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 .
Priority
Acknowledgement is made of applicant’s claim for priority based on provisional Application No. 63/196,125, filed on 2 June 2021.
Information Disclosure Statement
Applicant is reminded of the continuing obligation under 37 CFR 1.56 to timely apprise the Office of any information which is material to patentability of the claims under consideration in this application.
Specification
The attempt to incorporate subject matter into this application by reference to various non-patent literature publications throughout the Specification1, are ineffective. These cited references can be found, for example, in Specification, [0004-0006], [0016], [0018-0021], [0026], [0032], and [0035].
Applicant indicates that “These methodologies are described in detail in one or more of the cited references”, all of which are non-patent literature. However, this means that Applicant intended for such references to be regarded as “essential material”.2
However, MPEP § 608.01(I)(A) states that “In any application that is to issue as a U.S. patent, essential material may only be incorporated by reference to a U.S. patent or patent application publication”. Yet the Specification attempts to incorporate by reference non-patent literature, which cannot qualify as “essential material”3, but only as “nonessential material”, e.g., for providing background.4 However, given the sparsity of details concerning how the various methodologies are operating, Applicant appears to be attempting to rely upon the non-patent literature references in order to enable one skilled in the art to make or use the invention, which is improper. Thus, the lack of particular details of the claimed invention has raised enablement issues and indefiniteness issues, as seen in the 112 rejections below.
Claim Objections
Claims 1, 4 and 7 are objected to because of the following informalities: the claims recite “MSA-Transformer” in claims 1 and 7, and “MSA transformer” in claim 4. However, the acronym MSA (which stands for “multiple sequence alignment”) has not been properly defined in the claims for this acronym. Appropriate correction is required.
Claim 4 is objected to because of the following informalities: the claim recites “from the embeddings”. This should be “from the information-dense embeddings” to be consistent with the earlier language in the claim. Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-8 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the enablement requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to enable one skilled in the art to which it pertains, or with which it is most nearly connected, to make and/or use the invention.
Independent Claims 1, 4, and 7 recite an “MSA-Transformer” (or “MSA transformer” in the case of claim 4) that produces enriched individual and pairwise embeddings from the multiple sequence alignments (claims 1 and 7), or produces information-dense embeddings from the protein sequence (claim 4). There is a lack of enablement for such a limitation or producing enriched individual and pairwise embeddings from the multiple sequence alignments (claims 1 and 7) and for producing information-dense embeddings from the protein sequence (claim 4) using the MSA transformer, as the Specification only recites the use of the MSA transformer without stating how the MSA transformer arrives at the disclosed step of producing these embeddings. Thus, one of ordinary skill in the art would not be enabled to make and/or use this invention, which is directed to the result rather than how the disclosure enables such a function.
Independent Claims 1 and 7 further recite “extracting, from the enriched individual and pairwise embeddings, relevant features and structure latent states for use by a downstream graph transformer”. Similar to above, there is a lack of enablement with such a limitation, as it also cannot be ascertained what the form of the enriched individual and pairwise embedding is to take, in relation to the lack of enablement issue raised above.
Independent Claims 1 and 7 further recite “using the downstream graph transformer, operating on node representations through an attention-based mechanism that considers pairwise edge attributes to obtain final node encodings”. Independent Claim 4 recites “using the attention-based graph transformer architecture, processing and structuring geometric information, to obtain final node representations”. Similar to above, there is a lack of enablement for such limitations, as the details of the attention-based mechanism operating on node representations and considers pairwise edge attributes to obtain final node encodings (in claims 1 and 7) and processing and structuring geometric information to obtain final node representations (in claim 4) are not described in sufficient detail within the Specification, and thus lack enablement to one of ordinary skill in the art.
Lastly, independent claims 1 and 7 recite “projecting the final node encodings to form the computer-modeled folded protein structure”. Independent claim 4 recites “projecting the final node representations into Cartesian coordinates through a learnable transformation to obtain the folded protein sequence”. There is a lack of enablement for how the final node encodings/representations are projected, what exactly constitutes a “learnable transformation” in the case of claim 4 to achieve such a result, and how the computer-modeled folded protein structure (in claims 1 and 7) / folded protein sequence (in claim 4) is achieved.
The dependent claims are rejected for at least by virtue of their dependency on their respective independent claims, and for failing to cure the deficiencies of their respective independent claims.
Claims 2 and 8 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the enablement requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to enable one skilled in the art to which it pertains, or with which it is most nearly connected, to make and/or use the invention.
The claims recite “computing an induced distogram of the computer-modeled folded protein structure”. There is a lack of detail within the Specification for such a limitation (in addition to the 112(b) issues), and thus raises enablement issues.
Claim 3 is rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the enablement requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to enable one skilled in the art to which it pertains, or with which it is most nearly connected, to make and/or use the invention.
The claim recites “storing any individual and pairwise embeddings that are from the original protein sequence”. Given that there was no mention of an original protein sequence, or how any individual and pairwise embeddings would have been produced from such an original protein sequence, there is a lack of enablement with such a limitation (e.g., the production of individual and pairwise embeddings from an original protein sequence, or how the original protein sequence would be recognized from (enriched) individual and pairwise embeddings).
Claims 5-6 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the enablement requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to enable one skilled in the art to which it pertains, or with which it is most nearly connected, to make and/or use the invention.
Claim 5 recites “calculating induced distance maps from the projected final node representations”. There is a lack of enablement for such a limitation, as the Specification does not indicate how such a calculation is performed “from the projected final node representations”, as claimed.
Claim 6 is rejected for at least by virtue of its dependency on claim 5, and for failing to cure the deficiencies of claim 5.
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 1-3 and 7-8 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.
Independent Claims 1 and 7 recite “using an MSA-Transformer, producing enriched individual and pairwise embeddings from the multiple sequence alignments”. It is unclear what the metes and bounds of “enriched” are within the context of the claimed invention, as the only mention is that the protein sequence is augmented for obtaining multiple sequence alignments, and the details concerning the MSA-Transformer are sparse. Thus, the metes and bounds of “enriched” cannot be ascertained.
Independent Claims 1 and 7 recite “extracting…relevant features and structure latent states for use by a downstream graph transformer”.
Firstly, it is unclear what is meant by “structure latent states” within the context of the claimed invention, as this is never utilized again in the later steps. The Specification is also sparse on such details.
Secondly, the limitation “for use by a downstream graph transformer” is unclear, as the features and structure latent states do not show up again within the claimed language. Therefore, it is unclear how the “relevant features and structure latent states” which were extracted, were utilized by the downstream graph transformer (which is later claimed within the context of operating on node representations, but no mention of “relevant features and structure latent states”).
Therefore, the metes and bounds of such limitations cannot be ascertained within the context of the claimed invention.
The independent claims further recite “using the downstream graph transformer, operating on node representations through an attention-based mechanism that considers pairwise edge attributes…”). It is unclear what is meant by “considering” pairwise edge attributes, e.g., is it based on, partially based on, indirectly utilized in various iterations of equations, etc.? Therefore, the metes and bounds of such a limitation cannot be ascertained within the context of the claimed invention.
The independent claims further recite “projecting the final node encodings to form the computer-modeled folded protein structure”. It is unclear what is meant by “to form” a “computer-modeled folded protein structure”, e.g., is this simply a numerical output representation of the folded protein structure (such as the Cartesian coordinates that are claimed in independent claim 4), is this a projection into 3-dimensional space such as for graphical rendering, etc.? Thus, the metes and bounds of such a limitation cannot be ascertained within the context of the claimed invention, e.g., whether it is the same or different from, e.g., the Cartesian coordinates seen in claim 4.
The dependent claims are rejected for at least by virtue of their dependency on their respective independent claims, and for failing to cure the deficiencies of their respective independent claims.
Claims 2, 5-6, and 8 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.
Claims 2 and 8 recite “induced distogram”. Taking the broadest reasonable interpretation, a “distogram” appears to be some sort of visual display (e.g., a histogram of distance values). However, Specification, [0028] states that “To train the network, a distogram-based loss function is used on the resulting distance map”. It is unclear what a “distogram” is within the context of the claimed invention, as the distogram does not appear to be a visual display, but rather some type of series of values that are not displayed, such as a matrix. Thus, the metes and bounds of “distogram” cannot be ascertained within the context of the claimed invention.
Furthermore, Claims 5-6 recite “induced distance map”. As stated previously, Specification, [0028] states that there is some sort of relationship between the distogram-based loss function and the distance map. However, it is unclear what a “distance map” is within the context of the claimed invention, as it is claimed in a similar manner as the “distogram” claimed in claims 2 and 8, yet appears to be indicated as a separate element in Specification, [0028].
Additionally, it is unclear what is meant by “induced” within the context of the claimed invention.
For purposes of interpretation, “distogram” and “distance map” have been taken to be interchangeable, e.g., in secondary reference Rosenbluth, “distogram” and “distance matrix” are interchangeable. Additionally, “induced” has been taken to mean some sort of constraint on the distance.
Claim 3 is 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 claim recites “storing any individual and pairwise embeddings that are from the original protein sequence”.
Firstly, there is a lack of antecedent basis issue with both “any” individual and pairwise embeddings (e.g., is this the same as the “enriched” individual and pairwise embeddings of claim 1?).
Secondly, independent claim 1 never mentioned an “original protein sequence” as claimed; thus, there is a lack of antecedent basis issue with respect to “the” original protein sequence.
Thirdly, even if Applicant were to amend this to be “an” original protein sequence, it is unclear where an “original protein sequence” fits within the context of the claimed invention, as the independent claim only mentions performing a processing on multiple sequence alignments, which are augmentations of the protein sequence (i.e., “augmenting the protein sequence to obtain multiple sequence alignment”).
Lastly, in a related vein to the second point above, it is unclear how there would be “any individual and pairwise embeddings that are from the original protein structure”, as all individual and pairwise embeddings that were “enriched” (i.e., “producing enriched individual and pairwise embeddings from the multiple sequence alignments [which are augmented protein sequences]”). Thus, none of the data appears to be “original”, but rather augmentations/enriched forms, particularly the individual and pairwise embeddings that were generated. Thus, it cannot be ascertained what is meant by “any individual and pairwise embeddings that are from the original protein structure” as claimed, given that the individual and pairwise embeddings are enriched, and are produced after the protein sequence was already augmented.
Claims 4-6 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.
Independent Claim 4 recites “folding a protein sequence in silico”. It is unclear how this is different from the “modelling of a three-dimensional folded protein structure” as seen in independent claims 1 and 7.
Independent Claim 4 further recites “producing information-dense embeddings from the protein sequence”. It is unclear how this is different from the claimed “individual and pairwise embeddings” of independent claims 1 and 7.
Independent Claim 4 further recites “producing initial node and edge hidden representations in a complete graph”. It is unclear what is meant by “hidden” within the context of the claimed invention, as “hidden” is utilized within the context of neural networks. Although neural networks are graphs, they are directed graphs where every node may not be connected. A “complete graph” is an undirected graph where every node is connected to every other node in the graph by an edge. Therefore, “hidden representations” within the context of “a complete graph” as claimed does not make sense, as the “complete graph” does not appear to be a neural network, and “hidden representations” do not appear within the context of “complete graphs” as known in the technology.
Furthermore, it is unclear whether or how this might overlap with independent claims 1 and 7, which recite “extracting…relevant features and structure latent states” (followed by a limitation of “assigning individual and pairwise embeddings to nodes and edges, respectively”).
Independent Claim 4 further recites “processing and structuring geometric information, to obtain final node representations”. It is unclear whether “structuring” is an active verb, or whether it is referring to, e.g., the protein structure. Thus, the metes and bounds of such a limitation cannot be ascertained.
Independent Claim 4 further recites “using the attention-based graph transformer architecture”. It is unclear whether this is the same or different from independent claims 1 and 7, which recite a “downstream graph transformer” which operates on node representations “through an attention-based mechanism”.
Independent Claim 4 further recites “projecting the final node representations into Cartesian coordinates through a learnable transformation to obtain the folded protein sequence”.
Firstly, it is unclear whether “final node representations” is the same as “final node encodings” of claims 1 and 7.
Secondly, it is uncertain/unclear what is meant by “learnable transformation”.
Thirdly, it is unclear what a folded protein “sequence” is. Although the Specification states that three-dimensional folded structure models may be produced from protein sequences, and that the geometric deep learning architecture maps protein sequences to folded, three-dimensional structures (see, e.g., Specification, [0007-0008]), claim 4 tends to overlap in its steps with claims 1 and 7; thus, it is unclear what belongs to the protein sequence aspect (as claimed in claim 4), and what belongs to the folded protein structure aspect (as claimed in claims 1 and 7), as there appears to be a great amount of overlap, e.g., elements contextually share the same function, yet are mentioned as distinct elements within the Specification.
For purposes of examination, the interpretation that “node representation” and “node encoding” are interchangeable, as are “protein sequence” and “protein structure”, and that some sort of machine learning algorithm or structure capable of executing these steps represents a “learnable transformation”.
Lastly, independent claim 4 recites obtaining “the folded protein sequence”. It is unclear whether this is the same or different from the claimed “computer-modeled folded protein structure” seen in claims 1 and 7 (i.e., given that claim 4 pertains to “folding a protein sequence in silico”, as seen in the preamble).
Dependent Claims 5-6 are rejected for at least by virtue of their dependency on claim 4, and for failing to cure the deficiencies of claim 4.
Claim 6 is 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.
Claim 6 recites “comparing the induced distance maps to ground truth counterparts in order to define the loss”. The metes and bounds of “in order to” cannot be ascertained, e.g., it cannot be ascertained whether the definition of the loss is actually being claimed or not.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-8 are rejected under 35 U.S.C. 101 because the claims are directed to a judicial exception (i.e., an abstract idea) without significantly more.
Independent Claims 1 and 7 recite augmenting the protein sequence to obtain multiple sequence alignments; producing enriched individual and pairwise embeddings from the multiple sequence alignments; extracting relevant features and structure latent states from the enriched individual and pairwise embeddings; assigning individual and pairwise embeddings to nodes and edges, respectively; [using] node representations and considering pairwise edge attributes to obtain final node encodings; and projecting the final node encodings to form a folded protein structure. Similarly, independent Claim 4 recites producing information-dense embeddings from the protein sequence, producing initial node and edge hidden representations in a complete graph from the (information-dense) embeddings, processing and structuring geometric information to obtain final node representations, and projecting the final node representations into (spatial) coordinates through a transformation to obtain the folded protein sequence. These encompass an evaluation, observation and/or judgment (including representing the data as nodes and edges of a (complete) graph, as this is no different than a person using a pen and paper to draw out the graph with labeled nodes and edges), as well as mathematical concepts (e.g., projecting the final node encodings to form a folded protein structure), both of which fall under the “Mental Processes” grouping of abstract ideas.
Dependent Claims 2 and 8 recite computing an induced distogram of the folded protein structure. Similarly, dependent claim 5 recites calculating induced distance maps from the projected final node representations. This encompasses an evaluation, observation and/or judgment, as well as mathematical concepts, both of which fall under the “Mental Processes” grouping of abstract ideas.
Dependent Claim 6 recites comparing the induced distance maps to ground truth counterparts in order to define the loss. This encompasses an evaluation, observation and/or judgment, as well as mathematical concepts, both of which fall under the “Mental Processes” grouping of abstract ideas.
Because the claims cover performance of the limitation in the mind but for the recitation of generic computer components, the claims fall within the “Mental Processes” grouping of abstract ideas.
The judicial exception is not integrated into a practical application of the idea. The independent claims recite various computing components, including a computer processor (claims 1 and 7), an MSA transformer (claims 1, 4, and 7), downstream graph transformer (claims 1 and 7), that the claimed invention is within the context of computer-modelling (claims 1 and 7), a graph transformer architecture (claim 4), and that the folding of a protein sequence is done in silico. These are recited at a high level of generality and recited so generically that they represent no more than mere instructions to apply the judicial exception on a computer (see MPEP 2106.05(f)). These limitations can also be viewed as nothing more than an attempt to generally link the use of the judicial exception to the technological environment of a computer (see MPEP 2106.05(h)).
Independent Claims 1 and 7 further recite that the downstream graph transformer operates on node representations through an attention-based mechanism. Similarly, independent Claim 4 recites that the graph transformer architecture is attention-based, and that the projection of the final node representations into coordinates that are in Cartesian space, and that such a projection is performed through a “learnable” transformation.
However, these features simply provide further narrowing of what are still mental processes (or in the case of “learnable” transformation, a mathematical operation), adding nothing outside the abstract realm. See, e.g., SAP America, Inc. v. InvestPic, LLC, 890 F.3d 1016, 126 USPQ2d 1638 (Fed. Cir. 2018) at p. 12 (citing Mayo Collaborative Services v. Prometheus Laboratories, Inc., 132 S. Ct. 1289 (2012), 566 U.S. at 88-89 (stating that narrow embodiments of ineligible matter, citing mathematical ideas as an example, are still ineligible)). As a matter of law, narrowing or reformulating an abstract idea does not add “significantly” to it. See Id., slip op. at 14.
Thus, such limitations are, at best, nothing more than insignificant field-of-use limitations, describing the context rather than a particular manner of achieving the result.
Dependent Claim 3 recites storing data. This is an insignificant extra-solution activity, which is a nominal or tangential addition to the claim. The fact that that data pertains to individual and pairwise embeddings from the original protein sequence is nothing more than an insignificant field-of-use limitation, describing the context rather than a particular manner of achieving the result.
As such, the additional elements do not integrate the abstract idea into a practical application of that idea.
With respect to the well-understood, routine, and conventional elements, as stated previously above, the claims do not include any additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to the integration of the abstract idea into a practical application, the additional elements reciting the use of various computing components amount to no more than mere instructions to apply the judicial exception using generic components. Mere instructions to apply an exception using generic computer components cannot provide an inventive concept (or even components slightly narrower than generic computer components).
Additionally, dependent claim 3 recites storing data. However, this is a well-understood, routine, and conventional activity within the computing realm. See, e.g., MPEP § 2106.05(d)(II) (“Storing and retrieving data in memory” and “Electronic recordkeeping”).
Even as an ordered combination, the claims as a whole do not contain any additional elements that amount to significantly more. The focus of the claims is on performing data manipulation on certain data, manipulating/transforming them into a graph representation, and then using that graph representation to transform it into another form in order to obtain the folded protein structure/sequence, sometimes alluding to mathematical operations such as projections (claims 1, 4, and 7), and into a Cartesian space (claim 4). However, the claims contain no hint as to how any of these steps are particularly performed by a computer, other than reciting, at a high level of generality, that certain computing components such as a downstream graph transformer using attention-based mechanisms (claims 1 and 7) or attention-based graph transformer architecture (claim 4) are utilized. However, the incorporation of such computing components does nothing more than attempt to link the claims to a particular technological environment—namely, implementation via computers—which does not add significantly more.
Thus, when removing the computing elements, the claims do no more than describe a desired function or outcome, without providing any limiting detail that confines the claims to a particular solution to an identified problem by a computer aside from invoking the computer as a tool to be used in executing the claimed steps, i.e., applying the abstract idea with a computer. The purely functional nature of the claim confirms that it is directed to an abstract idea, not to a concrete embodiment of that idea (see Affinity Labs of Texas LLC v. Amazon.com Inc., 838 F.3d 1253 (Fed. Cir. 2016) at p. 7-8, citing Elec. Power Grp., LLC v. Alstom S.A., 830 F.3d 1350 (Fed. Cir. 2016), slip op. 12 (“[T]he essentially result-focused, functional character of claim language has been a frequent feature of claims held ineligible under § 101”)).
As a whole, the claims do not go beyond stating the relevant functions in general terms, without limiting them to a technical means for performing the functions that are arguably an advance over conventional computing technologies. Neither stating an abstract idea while adding the words “apply it” with a computer, nor limiting the use of an abstract idea to a particular technological environment is enough for patent eligibility. Stating the abstract idea while adding the words “apply it with a computer” simply combines those two steps, with the same deficient result.
Therefore, for at least the aforementioned reasons, the claims are rejected under 35 U.S.C. 101 for being directed to a judicial exception (i.e., an abstract idea) without significantly more.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-3 and 7-8 are rejected under 35 U.S.C. 103 as being unpatentable over Jumper et al. (“Jumper”) (US 2021/0166779 A1), in view of Rosenbluth et al. (“Rosenbluth”) (US 2021/0174893 A1), in further view of Eyal et al. (“Eyal”) (US 2008/0215301 A1).
Regarding claim 1: Jumper teaches A method for computer modelling of a three-dimensional folded protein structure based on a protein sequence, comprising: using a computer processor, performing the steps of (Jumper, [0097-0099] and [0103], where the disclosed system may be implemented as a data processing apparatus, which includes a programmable processor that executes a computer program for performing the disclosed steps):
augmenting the protein sequence to obtain multiple sequence alignments (Jumper, [0052], where the system 100 obtains a multiple sequence alignment (MSA) 108 corresponding to the amino acid sequence 102 of the protein 104 (i.e., “protein sequence”) by specifying a sequence alignment of the amino acid sequence 102 with multiple additional amino acid sequences, e.g., from other homologous proteins (i.e., “augmenting the protein sequence”));
using an MSA-Transformer, producing enriched individual and pairwise embeddings from the multiple sequence alignments (Jumper, [0053], where the system 100 generates the predicted structure 106 from the amino acid sequence 102 and the MSA 108 using (1) a multiple sequence alignment (MSA) embedding system 110, (2) a pair embedding neural network 112, and (3) a folding neural network 114. See Jumper, [0054-0060], where the MSA embedding system 110 processes the MSA 108 to generate an MSA embedding 116 (i.e., “individual embedding”), where after generating the MSA embedding 116, the system 100 transforms the MSA embedding 116 into an alternative representation as a collection of pair embeddings (i.e., “pairwise embeddings”));
using the downstream graph transformer, operating on … representations through an attention-based mechanism that considers pairwise edge attributes to obtain final … encodings (Jumper, [0095], where the system processes initial embeddings of the pairs of amino acids using a pair embedding neural network that includes multiple self-attention neural network layers to generate a final embedding of each pair of amino acids (i.e., “final encodings”)); and
projecting the final … encodings to form the computer-modeled folded protein structure (Jumper, [0045], where the system processes an amino acid sequence 102 of a protein 104 to generate a predicted structure 106 of the protein, the predicted structure 106 defining an estimate of a three-dimensional (3-D) configuration of the atoms in the amino acid sequence 102 of the protein 104 after the protein 104 undergoes protein folding. See Jumper, [0096], where the system determines the predicted structure of the protein based on the final embedding using a folding neural network (implying that the final embedding (i.e., “final encodings”) are projected to the 3-D configuration of the atoms, i.e., “projecting the final encodings to form the computer-modeled folded protein structure”) as claimed).
Jumper does not appear to explicitly teach extracting, from the enriched individual and pairwise embeddings, relevant features and structure latent states for use by a downstream graph transformer; assigning individual and pairwise embeddings to nodes and edges, respectively; and that the representations involve nodes and edges.
Rosenbluth teaches extracting, from the enriched individual and pairwise embeddings, relevant features and structure latent states for use by a downstream graph transformer (Rosenbluth, [0245], where a multiple sequence alignment (MSA) is first performed for a protein, followed by feature extraction by computing Potts model parameter and applying the direct-coupling analysis (DCA), where prior and posterior distograms are obtained using these features, where the output of the DCA is a matrix that represents the “strength” of the coupling between all pairs of residues, where a high DCA output value indicates that two residues are physically in contact (i.e., “structure latent states”)5).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Jumper and Rosenbluth (hereinafter “Jumper as modified”) with the motivation of focusing attention on certain features that may be more relevant, thereby leading to greater accuracy and greater efficiency (e.g., reduced number of features to consider).
The Examiner notes that “for use by a downstream graph transformer” has been considered as an intended use/result, and is not afforded patentable weight.
The Examiner notes that “A claim containing a ‘recitation with respect to the manner in which a claimed apparatus is intended to be employed does not differentiate the claimed apparatus from a prior art apparatus’ if the prior art apparatus teaches all the structural limitations of the claim.” Ex parte Masham, 2 USPQ2d 1647 (Bd. Pat. App. & Inter. 1987); see also MPEP § 2114. The recited prior art has the capability to perform these intended use limitations, and therefore, the prior art meets the claimed limitations. See MPEP § 2111.02; see also In re Schreiber, 128 F.3d 1473, 1477, 44 USPQ2d 1429, 1431 (Fed.Cir. 1997).
Because the combination of Jumper and Rosenbluth (“Jumper as modified”) disclose all the claimed features, the claimed invention does not distinguish over the prior art since Jumper as modified would confer the same intended use/result as claimed.
Jumper as modified does not appear to explicitly teach assigning individual and pairwise embeddings to nodes and edges, respectively; and that the representations involve nodes and edges.
Eyal teaches assigning individual and pairwise embeddings to nodes and edges, respectively; and that the representations involve nodes and edges (Eyal, [0138], where a graph G(N,E) is constructed in which nodes correspond to amino acids (columns in the alignment) and edges correspond to contacting amino acids).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Jumper as modified and Eyal (hereinafter “Jumper as modified”) with the motivation of conveniently representing spatial conformations by utilizing graphs to represent protein structures, given that graphs represent the most appropriate data structures for modeling the complex structures of proteins, as the graph representation can preserve the overall structure of the protein and its components.6
Regarding claim 2: Jumper as modified teaches The method of claim 1, further comprising computing an induced distogram of the computer-modeled folded protein structure (Rosenbluth, [0245], where a multiple sequence alignment (MSA) is first performed for a protein, followed by feature extraction by computing Potts model parameter and applying DCA, where prior and posterior distograms are obtained using these features. See also Rosenbluth, [0296], where the protein structure prediction model predicts the relative locations of all amino acids on the surface of the protein relative to one another in order to produce a distogram or distance matrix. See Rosenbluth, [0049-0052], where the distogram output of the algorithm can be constrained, e.g., limiting selection to only non-neighboring residue pairs (i.e., “induced distogram”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Jumper as modified and Rosenbluth with the motivation of generating a more accurate and refined protein structure (Rosenbluth, [0052] and [0062]), the distogram (or “distance matrix”) being completely descriptive of the 3D structure (Rosenbluth, [0046]).
Regarding claim 3: Jumper as modified teaches The method of claim 1, further comprising storing any individual and pairwise embeddings that are from the original protein sequence (Jumper, [0103], where the one or more memory devices may be used for storing data. See Jumper, [0053] in Claim 1 above with respect to the “individual and pairwise embeddings that are from the original protein sequence”).
Although Jumper does not appear to explicitly state that the data stored pertains to the embeddings generated as seen in Jumper, [0053], one of ordinary skill in the art would have found it obvious to have modified Jumper to have explicitly included the step of storing such generated data with the motivation of saving processing resources in future iterations of the same process, e.g., instead of having to recompute the individual and pairwise embeddings, which may be resource-intensive, the system may perform a simple lookup operation instead.
Regarding claim 7: Claim 7 recites substantially the same claim limitations as claim 1, and is rejected for the same reasons.
Regarding claim 8: Claim 8 recites substantially the same claim limitations as claim 2, and is rejected for the same reasons.
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Jumper et al. (“Jumper”) (US 2021/0166779 A1), in view of Creed et al. (“Creed”) (US 2021/0081717 A1).
Regarding claim 4: Jumper teaches A method for folding a protein sequence in silico using an attention-based graph transformer architecture, comprising:
using the MSA transformer, producing information-dense embeddings from the protein sequence (Jumper, [0053], where the system 100 generates the predicted structure 106 from the amino acid sequence 102 and the MSA 108 using (1) a multiple sequence alignment (MSA) embedding system 110, (2) a pair embedding neural network 112, and (3) a folding neural network 114. See Jumper, [0054-0060], where the MSA embedding system 110 processes the MSA 108 to generate an MSA embedding 116, where after generating the MSA embedding 116, the system 100 transforms the MSA embedding 116 into an alternative representation as a collection of pair embeddings (i.e., “information-dense embeddings”));
using the attention-based graph transformer architecture, processing and structuring geometric information, to obtain final node representations (Jumper, [0095], where the system processes initial embeddings of the pairs of amino acids using a pair embedding neural network that includes multiple self-attention neural network layers to generate a final embedding of each pair of amino acids (i.e., “final representations”). See Jumper, [0048-0049] and [0068-0071], with respect to “processing and structuring geometric information”, where the predicted structure 106 of the protein 104 is partially defined by location parameters, where the location parameters for an amino acid may specify a predicted 3-D spatial location of a specified atom in the amino acid in the structure of the protein); and
projecting the final node representations into Cartesian coordinates through a learnable transformation to obtain the folded protein sequence (Jumper, [0045], where the system processes an amino acid sequence 102 of a protein 104 to generate a predicted structure 106 of the protein, the predicted structure 106 defining an estimate of a three-dimensional (3-D) configuration of the atoms in the amino acid sequence 102 of the protein 104 after the protein 104 undergoes protein folding. See Jumper, [0096], where the system determines the predicted structure of the protein based on the final embedding (i.e., analogous to the step of “obtain[ing] the folded protein sequence”) using a folding neural network (i.e., “learnable transformation”), implying that the final embedding (i.e., “final representations”) are projected to the 3-D configuration of the atoms, e.g., a location represented in a three-dimensional [x,y,z] Cartesian coordinate system (i.e., “projecting the final representations into Cartesian coordinates”)).
Jumper does not appear to explicitly teach from the embeddings, producing initial node and edge hidden representations in a complete graph.
Creed teaches from the embeddings, producing initial node and edge hidden representations in a complete graph (Creed, [0127], where the embedding hi for entity i for a particular layer can be computed as a series of matrix multiplications, which includes a hidden representation H(l) ϵ. See also Creed, [0132], where the embedding of the conventional GCNN is modified to include an attention weight for each relationship edge, which includes the associated relationship adjacency matrix for the i-th hidden layer associated with the entity node i. These imply that there are “initial node and edge hidden representations” as claimed (and thus imply that these were produced). See Creed, [FIG. 1b] and [0092], which shows an example entity-entity graph for input to a GNN model).
Although Creed does not appear to explicitly state that the graph is a “complete graph” as claimed, one of ordinary skill in the art would have found it obvious to have modified Creed to have explicitly included graphs that are complete, as claimed, with the motivation of taking into account all possible linkages.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Jumper and Creed (hereinafter “Jumper”) with the motivation of efficiently performing matrix computations, thus speeding up (i.e., reducing) the processing time.
Claims 5-6 are rejected under 35 U.S.C. 103 as being unpatentable over Jumper et al. (“Jumper”) (US 2021/0166779 A1), in view of Creed et al. (“Creed”) (US 2021/0081717 A1), in further view of Rosenbluth et al. (“Rosenbluth”) (US 2021/0174893 A1).
Regarding claim 5: Jumper as modified teaches The method of claim 4, but does not appear to explicitly teach further comprising calculating induced distance maps from the projected final node representations.
Rosenbluth teaches further comprising calculating induced distance maps from the projected final node representations (Rosenbluth, [0046], where co-evolutionary statistical models can be used to generate “contact maps” that describe inter-residue contacts protein-wide, where contact maps are an important first step towards predicting all inter-residue (pairwise) distances for the amino acids in a protein. Such a distance matrix would be completely descriptive of the 3D structure. See also Rosenbluth, [0062], where the final model output is a distance matrix capturing the structure of the target protein, relative to random initialization. See Rosenbluth, [0049-0052], where the distogram output of the algorithm can be constrained, e.g., limiting selection to only non-neighboring residue pairs (i.e., “induced distance maps”)).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have combined the teachings of Jumper as modified and Rosenbluth with the motivation of generating a more accurate and refined protein structure (Rosenbluth, [0052] and [0062]), the distogram (or “distance matrix”) being completely descriptive of the 3D structure (Rosenbluth, [0046]).
Regarding claim 6: Jumper as modified teaches The method of claim 5, further comprising comparing the induced distance maps to ground truth counterparts in order to define the loss (Jumper, [0067-0069], where the structure loss characterize a similarity between a predicted protein structure by the system, and the target protein structure that should have been generated by the system (i.e., “ground truth counterparts”), where the structure loss is defined with reference to equations (10)-(12). These terms are sensitive to the predicted and actual rotations of amino acid i and j and therefore carry richer information than loss terms that are only sensitive to the predicted and actual distances between amino acids (implying that distance between amino acids (i.e., “induced distance maps”) may be used “to define the loss” as claimed; see, e.g., Jumper, [0062-0069] with respect to how the distance is utilized to determine attention weights, in turn are eventually used to define the structure loss).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure. See the enclosed 892 form.
Kinjo et al. (“Liquid-theory analogy of direct-coupling analysis of multiple-sequence alignment and its implications for protein structure prediction”, published 2015) is cited to show that the prior art’s disclosure of DCA implies an extraction of structure latent states, as claimed (see Kinjo et al., [Abstract]).
Dhifli et al. (“ProtNN: fast and accurate protein 3D-structure classification in structural and topological space”) is cited to show why it would have been obvious to one of ordinary skill in the art to represent the associated protein data as a graph (see Dhifli et al., [page 3] and [Figure 1]).
The prior art should be considered to define the claims over the art of record.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to IRENE BAKER whose telephone number is (408)918-7601. The examiner can normally be reached M-F 8-5PM PT.
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, NEVEEN ABEL-JALIL can be reached at (571)270-0474. 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 yo