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 .
Claim Status
Claims 15-16, 18-19, 21-87, 89, 91-117, and 120 are cancelled.
Claims 1-14, 17, 20, 88, 90, 118-119, and 121-122 are currently pending and under exam herein.
Claims 1-14, 17, 20, 88, 90, 118-119, and 121-122 are rejected.
Claim 88 is objected to.
Priority
This application is the U.S. National Stage of International Application No.
PCT/US2020/044646, filed on 31 July 2020, which claims the benefit of U.S. Provisional Application No. 62/882,150 and 62/882,159 both filed on 2 August 2019. The claimed priority is acknowledged. At this point in examination, the effective filing date of claims 1-14, 17, 20, 88, 90, 118-119, and 121-122 is 2 August 2019.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 26 January 2022, 31 May 2022, 16 October 2024, 25 July 2024, and 10 September 2024 comply with 37 CFR 1.98. Accordingly, all references listed have been considered by the examiner.
Drawings
The drawings filed on 26 January 2022 have been received and are accepted.
Claim Objections
Claim 88 is objected to because “an embedding at a to a system” should read “an embedding at a system.” Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-14, 17, 20, 88, 90, 118-119, and 121-122 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
The term “desired level” in claims 1, 4, and 88 is a relative term which renders the claims indefinite. The term “desired level” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention. This renders the claims indefinite because it is unclear when the first updated point is provided. Claims 2-14, 17, 20, 90, 118-119, and 122 are similarly rejected due to their dependency upon claim 1.
Claim 4 recites the limitation “the optionally iterated further updated point.” There is insufficient antecedent basis for this limitation in the claim. This rejection can be overcome by (1) amending claim 4 to remove the term optionally, or (2) amending claim 3 to indicate that iterating the process is optional.
Claim 8 recites the limitation “the two or more composite functions.” There is insufficient antecedent basis for this limitation in the claim. This rejection can be overcome by (1) amending claim 8 to recite “two or more composite functions,” (2) amending claim 8 to recite “the two or more component functions,” or (3) amending claim 7 to recite “two or more composite functions.”
Claim 12 recites the limitation “the marginal distribution.” There is insufficient antecedent basis for this limitation in the claim. This rejection can be overcome by (1) amending claim 12 to recite “a marginal distribution” or (2) amending claim 1 to provide sufficient antecedent basis.
Claim 88 recites the limitation "the function" in step (a). There is insufficient antecedent basis for this limitation in the claim. This rejection can be overcome by amending claim 88 to recite “a function.”
Claim 121 fails to recite a conjunction before the last limitation, rendering the claim unclear as to whether all limitations are required. This rejection may be overcome by amending the independent claims to include a conjunction (e.g. and/or) before the last limitation. For purposes of the present examination, the independent claims will be interpreted to require all listed limitations.
Claim Rejections - 35 USC § 101
Claim 122 is rejected under 35 U.S.C. 101 because it does not fall within one of the four enumerated categories of statutory subject matter. Claims 1-14, 17, 20, 88, 90, 118-119, and 121 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (abstract ideas) without significantly more. Under MPEP § 2106, subject matter is patent eligible when the claimed invention is to one of the four statutory categories of invention [Step 1], and the claim is not directed to a judicial exception [Step 2A] unless the claim as a whole includes additional limitations amounting to significantly more than the exception [Step 2B].
Step 1
Claims 1-14, 17, 20, 88, 90, 118-119, and 121 describe inventions that are to one of the statutory categories. In Step 1, a claim must fall within one of the four enumerated categories of statutory subject matter (process, machine, manufacture, or composition of matter); a claim falling outside these categories is ineligible without further analysis. See MPEP § 2106.03. Claims 1-14, 17, 20, 90, 118-119, and 121 are properly to one of the four statutory categories because the claimed invention is a method, which falls into the process category [Step 1: Yes]. Claim 88 is properly to one of the four statutory categories because the claimed invention is a system, which falls into the machine category [Step 1: Yes].
Claim 122 recites “a set of parameters that characterize the behavior of a supervised model, an encoder or a decoder.” A set of parameters is not directed to a statutory category of invention because it described mere data or information. Therefore, claim 122 is not eligible for patent protection at Step 1 because it is directed to non-statutory subject matter.
Step 2A
Under Step 2A, a claim is directed to a judicial exception if, under the broadest reasonable interpretation, it recites an abstract idea, law of nature, or natural phenomena [Prong One] without the claim as a whole integrating the exception into a practical application [Prong Two]. Abstract ideas include mathematical concepts, mental processes, and certain methods of organizing human activity. Mathematical concepts encompass mathematical relationships, formulas, equations, and mathematical calculations. See MPEP § 2106.04(a)(2)(I). Mental processes involve concepts that can be performed in the human mind or by a human with the aid of pen and paper, such as observations, evaluations, judgments, or opinions. See MPEP § 2106.04(a)(2)(III). Certain methods of organizing human activity include fundamental economic principles, commercial or legal interactions, and managing personal behavior or relationships. See MPEP § 2106.04(a)(2)(II). Laws of nature and natural phenomena, include naturally occurring principles/relations and nature-based products that are naturally occurring or that do not have markedly different characteristics compared to what occurs in nature. See MPEP § 2106.04(b)-(c).
Prong One
A claim recites a judicial exception when it sets forth or describes a law of nature, natural phenomenon, or abstract idea. Claims 1-14, 17, 20, 88, 90, 118-119, and 121 recite abstract ideas that fall into the groupings of mathematical concepts and mental processes.
Claim 1 recites the following limitations, which describe abstract ideas within the mathematical concepts and/or mental processes groupings:
(b) calculating a change in the function in relation to the embedding at the starting point according to a step size, the calculated change enabling providing a first updated point in the functional space;
(c) upon reaching a desired level of the function within a particular threshold at the first updated point in the functional space providing the first updated point; and
(d) obtaining a probabilistic improved biopolymer sequence from the decoder.
The limitation of calculating a change describes the foundation of iterative optimization or first-order mathematical approximations, which constitutes an abstract idea within the mathematical concepts or mental processes groupings when a person could mentally evaluate how a change would affect function. The limitation of upon reaching a desired level describes a mathematical condition/check and iterative optimization process in the abstract embedding space, which constitutes an abstract idea within the mathematical concepts and mental processes groupings when a person could mentally evaluate whether a candidate sequence is optimal with respect to its desired function. The limitation of obtaining a probabilistic improved biopolymer describes decoding from a latent/embedding vector to a probability distribution over amino acids, which constitutes an abstract idea within the mathematical concepts and mental processes groupings when a person having ordinary skill in the art could conceptualize probable amino acid choices at each position that would improve the function based on judgment and experience.
Claims 3-8 and 10-14 recite the following limitations, which describe abstract ideas within the mathematical concepts and/or mental processes groupings:
Claim 3 recites calculating a second change in the function with regard to the embedding at the first updated point in the functional space; and iterating the process of calculating the second change in the function with regard to the embedding at a further updated point.
Claim 4 recites wherein providing the first updated point can be performed upon reaching a desired level of the function within a particular threshold at the optionally iterated further updated point, and providing the further updated point includes providing the iterated further updated point to the decoder network.
Claim 5 recites wherein the embedding is a continuously differentiable functional space representing the function and having one or more gradients.
Claim 6 recites wherein calculating the change of the function with regard to the embedding comprises taking a derivative of the function with regard to the embedding.
Claim 7 recites wherein the function is a composite function of two or more component functions.
Claim 8 recites wherein the composite function is a weighted sum of the two or more composite functions.
Claim 10 recites wherein correlations between residues in a probabilistic sequence comprising a probability distribution of residue identities are considered in a sampling process using conditional probabilities that account for the portion of the sequence that has already been generated.
Claim 11 recites further comprising selecting the maximum likelihood improved biopolymer sequence from a probabilistic biopolymer sequence comprising a probability distribution of residue identities.
Claim 12 recites comprising sampling the marginal distribution at each residue of a probabilistic biopolymer sequence comprising a probability distribution of residue identities.
Claim 13 recites wherein the change of the function with regard to the embedding, is calculated by calculating the change of the function with regard to the encoder, then the change of the encoder to the change of the decoder, and the change of the decoder with regard to the embedding.
Claim 14 recites providing the first updated point in the functional space or further updated point in the functional space to the decoder network to provide an intermediate probabilistic biopolymer sequence, providing the intermediate probabilistic biopolymer sequence to the supervised model network to predict the function of the intermediate probabilistic biopolymer sequence, calculating the change in the function with regard to the embedding for the intermediate probabilistic biopolymer to provide a further updated point in the functional space.
The limitations of claim 3 describe repeated application of an optimization algorithm in vector space, which constitutes an abstract idea within the mathematical concepts and mental processes groupings when iteratively refining a conceptual design can be accomplished through repeated judgment. The limitation of claim 4 narrows the abstract idea of claim 3 by describing conditional checking of a threshold in the embedding space and routing the vector to the decoder, which are abstract ideas within the mathematical concepts and mental processes groupings. The limitation of claim 5 narrows the abstract ideas of claim 1 by describing a differentiable vector space with gradients, which is a mathematical concept. The limitation of claim 6 narrows the abstract ideas of claim 1 by describing derivative computation, which is a mathematical concept. The limitations of claims 7 and 8 narrow the abstract ideas of claim 1 by specifying that the function is a composite function and that the composite function is a weighted sum, which are mathematical concept. The limitation of claim 10 describes a mathematical framework for sequence generation by manipulating probability values in a distribution to generate a sequence while enforcing statistical dependencies between positions, which constitutes a mathematical concept. The limitation of claim 11 describes an argmax operation or maximum likelihood estimation over a probability distribution to select the sequence with the highest computed probability score, which is a mathematical concept and a mental process. The limitation of claim 12 describes random sampling from discrete probability distributions, which is a mathematical concept. The limitation of claim 13 narrows the abstract ideas of claim 1 by specifying how the change of the function is calculated, which is a mathematical concept. The limitations of claim 14 describe iterative mathematical optimization in an embedding space using feedback from a probabilistic intermediate, which is a mathematical concept and a mental process when a person having ordinary skill in the art could mentally perform the same conceptual loop of evaluating the candidate sequence’s performance of the desired function before refinement or adjustment.
Claim 88 recites the following limitations, which describe abstract ideas within the mathematical concepts and/or mental processes groupings:
(a) predict the function of a starting point in an embedding at a to a system comprising a supervised model network that predicts the function of a biopolymer sequence and a decoder network, the supervised model network comprising an encoder network providing the embedding of biopolymer sequences in a functional space representing the function and the decoder network trained to provide a predicted probabilistic biopolymer sequence, given an embedding of the predicted biopolymer sequence in the functional space;
(b) calculate a change in the function in relation to the embedding at the starting point according to a step size, thereby enabling providing a first updated point in the functional space;
(c) calculate, at the decoder network, a first intermediate probabilistic biopolymer sequence based on the first updated point in the functional space;
(d) predict the function of the first intermediate probabilistic biopolymer sequence, at the supervised model based on the first intermediate biopolymer sequence;
(e) calculate the change in the function with regard to the embedding at the first updated point in the functional space to provide an updated point in the functional space;
(f) calculate an additional intermediate probabilistic biopolymer sequence at the decoder network based on the updated point in the functional space;
(g) predict the function of the additional intermediate probabilistic biopolymer sequence, at the supervised model, based on the additional intermediate probabilistic biopolymer sequence;
(h) calculate the change in the function with regard to the embedding at the further first updated point in the functional space to provide a yet further updated point in the functional space, optionally iterating steps (g)-(i), where a yet further updated point in the functional space referenced in step (i) is regarded as the further updated point in the functional space in step (g); and
(i) upon approaching a desired level of the function in the functional space, provide the point in the embedding to the decoder network; and obtaining a probabilistic improved biopolymer sequence from the decoder.
The limitation of predict the function of a starting point describes computing a predicted scalar value from a numerical embedding vector, which constitutes an abstract idea within the mathematical concepts and mental processes groupings. The limitations of calculate a change in the function, (a), (e), and (h), describe gradient computation or directional update in embedding space according to a step size, which constitutes an abstract idea within the mathematical concepts or mental processes groupings when a person could mentally evaluate how a change would affect function. The limitations of calculate an intermediate probabilistic sequence, (c) and (f), describe decoding a numerical embedding into a full probability distribution over amino acids, which constitutes an abstract idea within the mathematical concepts and mental processes groupings when the mathematical reconstruction can be performed mentally or with the aid of pen and paper. The limitations of predict the function, (d) and (g), describe re-encoding or evaluating the intermediate sequence to obtain a new predicted function value, which constitutes an abstract idea within the mathematical concepts or mental processes groupings when a person could mentally assess how well the candidate sequence actually performs the desired function. The limitation of upon approaching a desired level of function describes a mathematical conditional check and final decoding to a probability distribution, which constitutes an abstract idea within the mathematical concepts or mental processes groupings when a person could mentally evaluate the sufficiency of the function and finalize the sequence design.
Claims 118-119 and 121 recite the following limitations, which describe abstract ideas within the mathematical concepts grouping:
Claim 118 recites:
a method for training a supervised model for use in the method of claim 1;
(b) mapping, using the encoder, each training biopolymer sequence to a representation in the embedding functional space;
(c) predicting, using the supervised model, based on these representations, the function of each training biopolymer sequence;
(d) determining, using a predetermined prediction loss function, for each training biopolymer sequence, how well the predicted function is in agreement with the function as per the label of the respective training biopolymer sequence; and
(e) optimizing parameters that characterize the behavior of the supervised model with the goal of improving the rating by said prediction loss function that results when further training biopolymer sequences are processed by the supervised model.
Claim 119 recites:
a method for training a decoder for use in a method or system according to claim 1;
(b) mapping, using the decoder, each representation to a probabilistic biopolymer sequence;
(c) drawing a sample biopolymer sequence from each probabilistic biopolymer sequence;
(d) mapping, using a trained encoder, this sample biopolymer sequence to a representation in said embedding functional space;
(e) determining, using a predetermined reconstruction loss function, how well each so-determined representation is in agreement with the corresponding original representation; and
(f) optimizing parameters that characterize the behavior of the decoder with the goal of improving the rating by said reconstruction loss function that results when further representations of biopolymer sequences from said embedding functional space are processed by the decoder.
Claim 121 recites:
a method for training an ensemble of a supervised model and a decoder;
(b) mapping, using the encoder, each training biopolymer sequence to a representation in the embedding functional space;
(c) predicting, using the supervised model, based on these representations, the function of each training biopolymer sequence;
(d) mapping, using the decoder, each representation in the embedding functional space to a probabilistic biopolymer sequence;
(e) drawing a sample biopolymer sequence from the probabilistic biopolymer sequence;
(f) determining, using a predetermined prediction loss function, for each training biopolymer sequence, how well the predicted function is in agreement with the function as per the label of the respective training biopolymer sequence;
(g) determining, using a predetermined reconstruction loss function, for each sample biopolymer sequence, how well it is in agreement with the original training biopolymer sequence from which it was produced;
(h) optimizing parameters that characterize the behavior of the supervised model and parameters that characterize the behavior of the decoder with the goal of improving the rating by a predetermined combination of the prediction loss function and the reconstruction loss function.
The limitations of training a model/decoder involves optimization via linear algebra, calculus, and statistics, which constitutes an abstract idea within the mathematical concepts grouping. The limitations of mapping sequences describe encoding a sequence into a numerical vector in a latent/functional space, which constitutes an abstract idea within the mathematical concepts grouping. The limitations of predicting the function describe computing a predicted output from the embedding vector, which constitutes an abstract idea within the mathematical concepts grouping. The limitations of determining agreement describe a mathematical comparison/measurement of error between predicted and ground truth, which constitutes an abstract idea within the mathematical concepts grouping. The limitations of optimizing parameters describe parameter optimization via minimization of a loss function, which relies on multivariable calculus and numerical optimization theory, constituting an abstract idea within the mathematical concepts grouping. The limitations of drawing a sample describe random sampling from discrete probability distributions, which is a mathematical concept. Claims 2, 9, 17, 20, and 90 do not recite or narrow any judicial exceptions, but inherit the exceptions from the claim upon which they depend.
Therefore, claims 1-14, 17, 20, 88, 90, 118-119, and 121 recite abstract ideas – namely mathematical concepts and mental processes [Step 2A, Prong One: Yes].
Prong Two
Claims 1-14, 17, 20, 88, 90, 118-119, and 121 as a whole do not integrate the recited judicial exception into a practical application. A claim that recites a judicial exception [Prong One] is deemed to be directed to a judicial exception [Step 2A] unless the claim as a whole contains additional elements that integrate the exception into a practical application [Prong Two]. A claim that integrates a judicial exception into a practical application will apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, beyond generally linking the use of the judicial exception to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception. See MPEP §§ 2106.04(d) and 2106.05(e). A claim does not integrate a judicial exception into a practical application by reciting insignificant extra-solution activity, generally linking the exception to a particular technological environment or field of use, merely reciting to apply the exception, merely including instructions to implement an abstract idea on a computer, or merely using a computer as a tool to perform an abstract idea. See MPEP § 2106.04(d)(I). Insignificant extra-solution activities are nominal or tangential additions to a claim that are incidental to the primary process or product, including both pre-solution and post-solution activity (e.g. pre-solution data gathering for use in a process). If integrated into a practical application, the claim is eligible; otherwise, it is directed to the judicial exception, necessitating further analysis at Step 2B.
Claim 1 recites (a) providing a starting point in an embedding to a system comprising a supervised model that predicts the function of a biopolymer sequence and a decoder network, the supervised model network comprising an encoder network providing the embedding of biopolymer sequences in a functional space representing the function, and the decoder network trained to provide a probabilistic biopolymer sequence, given an embedding of a biopolymer sequence in the functional space;
Claim 88 recites a system comprising a processor and non-transitory computer readable medium comprising instructions;
Claim 90 recites a method of making a biopolymer comprising synthesizing an improved biopolymer sequence obtainable by the method of claim 1;
Claim 118 recites wherein this supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations; and (a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function;
Claim 119 recites wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence; and (a) providing a plurality of representations of biopolymer sequences in the embedding functional space;
Claim 121 recites wherein the supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations, wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence; and (a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function;
The limitation of claim 1 reciting providing a starting point describes giving an input to a supervised model that uses an encoder network and decoder network to predict function. This additional element does not recite a particular machine or transformation, nor does it recite some technological improvement sufficient to integrate the judicial exceptions into a practical application. Providing a starting point constitutes insignificant extra-solution activity, while the supervised model is a generic computer component used to implement the abstract mathematical optimization and merely apply it to the field of biopolymers. See MPEP §§ 2106.05(f)-(h). This additional element is narrowed by the limitations of claims 2, 9, 17, and 20 because claim 2 specifies that the starting point is the embedding of a seed biopolymer sequence, claim 9 specifies that two or more starting points are used concurrently, claim 17 specifies that the biopolymer is a protein, and claim 20 specifies that the encoder is trained using a dataset of at least 20 sequences.
The limitation of claim 88 reciting a processor and non-transitory computer readable medium describes generic computer components that amount to nothing more than mere instructions to apply the judicial exceptions, which do not integrate into a practical application. See MPEP §§ 2106.05(b) and (f). The limitation of claim 90 is recited at such a high level of generality that it is equivalent to a recitation of the words “apply it,” which does not integrate the exceptions into a practical application. See MPEP § 2106.05(f). The additional elements of claims 118-119 and 121 recite field of use limitations (i.e. training for a specific purpose) and pre-solution data gathering activity that does not transform the nature of the claim into a patent-eligible application of the judicial exception. See MPEP §§ 2106.04(f)-(h).
Finally, claims 3-8 and 10-14 do not include any additional elements.
The claims as a whole merely recite insignificant extra-solution activities and abstract ideas implemented on generic computer components without meaningful limitations that tie it to a specific technological improvement. Therefore, claims 1-14, 17, 20, 88, 90, 118-119, and 121 do not contain additional elements that integrate the recited abstract ideas into a practical application [Step 2A, Prong Two: No].
Step 2B
Claims 1-14, 17, 20, 88, 90, 118-119, and 121 do not include additional elements, whether considered individually or in combination, that are sufficient to amount to significantly more than the judicial exception itself. Under Step 2B, the claim is analyzed to determine whether there are any additional elements that, individually or in combination, constitute an “inventive concept" sufficient to ensure that the claim, as a whole, amounts to significantly more than the judicial exception itself. See MPEP § 2106.05; and Alice Corp. Pty. Ltd. v. CLS Bank Int'l, 573 U.S. 208, 217-18, 110 USPQ2d 1976, 1981 (2014).
Claim 1 recites (a) providing a starting point in an embedding to a system comprising a supervised model that predicts the function of a biopolymer sequence and a decoder network, the supervised model network comprising an encoder network providing the embedding of biopolymer sequences in a functional space representing the function, and the decoder network trained to provide a probabilistic biopolymer sequence, given an embedding of a biopolymer sequence in the functional space;
Claim 88 recites a system comprising a processor and non-transitory computer readable medium comprising instructions;
Claim 90 recites a method of making a biopolymer comprising synthesizing an improved biopolymer sequence obtainable by the method of claim 1;
Claim 118 recites wherein this supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations; and (a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function;
Claim 119 recites wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence; and (a) providing a plurality of representations of biopolymer sequences in the embedding functional space;
Claim 121 recites wherein the supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations, wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence; and (a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function;
The limitation of claim 1 reciting providing a starting point constitutes conventional insignificant extra-solution activity with conventional computer components used to implement the abstract mathematical optimization and merely apply it to the field of biopolymers. See MPEP §§ 2106.05(f)-(h); Yutaka Saito et al., Machine-Learning-Guided Mutagenesis for Directed Evolution of Fluorescent Proteins, 7(9) ACS Synth. Biol. 2014, 2015 col.1 para.2 (13 August 2018); and Diederik P. Kingma and Max Welling, Auto-Encoding Variational Bayes, arXiv 1312.6114, Abstract (20 December 2013). The limitations of claims 2, 9, 17, and 20 are well-understood, routine, and conventional in the field of optimization via machine learning, and do not provide an inventive concept sufficient to amount to significantly more than the judicial exception itself. See MPEP § 2106.05(d); Rafael Gómez-Bombarelli et al., Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules, 4(2) ACS Cent. Sci. 268, 272 col.2 para.1 (12 January 2018) (embedding seed molecules and training on dataset of 108,000 molecules) (IDS document); and Yutaka Saito et al., at 2016 (embeds 155 seed proteins for training model).
The limitation of claim 88 reciting a processor and non-transitory computer readable medium describes generic computer components that amount to nothing more than mere instructions to apply the judicial exceptions, which do not add significantly more than the exceptions themselves. See MPEP §§ 2106.05(b) and (f); and C3.ai., Infrastructure: Machine Learning Hardware Requirements, §§ Processors: CPUs, GPUs, TPUs, and FPGAs – Memory and Storage (15 May 2021). The limitation of claim 90 is recited at such a high level of generality that it is equivalent to a recitation of the words “apply it,” which does add significantly more than the judicial exceptions themselves. See MPEP § 2106.05(f); and Ehecatl Antonio del Rio-Chanona et al., Sustainable biopolymer synthesis via superstructure and multiobjective optimization, 64(1) AIChE J. 91, abstract (20 July 2017) (synthesizing biopolymer based on optimization framework is well-understood, routine, and conventional). The additional elements of claims 118-119 and 121 recite conventional field of use limitations (i.e. training for a specific purpose) and conventional pre-solution data gathering activity that does not transform the nature of the claim into a patent-eligible application of the judicial exception. See MPEP §§ 2106.04(f)-(h); Yutaka Saito et al., at 2015 col.1 para.2; and Diederik P. Kingma and Max Welling, at 2 para.6.
Overall, claims 1-14, 17, 20, 88, 90, 118-119, and 121 amount to no more than insignificant extra-solution activities and implementing the abstract ideas on conventional computers in a routine way. The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception itself because the claims recite additional elements that equate to conventional insignificant extra-solution activity and mere instructions to apply the recited abstract ideas in a generic way or in a generic computing environment. Therefore, claims 1-14, 17, 20, 88, 90, 118-119, and 121 are rejected for failing to set forth patent eligible subject matter under 35 U.S.C. 101 because the claimed invention recites abstract ideas [Step 2A, Prong One: Yes] and the additional elements do not integrate the judicial exception into a practical application [Step 2A, Prong Two: No] and do not amount to claiming significantly more than the recited exception [Step 2B: No].
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-2, 5, 9-13, 17, 20, and 90 are rejected under 35 U.S.C. 103 as being unpatentable over Sam Sinai et al., Variational auto-encoding of protein sequences, arXiv 1712.03346 (3 January 2018) (hereinafter “Extended abstract”), Sam Sinai, Using a Variational Auto-encoder to predict protein function, (14 August 2017) (hereinafter “Blog”), and Sam Sinai and Eric Kelsic, Using a Variational Auto-encoder to predict protein fitness from evolutionary data, GitHub (20 July 2017) (hereinafter “GitHub”). The italicized text within parenthesis corresponds to the instant claim limitations.
Regarding claim 1, Sinai discloses a model that can provide protein sequences that can function similarly or better than an initial protein sequence. Extended abstract, at 2 para.2 (a method of engineering an improved biopolymer sequence as assessed by a function). Sinai discloses that the model predicts the function of a sequence using an encoder to embed the sequence as a representation in latent space and a decoder trained to predict a probabilistic sequence given the sequence representation. Id. at 2 para.3; 3 paras. 1, 3, & 6; 4 para.2 (a system comprising a supervised model that predicts the function of a biopolymer sequence and a decoder network, the supervised model network comprising an encoder network providing the embedding of biopolymer sequences in a functional space representing the function, and the decoder network trained to provide a probabilistic biopolymer sequence, given an embedding of a biopolymer sequence in the functional space). Sinai discloses that sequences are used as input to be embedded as one-hot encodings. Id. at 3 para.3 (providing a starting point in an embedding to a system). Sinai discloses that the model learns parameters for the distributions through gradient descent. Id. at 2 para.8. Mutant sequences with a higher fitness/probability than the wildtype have an improved/desired level of function. Id. at 3 para.7 (upon reaching a desired level of the function). Sinai teaches that to predict the fitness of a mutant (updated point), the one-hot vector for the sequence is fed into the model, which reconstructs the sequence and generates a probability weight matrix. Id. at 3 para.4 (providing the first updated point). Sinai discloses traversing the latent space in the neighborhood of the sequence of interest to obtain the sequence with the highest probability value. Id. at 2 para.4 (obtaining a probabilistic improved biopolymer sequence from the decoder).
While Sinai’s extended abstract does not explicitly disclose calculating a change in the function and reaching a level of the function within a particular threshold, Sinai notes that the code to reproduce the analysis is publicly available. Id. at 5 para.4. Additionally, Sinai discloses that the code notebook is associated with a blog post that highlights general results from the workflow. GitHub, para.1. Within the notebook and blog, Sinai discloses that for mutant sequences, which represent discrete steps away from the starting point in latent space, the model calculates a fitness value, which is the probability that a protein will have the target functionality or relative performance, and compares it to the wildtype sequence. GitHub, § 4.1 Single mutants paras.1 & 3; Blog, § Results para.2 (calculating a change in the function in relation to the embedding at the starting point according to a step size, the calculated change enabling providing a first updated point in the functional space). Additionally, Sinai discloses reweighting the sequences based on a similarity threshold to feed the network more informative samples by discounting sequences that are very close to each other. GitHub, para.1 after In [9] (within a particular threshold at the first updated point in the functional space).
A person having ordinary skill in the art would be motivated to combine the disclosures of the extended abstract, code notebook, and blog post because Sinai notes that the disclosures relate to the same workflow. One of ordinary skill in the art would reasonably expect success in this combination because the code notebook reproduces the analysis in the extended abstract, with general results reported in the blog post. 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 is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007); and MPEP § 2143, G.
Regarding claim 2, Sinai teaches starting with a reference/wildtype protein sequence embedding. GitHub, § 3. Exploring the Latent Space para.2 after In [43] (the method of Claim 1, wherein the starting point is the embedding of a seed biopolymer sequence).
Regarding claim 5, Sinai discloses that the protein is embedded into a continuous representation in latent space having a gradient. Extended abstract, at 4 para.2; 2 para.8 (the method of Claim 1, wherein the embedding is a continuously differentiable functional space representing the function and having one or more gradients).
Regarding claim 9, Sinai discloses superimposing one hot embeddings (input) from two sequences, one wildtype and one mutant. Blog, § Results paras.6-7 (the method of Claim 1, wherein two or more starting points in the embedding are used concurrently).
Regarding claim 10, Sinai discloses that the model learns parameters for the probability distribution through gradient descent, and the decoder outputs a full probability weight matrix over amino acids. Extended abstract, at 2 para.8; 3 para.4; Blog, § Results para.8. Sinai teaches autoregressive-style sampling where probabilities at later positions can depend on previous choices such that an update in probability occurs not only in the position where two sequences differ, but also in places that they are the same. Blog, § Results paras.3 & 9-10 (the method of Claim 1, wherein correlations between residues in a probabilistic sequence comprising a probability distribution of residue identities are considered in a sampling process using conditional probabilities that account for the portion of the sequence that has already been generated).
Regarding claim 11, Sinai discloses that an approximation for the maximum likelihood of the data is obtained by maximizing the lower bound on the evidence through gradient ascent, where the output is a reconstruction sequence in the form of a probability weight matrix with each position column constituting probabilities for each amino-acid at that location. Extended abstract, at 3 para.1; Blog, § Results para.8 (the method of Claim 1, further comprising selecting the maximum likelihood improved biopolymer sequence from a probabilistic biopolymer sequence comprising a probability distribution of residue identities).
Regarding claim 12, Sinai discloses that the decoder produces a probability weight matrix containing the probability for each amino acid independently, which is used to select the amino acid with the highest fitness. Blog, § Results paras.8 & 11 (the method of Claim 1, comprising sampling the marginal distribution at each residue of a probabilistic biopolymer sequence comprising a probability distribution of residue identities).
Regarding claim 13, Sinai discloses a variational autoencoder (VAE) implemented with the Evidence Lower Bound (ELBO) loss function to learn parameters for the distributions through gradient descent. Extended abstract, at 2 paras.7-8 (the method of Claim 1, wherein the change of the function with regard to the embedding, is calculated by calculating the change of the function with regard to the encoder, then the change of the encoder to the change of the decoder, and the change of the decoder with regard to the embedding).
Regarding claim 17, Sinai discloses a model that embeds and provides protein sequences. Extended abstract, Abstract (the method of Claim 1, wherein the biopolymer is a protein).
Regarding claim 20, Sinai discloses training the model on protein sequences with a batch size of 20. Blog, § Results para.3; GitHub, § 2. Training the model, In [27] (the method of Claim 1, wherein the encoder is trained using a training data set of at least 20 biopolymer sequences).
Regarding claim 90, Sinai discloses traversing the latent space in the neighborhood of the sequence of interest to obtain the sequence with the highest probability value. Blog, § Under the hood of the VAE paras.5-6; Extended abstract, at 2 para.4. While Sinai does not explicitly disclose making the improved protein, Sinai notes that the model was developed to design novel proteins that perform a particular function. Extended abstract, at 1 para.1. A person having ordinary skill in the art would understand that in searching for a protein with a particular function, the sequence with the highest probability value should be synthesized and deployed to perform the function. One of ordinary skill in the art would reasonably expect success in this because the sequence with the highest probability value obtained via Sinai’s model has the greatest chance of performing the target function (a method of making a biopolymer comprising synthesizing an improved biopolymer sequence obtainable by the method of Claim 1). 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 is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007); and MPEP § 2143, G.
Claims 3-4, 6-8, 14, 88, 118-119, and 121-122 are rejected under 35 U.S.C. 103 as being unpatentable over Sinai Extended abstract, Blog, and GitHub as applied to claims 1-2, 5, 9-13, 17, 20, and 90 above, and further in view of IDS document Rafael Gómez-Bombarelli et al., Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules, 4(2) ACS Cent. Sci. 268–76 (12 January 2018) (hereinafter “Gómez-Bombarelli”), Jennifer Wei, chemvae/hyperparameters.py, GitHub (15 January 2018) (hereinafter “Wei”), and beangoben, chemvae/train_vae.py, GitHub (18 January 2018) (hereinafter “beangoben”), as evidenced by C3.ai. (Infrastructure: Machine Learning Hardware Requirements (15 May 2021)) and Jason Brownlee, What Is a Gradient in Machine Learning?, Machine Learning Mastery (12 October 2021) (hereinafter “Brownlee”). The italicized text within parenthesis corresponds to the instant claim limitations.
Regarding claim 3, Sinai discloses a model that can provide protein sequences that can function similarly or better than an initial protein sequence. Extended abstract, at 2 para.2. Sinai notes that the model is similar to Gómez-Bombarelli’s model for chemical molecules, and suggests that the continuous representation of the protein may similarly be used together with gradient-based optimization to achieve a desirable property. Extended abstract, at 4 para.2. Sinai fails to disclose calculating a second change in the function with regard to the embedding at the first updated point in the functional space, and iterating the process of calculating the second change in the function with regard to the embedding at a further updated point.
However, Gómez-Bombarelli discloses an autoencoder for molecular design with a model, f(z), for property prediction. At 270 col.2 para.2; Figure 1 caption. Gómez-Bombarelli teaches training the model to predict the properties of molecules based on their latent representation z before performing gradient-based optimization of f(z) with respect to z to find new latent representations expected to have high values of desired properties. Figure 1 caption. Gómez-Bombarelli discloses that gradient-based optimization involves computing a gradient where molecules are organized by property value. Figure 1 caption, 272 col.2 paras.2-3. Gómez-Bombarelli teaches that a prominent molecule can be decoded from a latent point, and the resulting sequence is re-encoded into the latent space for continued optimization. At 272 col.1 para.1; 273 col.2 para.1. Gómez-Bombarelli continues the gradient-based optimization until a molecule with the desired properties is achieved. At 273 col.2 para.2; Figure 4 caption.
A person having ordinary skill in the art would be motivated to combine the disclosure of Sinai with the teachings of Gómez-Bombarelli because Sinai suggests that the continuous representation of the protein may be used together with gradient-based optimization to achieve a desirable property. One of ordinary skill in the art would reasonably expect success in this combination because Gómez-Bombarelli supplies the gradient-based method for exploring chemical space based on continuous encodings of molecules, which is applicable to proteins when Sinai’s model effectively encodes protein sequences into a continuous latent space that can be traversed. Therefore, one of ordinary skill in the art would find it obvious to apply gradient-based optimization to Sinai’s protein model to generate sequences with improved function. 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 is likely to be obvious. See KSR International Co. v. Teleflex Inc., 550 U.S. 398, 415-421, USPQ2d 1385, 1395 – 97 (2007); and MPEP § 2143, G.
Regarding claim 4, Gómez-Bombarelli discloses that for a given latent point, a predominant molecule is decoded, along with many slight variations with lower property values, and the predominant molecular sequence is re-encoded. At 272 col.1 para.1 (wherein providing the first updated point can be performed upon reaching a desired level of the function). Gómez-Bombarelli iterates gradient-based optimization until a molecule with the desired properties is achieved. At 273 col.2 para.2; Figure 4 caption (providing the further updated point includes providing the iterated further updated point to the decoder network). Sinai discloses reweighting the sequences based on a similarity threshold to feed the network more informative samples by discounting sequences that are very close to each other. GitHub, para.1 after In [9] (within a particular threshold at the optionally iterated further updated point).
Regarding claim 6, Gómez-Bombarelli computes the gradient of the property prediction model with respect to the latent embedding z. At 273 col.1 para.1; Figure 1(b) caption. A gradient is a derivative of a function, as evidenced by Brownlee. § What Is a Gradient? para.1 (the method of Claim 1, wherein calculating the change of the function with regard to the embedding comprises taking a derivative of the function with regard to the embedding).
Regarding claim 7, Gómez-Bombarelli trains a model that can predict multiple properties and optimize a combined objective that captures and balances multiple desired traits. At 273 col.2 paras.2-3; 274 col.1 para.6 – col.2 para.1 (the method of Claim 1, wherein the function is a composite function of two or more component functions).
Regarding claim 8, Gómez-Bombarelli discloses optimizing the 5×QED−SAS objective, which is a weighted sum, where QED is the Quantitative Estimation of Drug-likeness and SAS is the Synthetic Accessibility score. At 273 col.2 para.3 (the method of Claim 7, wherein the composite function is a weighted sum of the two or more composite functions).
Regarding claim 14, Gómez-Bombarelli teaches inputting an initial sequence to be embedded into the autoencoder, obtaining an updated latent point, decoding the updated point into a probabilistic sequence, re-encoding the sequence, and iterating the process until a sequence with the target properties is achieved. At 272 col.1 para.1; Figure 4 caption (the method of Claim 1, the method comprising: providing the first updated point in the functional space or further updated point in the functional space to the decoder network to provide an intermediate probabilistic biopolymer sequence, providing the intermediate probabilistic biopolymer sequence to the supervised model network to predict the function of the intermediate probabilistic biopolymer sequence). Gómez-Bombarelli discloses using gradient-based optimization of f(z) with respect to the embedding z to optimize the sequence, which involves computing a gradient where molecules are organized by property value. Figure 1 caption, 272 col.2 paras.2-3. (calculating the change in the function with regard to the embedding for the intermediate probabilistic biopolymer to provide a further updated point in the functional space).
Regarding claim 88, Sinai discloses a model that predicts the function of a sequence using an encoder to embed the sequence as a representation in latent space and a decoder trained to predict a probabilistic sequence given the sequence representation. Extended abstract, at 2 para.3; 3 paras. 1, 3, & 6; 4 para.2 ((a) predict the function of a starting point in an embedding at a to a system comprising a supervised model network that predicts the function of a biopolymer sequence and a decoder network, the supervised model network comprising an encoder network providing the embedding of biopolymer sequences in a functional space representing the function and the decoder network trained to provide a predicted probabilistic biopolymer sequence, given an embedding of the predicted biopolymer sequence in the functional space). Sinai suggests that the continuous representation of the protein may be used together with gradient-based optimization to achieve a desirable property, similar to Gómez-Bombarelli’s method for chemical molecules. Extended abstract, at 4 para.2. Gómez-Bombarelli teaches training the model to predict the properties of molecules based on their latent representation z before performing gradient-based optimization of f(z) with respect to z using a constant step size to find new latent representations expected to have high values of desired properties. Figures 1 & 4 caption ((b) calculate a change in the function in relation to the embedding at the starting point according to a step size, thereby enabling providing a first updated point in the functional space).
Gómez-Bombarelli teaches that a prominent molecule can be decoded from a latent point, with the last layer of the decoder defining a probability distribution over all possible characters at each position in the sequence. At 272 col.1 para.1; 274 col.2 para.5 ((c) calculate, at the decoder network, a first intermediate probabilistic biopolymer sequence based on the first updated point in the functional space). Gómez-Bombarelli discloses that the resulting sequence is re-encoded into the latent space for property prediction and continued optimization by moving in the direction most likely to improve the desired attributes using a consistent step size. At 270 col.2 para.2; 272 col.1 para.1; 273 col.2 para.1 Figure 4 caption ((d) predict the function of the first intermediate probabilistic biopolymer sequence, at the supervised model based on the first intermediate biopolymer sequence). Gómez-Bombarelli iterates the gradient-based optimization, calculating the change in property to obtain an updated latent point that is decoded into a probabilistic sequence and re-embedded for property prediction. At 272 col.1 para.1; 273 col.2 para.1; Figure 4 caption ((e) calculate the change in the function with regard to the embedding at the first updated point in the functional space to provide an updated point in the functional space; (f) calculate an additional intermediate probabilistic biopolymer sequence at the decoder network based on the updated point in the functional space; (g) predict the function of the additional intermediate probabilistic biopolymer sequence, at the supervised model, based on the additional intermediate probabilistic biopolymer sequence; (h) calculate the change in the function with regard to the embedding at the further first updated point in the functional space to provide a yet further updated point in the functional space, optionally iterating steps (g)-(i), where a yet further updated point in the functional space referenced in step (i) is regarded as the further updated point in the functional space in step (g)). Gómez-Bombarelli continues the gradient-based optimization until a molecule with the desired properties is obtained. At 273 col.2 para.2; Figure 4 caption ((i) upon approaching a desired level of the function in the functional space, provide the point in the embedding to the decoder network; and obtaining a probabilistic improved biopolymer sequence from the decoder).
While neither Sinai nor Gómez-Bombarelli explicitly disclose a system comprising a processor and non-transitory computer readable medium comprising instructions, Sinai and Gómez-Bombarelli disclose machine learning models, which necessarily involves a processor and a non-transitory computer readable medium comprising instructions. See C3.ai., §§ Processors: CPUs, GPUs, TPUs, and FPGAs – Memory and Storage.
Regarding claim 118, Gómez-Bombarelli trains an autoencoder for molecular design with an encoder to convert sequences into a fixed-dimensional vector and a model, f(z), for property prediction from the latent vectors. At 269 col.2 para.3; 270 col.2 para.2; Figure 1 caption (a method for training a supervised model for use in the method of Claim 1, wherein this supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations). Gómez-Bombarelli trains the property prediction model on a set of labeled examples. At 269 col.1 para.2 ((a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function). Gómez-Bombarelli discloses that the encoder is jointly trained such that the encoder maps the training sequences to the latent space representation. At 270 col.2 para.2 ((b) mapping, using the encoder, each training biopolymer sequence to a representation in the embedding functional space). Gómez-Bombarelli uses the property prediction model to predict the property from the latent vector of the encoded sequence. Id. ((c) predicting, using the supervised model, based on these representations, the function of each training biopolymer sequence). Gómez-Bombarelli discloses using a loss function to determine the accuracy of the property prediction model. At 269 col.2 para.2 ((d) determining, using a predetermined prediction loss function, for each training biopolymer sequence, how well the predicted function is in agreement with the function as per the label of the respective training biopolymer sequence). Gómez-Bombarelli teaches that the property prediction loss was annealed according to sigmoid schedule after 29 epochs, running for a total 120 epochs. At 274 col.2 para.5; 275 col.1 para.1 ((e) optimizing parameters that characterize the behavior of the supervised model with the goal of improving the rating by said prediction loss function that results when further training biopolymer sequences are processed by the supervised model).
Regarding claim 119, Gómez-Bombarelli trains an autoencoder for molecular design with a decoder to convert latent vector representations back into probabilistic sequences. At 269 col.2 para.3; 271 col.2 para.2 (a method for training a decoder for use in a method or system according to Claim 1, wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence). Gómez-Bombarelli discloses jointly training the autoencoder with a plurality of strings, which are embedded as latent vectors. At 270 col.2 para.2 – 271 col.1 para.1 ((a) providing a plurality of representations of biopolymer sequences in the embedding functional space). Gómez-Bombarelli teaches that the last layer of the decoder defines a probability distribution, and the decoder model samples a string from the probability distribution over characters in each position generated by its final layer. At 271 col.2 para.2; 274 col.2 para.5 ((b) mapping, using the decoder, each representation to a probabilistic biopolymer sequence; (c) drawing a sample biopolymer sequence from each probabilistic biopolymer sequence). Gómez-Bombarelli discloses that the probabilistic output is re-encoded into a latent point. Figure 4 caption; 272 col.1 para.1 ((d) mapping, using a trained encoder, this sample biopolymer sequence to a representation in said embedding functional space). Gómez-Bombarelli teaches using a decoder reconstruction loss to assess the accuracy of the decoder. At 269 col.2 para.2; Table 2 caption ((e) determining, using a predetermined reconstruction loss function, how well each so-determined representation is in agreement with the corresponding original representation). Gómez-Bombarelli optimizes hyperparameters of the model to anneal the decoder loss. At 271 col.1 para.1; 274 col.2 para.5; see also Wei, line 62; and beangoben, lines 254-61 ((f) optimizing parameters that characterize the behavior of the decoder with the goal of improving the rating by said reconstruction loss function that results when further representations of biopolymer sequences from said embedding functional space are processed by the decoder).
Regarding claim 121, Gómez-Bombarelli trains an autoencoder for molecular design with an encoder to convert sequences into a fixed-dimensional vector, a model, f(z), for property prediction from the latent vectors, and a decoder to convert latent vector representations back into probabilistic sequences. At 269 col.2 para.3; 270 col.2 para.2; 271 col.2 para.2; Figure 1 caption ( a method for training an ensemble of a supervised model and a decoder, wherein the supervised model comprises an encoder network that is configured to map biopolymer sequences to representations in an embedding functional space, wherein the supervised model is configured to predict a function of the biopolymer sequence based on the representations, wherein the decoder is configured to map a representation of a biopolymer sequence from an embedding functional space to a probabilistic biopolymer sequence). Gómez-Bombarelli trains the property prediction model on a set of labeled examples. At 269 col.1 para.2 ((a) providing a plurality of training biopolymer sequences, wherein each training biopolymer sequence is labelled with a function). Gómez-Bombarelli discloses that the encoder is jointly trained such that the encoder maps the training sequences to the latent space representation. At 270 col.2 para.2 ((b) mapping, using the encoder, each training biopolymer sequence to a representation in the embedding functional space). Gómez-Bombarelli uses the property prediction model to predict the property from the latent vector of the encoded sequence. Id. ((c) predicting, using the supervised model, based on these representations, the function of each training biopolymer sequence).
Gómez-Bombarelli teaches that the last layer of the decoder defines a probability distribution, and the decoder model samples a string from the probability distribution over characters in each position generated by its final layer. At 271 col.2 para.2; 274 col.2 para.5 ((d) mapping, using the decoder, each representation in the embedding functional space to a probabilistic biopolymer sequence; (e) drawing a sample biopolymer sequence from the probabilistic biopolymer sequence). Gómez-Bombarelli discloses using a loss function to determine the accuracy of the property prediction model. At 269 col.2 para.2 ((f) determining, using a predetermined prediction loss function, for each training biopolymer sequence, how well the predicted function is in agreement with the function as per the label of the respective training biopolymer sequence). Gómez-Bombarelli teaches using a decoder reconstruction loss to assess the accuracy of the decoder. At 269 col.2 para.2; Table 2 caption; see also Wei, line 62; and beangoben, lines 254-61 ((g) determining, using a predetermined reconstruction loss function, for each sample biopolymer sequence, how well it is in agreement with the original training biopolymer sequence from which it was produced). Gómez-Bombarelli teaches that the property prediction loss was annealed according to sigmoid schedule after 29 epochs, running for a total 120 epochs, while the hyperparameters are optimized to anneal the decoder loss. At 271 col.1 para.1; 274 col.2 para.5; 275 col.1 para.1 ((h) optimizing parameters that characterize the behavior of the supervised model and parameters that characterize the behavior of the decoder with the goal of improving the rating by a predetermined combination of the prediction loss function and the reconstruction loss function).
Regarding claim 122, Gómez-Bombarelli discloses the set of parameters used in the optimized model. At 274 col.2 paras.4-6 (a set of parameters that characterize the behavior of a supervised model, an encoder or a decoder, obtained by the method of Claim 118).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Emily A Darrigrand whose telephone number is (571) 272-1098. The examiner can normally be reached Mon-Thursday 7:00AM-4:00PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Larry Riggs, can be reached at (571) 270-3062. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/E.A.D./Examiner, Art Unit 1686
/LARRY D RIGGS II/Supervisory Patent Examiner, Art Unit 1686
Prosecution Insights