1.
Claim 7 recites:
generating, by the graph convolutional neural network, a graph of the sample chemical compound based on the input; (which is a mathematical concept of a mathematical calculation)
encoding the graph to generate the latent vector representation of the sample chemical compound based on at least one of an adjacency matrix of the graph convolutional neural network, one or more characteristics of the graph, one or more activation functions of the graph convolutional neural network, one or more node aggregation functions, and one or more weights of the graph convolutional neural network, (which is a mathematical concept of a mathematical calculation)
Claim 8 recites:
identifying one or more fragments and one or more substructures of the input, (which is a mathematical concept of a mathematical calculation)
generating one or more nodes based on the one or more substructures; (which is a mathematical concept of a mathematical calculation)
generating one or more edges based on the one or more fragments, wherein the graph is further based on the one or more nodes and the one or more edges, (which is a mathematical concept of a mathematical calculation)
Claim 9 recites: …wherein the latent vector representation of the sample chemical compound is an order independent representation, (which is a mathematical concept of a mathematical calculation)
Claim 10 recites:
the input is one of an order-dependent representation of the sample chemical compound and a molecular graph representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
… the sample chemical compound is an order independent representation, which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
determining, by an output neural network, one or more properties of the sample chemical compound based on the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
performing, by the output neural network, an optimization routine to select a candidate latent vector representation from among a plurality of latent vector representations based on the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
wherein the plurality of latent vector representations includes the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
and wherein the optimization routine is one of a gradient descent routine, an iterative expansion routine, and a genetic algorithm routine; which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation) (which is a mathematical concept of a mathematical calculation) and
identifying, by the output neural network, a candidate chemical compound based on the candidate latent vector representation, (which is a mathematical concept of a mathematical calculation)
11.
wherein the optimization routine is the gradient descent routine, (which is a mathematical concept of a mathematical calculation)
and wherein performing the gradient descent routine to select the candidate latent vector representation further comprises: setting the latent vector representation of the sample chemical compound as an initial value of the gradient descent routine; which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
descending along a gradient model of the plurality of latent vector representations to determine a gradient value of a given latent vector representation from among a remaining set of the plurality of latent vector representations; which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
determining whether the gradient value satisfies a convergence condition; (which is a mathematical concept of a mathematical calculation)
and designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition, (which is a mathematical concept of a mathematical calculation)
12.
wherein the optimization routine is the iterative expansion routine, which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
and wherein performing the iterative expansion routine to select the candidate latent vector representation further comprises: setting the latent vector representation of the sample chemical compound as an initial value of the iterative expansion routine; which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
selecting a given latent vector representation from among the plurality of latent vector representations that is proximate to the latent vector representation of the sample chemical compound; (which is a mathematical concept of a mathematical calculation)
determining a gradient value of the given latent vector representation; (which is a mathematical concept of a mathematical calculation)
determining whether the gradient value satisfies a convergence condition; (which is a mathematical concept of a mathematical calculation)
and designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition, (which is a mathematical concept of a mathematical calculation)
13.
wherein the optimization routine is the genetic algorithm routine, which further limits claim 10 ; (which is a mathematical concept of a mathematical calculation)
and wherein performing the genetic algorithm routine to select the candidate latent vector representation further comprises: determining a fitness score for each latent vector representation of at least one set of the plurality of latent vector representations; (which is a mathematical concept of a mathematical calculation)
selecting a given latent vector representation from among each of the at least one set based on the fitness score; (which is a mathematical concept of a mathematical calculation)
performing, for each selected given latent vector representation, a reproduction routine to generate an additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
determining an additional fitness score associated with the additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
and designating the additional latent vector representation as the candidate latent vector representation in response to the additional fitness score satisfying a convergence condition, (which is a mathematical concept of a mathematical calculation)
14.
wherein the generative network further comprises a graph convolutional neural network and an input neural network, (which is a mathematical concept of a mathematical calculation)
15. The method of Claim 14,
wherein converting the input into the latent vector representation of the sample chemical compound further comprises: generating, by the graph convolutional neural network, a graph of the sample chemical compound based on the input; (which is a mathematical concept of a mathematical calculation)
and encoding the graph to generate the latent vector representation of the sample chemical compound based on at least one of an adjacency matrix of the graph convolutional neural network, one or more characteristics of the graph, one or more activation functions of the graph convolutional neural network, one or more node aggregation functions, and one or more weights of the graph convolutional neural network, (which is a mathematical concept of a mathematical calculation)
16. The method of Claim 15 further comprising:
identifying one or more fragments and one or more substructures of the input; mental step
generating one or more nodes based on the one or more substructures; which further limits claim 15,
and generating one or more edges based on the one or more fragments, wherein the graph is further based on the one or more nodes and the one or more edges, which further limits claim 15.
17.
wherein: the input is one of an order-dependent representation of the sample chemical compound and a molecular graph representation of the sample chemical compound, and the latent vector representation of the sample chemical compound is an order independent representation; (which is a mathematical concept of a mathematical calculation)
and an output neural network configured to: determine one or more properties of the sample chemical compound based on the latent vector representation of the sample chemical compound; (which is a mathematical concept of a mathematical calculation)
perform an optimization routine to select a candidate latent vector representation from among a plurality of latent vector representations based on the latent vector representation of the sample chemical compound, (which is a mathematical concept of a mathematical calculation)
wherein the plurality of latent vector representations includes the latent vector representation of the sample chemical compound, and wherein the optimization routine is one of a gradient descent routine, an iterative expansion routine, and a genetic algorithm routine; which further limits claim 17; (which is a mathematical concept of a mathematical calculation)
and identify a candidate chemical compound based on the candidate latent vector representation, (which is a mathematical concept of a mathematical calculation) .
18.
wherein the optimization routine is the gradient descent routine, which further limits claim 18; (which is a mathematical concept of a mathematical calculation)
and wherein the output neural network is configured to: set the latent vector representation of the sample chemical compound as an initial value of the gradient descent routine; (which is a mathematical concept of a mathematical calculation)
descend along a gradient model of the plurality of latent vector representations to determine a gradient value of a given latent vector representation from among a remaining set of the plurality of latent vector representations; which further limits claim 18; (which is a mathematical concept of a mathematical calculation)
determine whether the gradient value satisfies a convergence condition; mental step
and designate the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition, (which is a mathematical concept of a mathematical calculation)
19.
wherein the optimization routine is the iterative expansion routine, and wherein the output neural network is configured to: set the latent vector representation of the sample chemical compound as an initial value of the iterative expansion routine; (which is a mathematical concept of a mathematical calculation)
select a given latent vector representation from among the plurality of latent vector representations that is proximate to the latent vector representation of the sample chemical compound; (which is a mathematical concept of a mathematical calculation)
determine a gradient value of the given latent vector representation; (which is a mathematical concept of a mathematical calculation)
determine whether the gradient value satisfies a convergence condition; (which is a mathematical concept of a mathematical calculation)
and designate the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition, (which is a mathematical concept of a mathematical calculation)
20.
wherein the optimization routine is the genetic algorithm routine, (which is a mathematical concept of a mathematical calculation)
and the output neural network is configured to: determine a fitness score for each latent vector representation of at least one set of the plurality of latent vector representations; (which is a mathematical concept of a mathematical calculation).
select a given latent vector representation from among each of the at least one set based on the fitness score; (which is a mathematical concept of a mathematical calculation)
perform, for each selected given latent vector representation, a reproduction routine to generate an additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
determine an additional fitness score associated with the additional latent vector representation; (which is a mathematical concept of a mathematical calculation)
and designate the additional latent vector representation as the candidate latent vector representation in response to the additional fitness score satisfying a convergence condition, (which is a mathematical concept of a mathematical calculation)
The claims recite an abstract idea of analyzing and identifying chemical structures (See MPEP 2106.07(a)).
These recitations are similar to the concepts of collecting information, analyzing it and displaying certain results of the collection and analysis in Electric Power Group, LLC, v. Alstom (830 F.3d 1350, 119 USPQ2d 1739 (Fed. Cir. 2016)), organizing and manipulating information through mathematical correlations in Digitech Image Techs., LLC v Electronics for Imaging, Inc. (758 F.3d 1344, 111 U.S.P.Q.2d 1717 (Fed. Cir. 2014)) and comparing information regarding a sample or test to a control or target data in Univ. of Utah Research Found. v. Ambry Genetics Corp. (774 F.3d 755, 113 U.S.P.Q.2d 1241 (Fed. Cir. 2014)) and Association for Molecular Pathology v. USPTO (689 F.3d 1303, 103 U.S.P.Q.2d 1681 (Fed. Cir. 2012)) that the courts have identified as concepts that can be practically performed in the human mind or mathematical relationships. Therefore, these limitations fall under the “Mental process” and “Mathematical concepts” groupings of abstract ideas.
While claims 1-20 recite performing some aspects of the analysis using a “model”, there are no additional limitations that indicate that this model requires anything other than carrying out the recited mental process or mathematical concept in a generic computer environment. Merely reciting that a mental process is being performed in a generic computer environment does not preclude the steps from being performed practically in the human mind or with pen and paper as claimed. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation in the mind but for the recitation of generic computer components, then if falls within the “Mental processes” grouping of abstract ideas. As such, claim(s) 1-20 recite(s) an abstract idea/law of nature/natural phenomenon (Step 2A, Prong 1: YES).
Step 2A, Prong 2
Claims found to recite a judicial exception under Step 2A, Prong 1 are then further analyzed to determine if the claims as a whole integrate the recited judicial exception into a practical application or not (Step 2A, Prong 2). The claims recite no additional elements. As such, claims 1-20 is/are directed to an abstract idea/law of nature/natural phenomenon (Step 2A, Prong 2: NO).
Step 2B
Claims found to be directed to a judicial exception are then further evaluated to determine if the claims recite an inventive concept that provides significantly more than the judicial exception itself (Step 2B). The claims do not include additional elements. Therefore, the claims do not amount to significantly more than the judicial exception itself (Step 2B: No). As such, claims 1-20 is/are not patent eligible.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-4, 6-12, and 14-19 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Zaccary Alperstein et al. (arXiv:1905.13343v2 [cs.LG] 3 Jun 2019, pages. 1-23).
References to claim limitations will be italicized.
Regarding claims 1, 10, and 17, Alperstein et al. discloses a variational autoencoder (VAE) defined over SMILES strings and graph-based representations of molecules (abstract).
Alperstein et al. further discloses encoding multiple SMILEs strings of a single molecule (for example, Fig. 1, pg. 14 Tox21 dataset) using a set of stacked recurrent neural networks, pooling hidden representations of each atom between SMILES representations, and using attentional pooling to build a final fixed-length latent representation (Abstract, Fig 1, re: clms. 1, 10, 17, … A method[system] comprising: converting, by a generative network, an input into a latent vector representation of a sample chemical compound, wherein the input is one of an order-dependent representation of the sample chemical compound and a molecular graph representation of the sample chemical compound…).
Alperstein et al. teaches an order independent representation in which the All SMILES VAE examines the structure associated with text characters instead of the text characters themselves, stating on pg. 2 “From this latent representation, the decoder RNN reconstructs a set of SMILES strings disjoint from those input to the encoder, ensuring that the latent representation only captures features of the molecule, rather than its SMILES realization.” (re: clms. 10, 17, … the latent vector representation of the sample chemical compound is an order independent representation…).
Alperstein et al. further discloses that a neural network regressor from the latent space to molecular properties can be used to perform gradient descent on molecular property prediction (Introduction, pg. 1-2). Alperstein et al. further discloses: “Simple property regressors jointly trained on this latent representation surpass the state-of-the-art for molecular property prediction, and facilitate exceptional gradient-based molecular property optimization when constrained to the region of prior containing almost all probability…” which reads on the application of regressors on latent representation for determination (Introduction, pg. 1-2, re: clms. 1, 10, 17, …determining, by an output neural network [an output neural network configured to: determine], one or more properties of the sample chemical compound based on the latent vector representation of the sample chemical compound).
Alperstein et al. further discloses an optimization routine on pg. 5, stating: “…We then perform gradient-based optimization of the property of interest with respect to the latent space, and decode the result to produce an optimized molecule…” and on pg. 16 (“Latent space optimization”, re: clm. 1,10, 17, … [perform]performing, by the output neural network, an optimization routine to select a candidate latent vector representation from among a plurality of latent vector representations based on the latent vector representation of the sample chemical compound, wherein the plurality of latent vector representations includes the latent vector representation of the sample chemical compound…).
Alperstein et al. further discloses identification of candidate chemical compounds, stating “...The design of new pharmaceuticals, OLED materials, and photovoltaics all require optimization within the space of molecules…” (Introduction), and further discloses on pg. 16-17 that the SMILES VAE “…learns a SMILES-derived generative model of molecules, rather than SMILES strings. The powerful, learned, hierarchical prior of the All SMILES VAE regularizes molecular optimization and property prediction.”, which reads on predicting candidate compounds based on latent vector representation (re: clm. 1, 10, 17 … [identify]identifying, by the output neural network, a candidate chemical compound based on the candidate latent vector representation.) Alperstein et al. further discloses a gradient descent application in the introduction ln. 2, and further teaches a gradient descent routine, stating on pg. 5, sec. 3.1 ( “We then perform gradient-based optimization of the property of interest with respect to the latent space, and decode the result to produce an optimized molecule…” ,re: clm. 10, 17, … the plurality of latent vector representations includes the latent vector representation of the sample chemical compound, and wherein the optimization routine is one of a gradient descent routine…
Alperstein et al. teaches a SMILES VAE in anticipation of claims 1, 10, and 17.
Regarding claim 2, Alperstein et al. discloses a gradient descent application in the introduction ln. 2, and further teaches a gradient descent routine, stating on pg. 5, sec. 3.1: “We then perform gradient-based optimization of the property of interest with respect to the latent space, and decode the result to produce an optimized molecule.” (re: clm. 2, …wherein the optimization routine is one of a gradient descent routine, an iterative expansion routine, and a genetic algorithm routine). Alperstein et al. anticipates claim 2.
Regarding claims 3, 11, and 18, Alperstein et al. discloses a gradient descent application in the introduction ln. 2, and further teaches a gradient descent routine, stating on pg. 5, sec. 3.1.
Alperstein et al. further teaches setting the latent vector as an initial value on pg. 6, stating: “Using the approximating posterior as the encoder, but always selecting the mean of each conditional Gaussian distribution (the maximum conditional a posteriori point), and a using beam search over the conditional likelihood as the decoder, 87.4% ± 1% of a held-out test set of ZINC250k (80/10/10 train/val/test split) is reconstructed accurately.”, which reads on ZINC250k being he initial starting position for optimization. Alperstein further discusses initial state on pg. 5, stating “The decoder is a single-layer LSTM, for which the initial cell state is computed from the latent representation by a neural network, and a linear transformation of the latent representation is concatenated onto each input.” (re: clms. 3, 11, 18 … wherein the optimization routine is one of a gradient descent routine…, [set]setting the latent vector representation of the sample chemical compound as an initial value of the gradient descent routine…)
Alperstein et al. further teaches that SMILES VAE can be used to perform gradient descent on molecular properties in the introduction, stating: “Bayesian optimization of molecular properties within the latent space or a neural network regressor from the latent space to molecular properties can be used to perform gradient descent on molecular properties with respect to the latent space…SMILES, the simplified molecular-input line-entry system, defines a character string representation of a molecule by performing a depth-first pre-order traversal of a spanning tree of the molecular graph…emitting characters for each atom, bond, tree-traversal decision, and broken cycle…. The resulting character string corresponds to a flattening of a spanning tree of the molecular graph, as shown in Figure 1.” (re: clms. 3, 11, 18… [descend]descending along a gradient model of the plurality of latent vector representations to determine a gradient value of a given latent vector representation from among a remaining set of the plurality of latent vector representations…).
Alperstein et al. further teaches determining a convergence condition, stating on pg. 2 “Simple property regressors jointly trained on this latent representation surpass the state-of-the-art for molecular property prediction, and facilitate exceptional gradient-based molecular property optimization when constrained to the region of prior containing almost all probability…” and on pg. 16, “…To compare with previous work as fairly as possible, we optimize 1000 random samples from the prior to convergence, collecting the last point from each trajectory with a valid SMILES decoding.” (re: clm. 3, 11, 18… [determine]determining whether the gradient value satisfies a convergence condition…).
Alperstein et al. further teaches a candidate latent vector representation via the a posteriori point associated with a ZINC250k dataset used in training. Alperstein et al. discloses 50k novel molecules based upon coordinates used in the initial convergence floor, stating in the results on pg. 6 (Results) “Using the approximating posterior as the encoder, but always selecting the mean of each conditional Gaussian distribution (the maximum conditional a posteriori point), and a using beam search over the conditional likelihood as the decoder, 87.4% ± 1% of a held-out test set of ZINC250k (80/10/10 train/val/test split) is reconstructed accurately. With the same beam search decoder, 98.5% ± 0.1% of samples from the prior decode to valid SMILES strings. We expect that enforcing grammatical constraints in the decoder LSTM, as described in Appendix D, would further increase these rates. All molecules decoded from a set of 50,000 independent samples from the prior were unique…” (re: clms. 3, 11, 18 … and [designate]designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition….) Alperstein et al taches a gradient descent optimization routine in anticipation of claims 3, 11, and 18.
Regarding claims 4, 12, and 19, Alperstein et al. discloses an iterative expansion routine, stating in the abstract “…we encode multiple SMILES strings of a single molecule using a set of stacked recurrent neural networks, pooling hidden representations of each atom between SMILES representations, and use attentional pooling to build a final fixed-length latent representation.”, which reads on converting a known molecule into a set of coordinates which are set as the initial point for exploration (re: clms. 4, 12, 19…the optimization routine is the iterative expansion routine, and wherein performing the iterative expansion routine to select the candidate latent vector representation further comprises: [set]setting the latent vector representation of the sample chemical compound as an initial value of the iterative expansion routine…)
Alperstein et al. further discloses selecting a latent vector representation via use of recurrent neural networks constraining molecular property optimization to the region associated with chemical compounds, stating “A fixed-length latent representation is distilled from the variable length RNN output using attentional mechanisms. From this latent representation, the decoder RNN reconstructs a set of SMILES strings disjoint from those input to the encoder, ensuring that the latent representation only captures features of the molecule, rather than its SMILES realization.” on pg. 2.
Alperstein et al. continues on the same page (2) stating “Simple property regressors jointly trained on this latent representation surpass the state-of-the-art for molecular property prediction, and facilitate exceptional gradient-based molecular property optimization when constrained to the region of prior containing almost all probability. We further demonstrate that the latent representation forms a near-bijection with the space of molecules, and is smooth with respect to molecular properties, facilitating effective optimization…” which reads on selecting a latent vector representation similar to the given sample compound (re: clms. 4, 12, 19 … [select]selecting a given latent vector representation from among the plurality of latent vector representations that is proximate to the latent vector representation of the sample chemical compound…).
Alperstein et al. further teaches (as previously disclosed) that a neural network regressor from the latent space to molecular properties can be used to perform gradient descent on molecular property prediction (Introduction, pg. 1-2). Alperstein et al. further discloses: “Simple property regressors jointly trained on this latent representation surpass the state-of-the-art for molecular property prediction, and facilitate exceptional gradient-based molecular property optimization when constrained to the region of prior containing almost all probability…” which reads on the application of regressors on latent representation for determination and also reads on determining a gradient value of a given latent vector representation (Introduction, re: clms. 4, 12, 19 … [determine]determining a gradient value of the given latent vector representation…).
Alperstein et al. further teaches (as previously disclosed) determining whether the gradient value satisfies a convergence condition on pg. 2 “Simple property regressors jointly trained on this latent representation surpass the state-of-the-art for molecular property prediction, and facilitate exceptional gradient-based molecular property optimization when constrained to the region of prior containing almost all probability…” and on pg. 16, “…To compare with previous work as fairly as possible, we optimize 1000 random samples from the prior to convergence…” (re: clm. 4, 12,19, …[determine]determining whether the gradient value satisfies a convergence condition)
Alperstein et al. further teaches (as previously shown) a candidate latent vector representation on pg. 6 (Results, re: clm. 4, 12, 19 … [designate]designating the given latent vector representation as the candidate latent vector representation in response to the gradient value satisfying the convergence condition…). Alperstein teaches an iterative expansion routine in anticipation of claims 4, 12, and 19.
Regarding claims 6 and 14, Alperstein et al. teaches a graph convolutional neural network and an input neural network (Abstract, Model architecture “It takes multiple distinct SMILES strings of the same molecule as input, and applies RNNs to them in parallel….”, re: clm. 6, 14, … the generative network further comprises a graph convolutional neural network and an input neural network.) Alperstein et al. anticipates claims 6 and 14.
Regarding claims 7 and 15, Alperstein et al. teaches a VAE generating a graph on pg. 4, stating “The pooling effectively sums messages propagated from many adjacent nodes in the molecular graph, analogous to a graph convolution, but the GRUs efficiently transfer information through many edges in each layer, rather than just one.” (re: clms. 7,15, … wherein converting the input into the latent vector representation of the sample chemical compound further comprises: generating, by the graph convolutional neural network, a graph of the sample chemical compound based on the input…).
Alperstein et al. further teaches passing along branches of the molecular graph on page 2, which reads on a weight, stating: “…the All SMILES VAE, which uses recurrent neural networks (RNNs) on multiple SMILES strings to implicitly perform efficient message passing along and amongst many flattened spanning trees of the molecular graph in parallel. A fixed-length latent representation is distilled from the variable length RNN output using attentional mechanisms.” (re: clms. 7,15 … encoding the graph to generate the latent vector representation of the sample chemical compound based on at least one of an adjacency matrix of the graph convolutional neural network, one or more characteristics of the graph, one or more activation functions of the graph convolutional neural network, one or more node aggregation functions, and one or more weights of the graph convolutional neural network.) Alperstein et al. teaches a graph convolutional neural network in anticipation of claims 7 and 15.
Regarding claims 8 and 16, Alperstein et al. teaches SMILES fragments in figures 1 and 15 (re: clm. 8, 16 … identifying one or more fragments and one or more substructures of the input…).
Alperstein further teaches ordered nodes on pg. 3, stating “The message at each symbol in the string is a weighted sum of the previous message and the current input, followed by a pointwise nonlinearity and subject to gating, as reviewed in Appendix B.1. This differs from explicit graph-based message passing in that the molecular graph is flattened into a chain corresponding to a depth-first pre-order traversal of a spanning tree, and the set of adjacent nodes that affect a message only includes the preceding node in this chain.” The set of adjacent nodes including only the preceding node reads on generating nodes based on SMILES fragments Alperstein et al. relates nodes to graphs, stating on pg. 4, “The pooling effectively sums messages propagated from many adjacent nodes in the molecular graph” (re: clm. 8, 16, … generating one or more nodes based on the one or more substructures…)
Alperstein et al. further teaches a gated recurrent unit on pg. 4, stating “The characters of the multiple SMILES strings are linearly embedded, and each string is preprocessed by a BiGRU [7], followed by a linear transformation, to produce the initial hidden representation h0 i for each SMILES string i.” and further states adjacent node propagation on pg. 4 (“The pooling effectively sums messages propagated from many adjacent nodes in the molecular graph, analogous to a graph convolution, but the GRUs efficiently transfer information through many edges in each layer, rather than just one.”), which reads on substructure, fragment, and edges consideration (re: clms. 8, 16, …generating one or more nodes based on the one or more substructures; and generating one or more edges based on the one or more fragments… wherein the graph is further based on the one or more nodes and the one or more edges…). Alperstein et al. teaches a graph convolutional neural network in anticipation of claims 8 and 16.
Regarding claim 9, Alperstein et al. teaches an order independent representation in which the All SMILES VAE examines the structure associated with text characters instead of the text characters themselves, stating on pg. 2 “From this latent representation, the decoder RNN reconstructs a set of SMILES strings disjoint from those input to the encoder, ensuring that the latent representation only captures features of the molecule, rather than its SMILES realization.” (re: clm. 9, …the latent vector representation of the sample chemical compound is an order independent representation.). Alperstein teaches an order independent representation in anticipation of claim 9.
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 5, and 13, and 20 is rejected under 35 U.S.C. 103 as being unpatentable over Alperstein et al. as applied to claims 1-4, 6-12, and 14-19 in view of Ahn et al. (arXiv:2007.04897v3 [q-bio.QM] 27 Oct 2020 pages 1-20).
Alperstein et al. is applied to claims 1-4, 6-12, and 14-19.
Regarding claims 5, 13, and 20, Alperstein et al. further discloses an optimization routine on pg. 5, stating: “…We then perform gradient-based optimization…” and latent vector representation on pg. 6 (Results, (re: clms. 5, 13, and 20 … wherein the optimization routine is the genetic algorithm routine…)
Alperstein et al. does not teach a genetic algorithm routine.
Anh et al. teaches a genetic expert-guided learning (GEGL), a framework for training a deep neural network (DNN) to generate highly-rewarding molecules (according to the concept of natural selection) both de-novo and using the ZINC database and ChEMBL database (Abstract, Contribution, pg. 2, pg. 7-8, re: clms. 5,13, 20… the optimization routine is the genetic algorithm routine, and wherein performing the genetic algorithm routine to select the candidate latent vector representation…).
Anh et al. further teaches determining a fitness score through the disclosure of the GEGL framework, as stated on pg. 3: “To apply our framework, we view de novo molecular design as a combinatorial optimization of discovering a molecule x, which maximizes the reward r(x), i.e., the desired property.”
Anh et al. continues to describe the scoring of molecules according to PenalizedLogP(x), an equation that determines the fitness of a molecule for use based on the drug-likeness of a molecule (the octanol-water experiment stated on pg. 2), along with the synthetic accessibility and ring penalty, which constrain the logP score based upon how complex synthesis of the molecule would be (pg. 6, sec 4.1). Anh et al. further teaches a three step procedure on pg. 3 which establishes policies to generates both known and novel molecules (based upon the ZINC dataset [“we pretrain the apprentice policy on the ZINC dataset”, pg. 6, and the ChEMBL dataset, [“For the experiments, we initialize the apprentice policy using the weights provided by Brown et al. [31],4 that was pretrained on the ChEMBL”, pg. 7]).
Anh et al. further teaches applying experiments (“benchmark[s]”, pg. 7) to de-novo molecules, listing results on Fig. 5, pg. 7.
Anh et al.’s teaching of molecules generated with deep learning (based on SMILES text rules, de-novo and known molecule policies, and structural viability scoring) reads on a fitness score, a reproduction routine, and selection of candidate molecules (re: clms. 5,13, 20 … determining a fitness score for each latent vector representation of at least one set of the plurality of latent vector representations… performing, for each selected given latent vector representation, a reproduction routine to generate an additional latent vector representation… determining an additional fitness score associated with the additional latent vector representation… designating the additional latent vector representation as the candidate latent vector representation in response to the additional fitness score satisfying a convergence condition).
Anh et al. does not teach latent vector representation (Anh et al. uses a long-short term memory network, as described on pg. 4) (re: clms. 5,13, 20 …latent vector representations…)
In KSR Int 'l v. Teleflex, the Supreme Court, in rejecting the rigid application of the teaching, suggestion, and motivation test by the Federal Circuit, indicated that “The principles underlying [earlier] cases are instructive when the question is whether a patent claiming the combination of elements of prior art is obvious. When a work is available in one field of endeavor, design incentives and other market forces can prompt variations of it, either in the same field or a different one. If a person of ordinary skill can implement a predictable variation, § 103 likely bars its patentability.” KSR Int'l v. Teleflex lnc., 127 S. Ct. 1727, 1740 (2007).
Applying the KSR standard of obviousness to Anh et al. and Alperstein et al., the examiner concludes that the combination of the GEGL according to Anh et al. with the all SMILES VAE as disclosed by Alperstein et al. represents a finding that there was some teaching, suggestion, or motivation, either in the references themselves or in the knowledge generally available to one of ordinary skill in the art, to modify the reference or to combine reference teachings.
One of ordinary skill in the art of computational chemistry would be motivated to combine the teachings of Anh et al. with Alperstein et al. because the combination would lead to a stronger chemical compound vector representation method. In support of this motivation, Anh et al. cites methods to optimize molecular embeddings on pg. 3, including one (“[17]”) using latent space, stating: “Methods such as gradient descent [10, 15], Bayesian optimization [7, 8, 11, 12], constrained Bayesian optimization [20], and particle swarm optimization [17] have been applied for continuous optimization of the embedding space.”
There would have been a reasonable expectation of success because the methods of both arts rely on and utilize openly accessible SMILES datasets. One skilled in the art of computational chemistry could use the teaching of Alperstein et al. to convert molecules to a latent vector representation which one skilled in the art could then apply to Anh et al.’s GEGL framework. Therefore, claims 5, 13, and 20 of the applicant’s invention would have been prima facie obvious to one of skill in the art at the time of filing of the application, absent evidence to the contrary.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOHN T STUBBS whose telephone number is (571)272-0340. The examiner can normally be reached M-F 8-5 EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Larry Riggs can be reached at 571-270-3062. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/J.T.S./Examiner, Art Unit 1686
/Karlheinz R. Skowronek/Supervisory Patent Examiner, Art Unit 1687
Prosecution Insights