DETAILED ACTION
Examiner Remarks
Examiner has withdrawn the section 112(b) rejection in light of Applicant’s amendments to claim 2. After the cancelation of claim 18, claims 1-17 and 19-21 are currently pending.
Response to Arguments
Applicant’s arguments with respect to claims 1 and 20 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged by Applicant’s argument submitted on 03/03/2026. Furthermore, with respect to the claim limitation of computational transformation to the genetic data, the prior art of Liu teaches this claim limitation. See the Current Office Action for the detailed teachings.
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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 03/04/2026 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1, 4-17, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zamft et al. US 2022/0301658 Al (“Zamft”) in view of Liu T et al., scElmo: Embeddings from language models are good learners for single-cell data analysis. bioRxiv. 2024 March 3(“Liu”) and in view of Serber et al., US 2017 /0159045 Al(“Serber”) and further in view of Hauck US 2023/0170040 Al(“Hauck”)
Regarding claim 1, Zamft teaches a method performed by one or more computers for predicting performance associated with genetic edits, the method comprising:
generating, the performance prediction for the strain [based on inputting the set of genetic embeddings] to a prediction model included in the machine learning model, wherein the prediction model is a neural network trained to predict strain performance based on training data including: information about a plurality of genetic edits corresponding to a plurality of strains of the microorganism(Zamft, paras. 0042-0050 see also fig. 2A, “FIG. 2A shows an exemplary deep neural network 200 (in this instance the exemplary deep neural network is a feedforward neural network... although the training-tuning-validation mechanisms described herein focus on training a new deep neural network 200. These training-tuning-validation mechanisms can also be utilized to fine tune existing deep neural networks 200 trained from other datasets. For example, in some instances, a deep neural network 200 might have been pre-trained using gene expression profile data for a first phenotype. In those cases, the deep neural network 200 can be used for transfer learning and retrained/validated using new sets of gene expression profiles for a second phenotype”);1
and target data indicating a corresponding performance of each of the plurality of strains of the microorganism(Zamft, paras. 0074-0079, see also fig. 8, “[A] set of genomic regions are identified, based on the identified set of candidate gene targets, that when edited provide a requisite change in a gene expression profile to realize an expected phenotypic change.”)
wherein the prediction model applies computational transformations [to the set of genetic embeddings] to generate the performance prediction(Zamft, paras. 0074-0079, see also fig. 8, “In instances in which the prediction model is a deep neural network, the gene edit model may model the gene edits by performing an adversarial attack on the deep neural network, the adversarial attack comprising freezing the weights of the deep neural network, and optimizing over a space of constrained inputs to maximize, minimize, or otherwise modulate the phenotype.”).2
Zamft does not teach: based on inputting the set of genetic embeddings; to the set of genetic embeddings; and applies computational transformations to the genetic edit data using a corresponding embedding model to generate a multi-dimensional vector representation for each of the plurality of genetic edits, wherein each multi- dimensional vector representation generated by the one or more embedding models is added to the set of genetic embeddings.
However, Liu teaches:
based on inputting the set of genetic embeddings(Liu, pgs., 14-16, see also fig. 1, “For a typical single-cell dataset
X
n
×
m
... with n cells and m features, our target is to utilize the text description from a mapping function
M
(
)
for feature-level metadata information
f
m
×
1
and cell-level metadata information
c
n
×
1
to learn the embeddings of cells. If we define the embedding generation layer of
M
(
)
as
M
e
(
)
, our cell embeddings can be represented as...
e
c
e
l
l
s
=
A
V
G
X
e
f
+
e
c
[based on inputting the set of genetic embeddings]where Prompt is a mapping function that can transfer the name of input data to the prompt space. The prompts can be used as the input of language models. The function
A
V
G
(
)
represents the method we used to average the embeddings of all genes for each cell... [o]ur default setting of the mapping function is a LLM. GenePT can be treated as a special case...that is, replacing the LLM with a known database and using aa mode.”);
to the set of genetic embeddings(Liu, pgs., 14-16, see also fig. 1, “For a typical single-cell dataset
X
n
×
m
... with n cells and m features, our target is to utilize the text description from a mapping function
M
(
)
for feature-level metadata information
f
m
×
1
and cell-level metadata information
c
n
×
1
to learn the embeddings of cells. If we define the embedding generation layer of
M
(
)
as
M
e
(
)
, our cell embeddings can be represented as...
e
c
e
l
l
s
=
A
V
G
X
e
f
+
e
c
where Prompt is a mapping function that can transfer the name of input data to the prompt space. The prompts can be used as the input of language models. The function
A
V
G
(
)
represents the method we used to average the embeddings of all genes for each cell[to the set of genetic embeddings]... [o]ur default setting of the mapping function is a LLM. GenePT can be treated as a special case...that is, replacing the LLM with a known database and using aa mode.”);
and applies computational transformations to the genetic edit data using a corresponding embedding model to generate a multi-dimensional vector representation for each of the plurality of genetic edits, wherein each multi- dimensional vector representation generated by the one or more embedding models is added to the set of genetic embeddings(Liu, pgs., 14-16, see also fig. 1, “For a typical single-cell dataset
X
n
×
m
[the input]... with n cells and m features, our target is to utilize the text description from a mapping function
M
(
)
for feature-level metadata information
f
m
×
1
and cell-level metadata information
c
n
×
1
[]to learn the embeddings of cells. If we define the embedding generation layer of
M
(
)
as
M
e
(
)
, our cell embeddings can be represented as...
e
c
e
l
l
s
=
A
V
G
X
e
f
+
e
c
where Prompt is a mapping function that can transfer the name of input data to the prompt space. The prompts can be used as the input of language models[and applies computational transformations to the genetic edit data using a corresponding embedding model to generate a multi-dimensional vector representation for each of the plurality of genetic edits,]. The function
A
V
G
(
)
represents the method we used to average the embeddings of all genes for each cell[wherein each multi- dimensional vector representation generated by the one or more embedding models is added to the set of genetic embeddings]... [o]ur default setting of the mapping function is a LLM. GenePT can be treated as a special case...that is, replacing the LLM with a known database and using aa mode.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the teachings of Liu the motivation to do so would be to use pre-trained Large Language Models for zero-shot learning to handled different tasks associated with cellular data(Liu, pg., 3, “The high-level idea of scELMo is to transfer the information of each cell from the sequencing data space to the embedding space from LLM... [a]fter having embeddings for features or cells, we can use them based on the zero shot learning framework for clustering or batch effect correction. Moreover, we can also combine the embeddings with known models to improve their performance for various downstream tasks.”).
Zamft in view of Liu does not teach: obtaining a model input comprising genetic edit data for a strain of a microorganism, wherein the genetic edit data defines: (i) a plurality of genes of the strain that are modified relative to a base strain of the microorganism, and (ii) a type of modification for each of the plurality of genes of the strain that are modified; processing the model input comprising the genetic edit data by a machine learning model to generate a model output defining a performance prediction for the strain
However, Serber teaches:
obtaining a model input comprising genetic edit data for a strain of a microorganism, wherein the genetic edit data defines: (i) a plurality of genes of the strain that are modified relative to a base strain of the microorganism, and (ii) a type of modification for each of the plurality of genes of the strain that are modified(Serber, paras. [0431-0439], see also figs. 23-24 and table 3, “[F]or the sake of ease of illustration, input data may comprise two components: (1) sets of genetic changes and (2) relative strain performance. Those skilled in the art will recognize that this model can be readily extended to consider a wide variety of inputs... [a]n example set of entries from a table of genetic changes is shown below in Table 3. Each row indicates a genetic change in strain 7000051473, as well as metadata about the mechanism of change....”);
processing the model input comprising the genetic edit data by a machine learning model to generate a model output defining a performance prediction for the strain(Serber, paras. [0434-0448], see also figs. 23-24 and table 4, “The goal of the taught model is to predict strain performance based on the composition of genetic changes introduced to the strain. To construct a standard for comparison, strain performance is computed relative to a common reference strain, by first calculating the median performance per strain, per assay plate. Relative performance is then computed as the difference in average performance between an engineered strain and the common reference strain within the same plate.”)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft in view of Liu with the teachings of Serber the motivation to do so would be to develop a microbial genomic engineering platform that includes machine learning for automating and speeding up the mutagenesis process of genetic edits to strains(Serber, para. [0003-0010], “[T]raditional microbial strain improvement
programs inefficient, but the process can also lead to industrial strains with a high degree of detrimental mutagenic load. The accumulation of mutations in industrial strains subjected to these types of programs can become significant and may lead to an eventual stagnation in the rate of performance improvement. Thus, there is a great need in the art for new methods of engineering industrial microbes, which do not suffer from the aforementioned drawbacks inherent with traditional strain improvement programs and greatly accelerate the process of discovering and consolidating beneficial mutations.”).
Zamft in view of Liu and Serber do not teach: comprising: processing the genetic edit data using one or more embedding models included in the machine learning model to generate a set of genetic embeddings that collectively characterize the genetic edit data wherein each of the one or more embedding models: receives the genetic edit data.
However, Hauck teaches:
comprising: processing the genetic edit data using one or more embedding models included in the machine learning model to generate a set of genetic embeddings that collectively characterize the genetic edit data wherein each of the one or more embedding models: receives the genetic edit data(Hauck, para. [0120-0138], see also fig. 5, “FIG. 5 depicts an example process for creating sampling probabilities from a metabolite in an embedding space. In the example, a Poincare embedding space 503 contains a key metabolite node 505, which serves as a query nodes, a plurality of gene nodes 507, which serve as candidate nodes for editing... predicted edits for nodes/elements of a microorganism from an embedded vector space are used to train a machine learning classification model. The resulting model predicts whether a particular edit can be successful.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft in view of Liu and Serber with the teachings of Hauck the motivation to do so would be to represent genetic edits in an embedding space to associate relations between different genes and their edits for better targeting(Hauck, para. [0003-0013], “One aspect of the disclosure provides methods of
identifying one or more elements for modification to cause a change in functioning of a microorganism. Such methods may be characterized by the following operations: (a) receiving a graph representation of a biological network of interacting elements of a microorganism or a plurality of related microorganisms; (b) converting the graph representation of the biological network to a vector representation having locations of the interacting elements, which locations are derived from positions of said interacting elements in the graph representation; ( c) determining distance relationships between the interacting elements as represented in the vector representation; and ( d) from the distance relationships determined in ( c ), recommending a subset of the interacting elements for modification in an engineered variant”).
Regarding claim 4, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the prediction model comprises a first stage that generates a strain embedding characterizing the strain of the microorganism and a second stage that generates the performance prediction based on the strain embedding(Liu, pgs., 11-12, see also fig. 5, “CPA is a tool based on Conditional Variational Auto-encoder (CVAE) to predict the gene expression levels for the out-of-distribution (OOD) samples of scRNA seq data under certain perturbations. Here we combined the gene embeddings from scELMo[a first stage that generates a strain embedding characterizing the strain of the microorganism] with the original input dataset and learned a new latent space for gene expression prediction[and a second stage that generates the performance prediction based on the strain embedding]. We investigated the contribution of gene embeddings by comparing the R2 score between these two different settings.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
Regarding claim 5, Zamft in view of Liu, Serber and Hauck teaches the method of claim 4, wherein the first stage is one or more of a long-short term memory (LSTM) model, a transformer model, or a convolutional neural network (CNN) model(Liu, pg., 3, “Here we choose GPT 3.5[a transformer model] as the tool to summarize the functions of features and to generate embeddings based on our evaluation....”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
Regarding claim 6, Zamft in view of Liu, Serber and Hauck teaches the method of claim 4, wherein the second stage is a multi-layer perceptron(Liu, pg., 11, “CPA is a tool based on Conditional Variational Auto-encoder (CVAE)[ wherein the second stage is a multi-layer perceptron] to predict the gene expression levels....”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
Regarding claim 7, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the performance prediction comprises at least one of a predicted growth rate, a predicted metabolite production rate, a predicted byproduct formation rate, or a predicted protein expression level(Zamft, paras. 0067-0073, see also fig. 7, “The modeling system 737 uses the received data (e.g., plant extracted data, multi-omics profiles, management practices profiles, environmental conditions profiles, etc.) for the development ( e.g., design, training, validation, and deployment) of various models (e.g., machine learning models) that the gene discovery and editing system 700 can then use to guide growth of present plants 727[wherein the performance prediction comprises at least one of a predicted growth rate] and the generation of new plants with desired phenotypes.”).3
Regarding claim 8, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, further comprising receiving process condition information, wherein the prediction model is trained to predict performance with respect to a set of process conditions as indicated by process inputs, and wherein generating the performance prediction for the strain is further based on process inputs corresponding to the process condition information(Zamft, paras. 0067-0073, see also fig. 7, “The modeling system 737 uses the received data (e.g., plant extracted data, multi-omics profiles, management practices profiles, environmental conditions profiles, etc.) for the development ( e.g., design, training, validation, and deployment) of various models (e.g., machine learning models) that the gene discovery and editing system 700 can then use to guide growth of present plants 727 and the generation of new plants with desired phenotypes.”).
Regarding claim 9, Zamft in view of Liu, Serber and Hauck teaches the method of claim 8, wherein the process condition information comprises at least one of bioreactor volume, temperature, pH, oxygen levels, or substrate concentrations(Zamft, paras. 0067-0073, see also fig. 7, “The data environment conditions profile data may include the location-specific environmental conditions of the plant 727 is exposed to at various times of growth or continuously though-out the life cycle of plant 727, e.g. temperature, precipitation, soil properties, etc[wherein the process condition information comprises at least one of bioreactor volume, temperature].”).4
Regarding claim 10, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the prediction model is trained, by pre-training the prediction model using the training data; and fine-tuning the prediction model using additional strain-specific data, wherein the training data is a larger data set compared to the additional strain-specific data(Zamft, paras. 0042-0050 see also fig. 2A, “These training-tuning-validation mechanisms can also be utilized to fine tune existing deep neural networks 200 trained from other datasets. For example, in some instances, a deep neural network 200 might have been pre-trained using gene expression profile data for a first phenotype[prediction model using the training data]. In those cases, the deep neural network 200 can be used for transfer learning and retrained/validated using new sets of gene expression profiles for a second phenotype[and fine-tuning the prediction model using additional strain-specific data, wherein the training data is a larger data set compared to the additional strain-specific data].”).
Regarding claim 11, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, further comprising updating the prediction model using an active learning process including generating a set of candidate genetic modifications(Zamft, paras. 0053-0056, see also figs. 1 and 2, “In order to generate candidate gene targets for the phenotype 115 of interest, XAI techniques are applied to obtain the importance of each feature for all predictions in
a holdout data set or a new set of input data (i.e., a set of gene expression profiles 120)[ updating the prediction model using an active learning process including, generating a set of candidate genetic modifications].”);
generating a corresponding performance prediction for each of the set of candidate genetic modifications using the prediction model(Zamft, paras. 0053-0056, see also figs. 1 and 2, “The deep learning architecture 130, which predicts the phenotypes 115, using gene expression profiles 120 as inputs is analyzed via XAI 135 to identify features (e.g., one or more genes 140) that have the largest contribution or influence on the deep learning architecture 130 output or prediction.”);
receiving experimental data associated with at least a portion of the set of candidate genetic modifications; updating the training data using the experimental data(Zamft, paras. 0053-0056, see also figs. 1 and 2, “The techniques used for XAI 135 ( e.g., SHAP) generate: (i) a set of feature importance scores ( quantitative value) for the features used in the prediction (some or all input features), and (ii) a ranking or otherwise sorting of the features through aggregation of the importance scores for each feature for all predictions in a holdout data set or a new set of input data (i.e., a set of gene expression profiles 120)[ receiving experimental data associated with at least a portion of the set of candidate genetic modifications;updating the training data using the experimental data].”);
and re-training the prediction model using the updated training data(Zamft, paras. 0053-0056, see also figs. 1 and 2, “The highest ranking or sorted feature(s) 140 may be the candidate genes to be involved in the molecular regulatory processes for that particular plant species and phenotype, and are used in the second component 110 for modeling gene edits[and re-training the prediction model using the updated training data].”).
Regarding claim 12, Zamft in view of Liu, Serber and Hauck teaches the method of claim 11, further comprising determining which portion of the set of candidate genetic modifications to test via experiment based at least in part on an uncertainty quantification generated by the prediction model(Zamft, para. 0065, see also fig. 6, “FIG. 6 shows an example of an ideal gene expression profile 600 determined using an adversarial based modeling approach for genes AT2G45660, AT2G45660, AT5G44590, AT3G52480 identified with SHAP based XAI of a DNN[determining which portion of the set of genetic modifications to test via experiment based at least in part on an uncertainty quantification generated by prediction model].”).
Regarding claim 13, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the prediction model is an ensemble of multiple prediction models(Zamft, para. 0066, “[T]wo or more of these approaches could be used in combination as an ensemble of approaches and the resulting gene expression profiles could be selected or combined to obtain the ideal gene expression profile. For example, every model makes a prediction (votes) for each test instance and the final output prediction is the one that receives more than half of the votes. If none of the predictions get more than half of the votes, then it may be determined that the ensemble method could not make a stable prediction for the given instance. Alternatively, an averaging technique could be used, where for every instance of the test dataset, the average predictions are calculated. Weights could also be implemented in either of these ensemble techniques to increase the importance of one or more models[wherein the prediction model is an ensemble of multiple prediction models].”).
Regarding claim 14, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the set of genetic embeddings captures functional relationships between genes and metabolic pathways(Liu, pg., 30, “We also considered genes whose silence might shift the cell embeddings from the control condition to diseased conditions[wherein the set of genetic embeddings]... [b]ecause the function of genes is closely related to the pathway[captures functional relationships between genes and metabolic pathways]... our findings can help analyze the pathogenesis of some diseases.”).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
Regarding claim 15, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the prediction model is trained to predict performance across multiple strains of different microorganisms(Zamft, paras. 0086-0087, “A sequential neural network was built to model the time taken by natural genetic variant lines of Arabidopsis thaliana to reach the reproductive stage (time to flower). In one example, the model was trained on publicly
available transcriptomes collected from leaves... [t]he transcriptomic data was available for 728 natural genetic variants[wherein the prediction model is trained to predict performance across multiple strains of different microorganisms]....”).
Regarding claim 16, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the information about the strain comprises information about the base strain(Zamft, para. 0065, see also fig. 6, “Also shown, is a comparison of the deal gene expression profiles 600/605 to a naturally occurring distribution of gene expression for the plant species over Samples 1-3[wherein the information about the strain comprises information about the base strain] to understand whether the gene edit recommendation is to upregulate or downregulate a particular
gene, a sub group of genes, or each gene within the ideal gene expression profiles 600/605.”).
Regarding claim 17, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the type of modification for each of the plurality of genes that are modified at least one of a gene knockout, a gene overexpression, or a gene under-expression(Zamft, paras. 0067-0074, see also fig. 7, “The gene editing system 712 makes genetic edits or perturbations to the genome of a given plant species ( e.g., plant 727) in accordance with the recommendation 775. Examples of gene editing systems include...CRISPR base editing...[f]or example, gene editing system 712 may make one or more combinatorial edits ("bashing") in the gene regulatory genomic regions (promoters, 5'UTR, 3'UTR, terminator) of one or more target genes in order to modify their expression (upregulation or downregulation)[a gene overexpression, or a gene under-expression]... [t]he modified genome of the given plant species may then be sent to the library design system 725 for use by the library 720....”).5
Regarding claim 19, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein generating the set of genetic embeddings occurs prior to training, the method further comprising caching the generated set of genetic embeddings for later use at prediction time(Liu, pgs., 3-4, As Fig. 1a of Liu details:
PNG
media_image1.png
272
756
media_image1.png
Greyscale
During the Zero-shot learning framework of scELMo the text description of metadata is extracted by using either databases or LLMs. Then we use GPT 3.5 to generate the embeddings of the text description as the embeddings of features or cell states. We then aggregate these embeddings with single-cell profiles to generate cell embeddings).
Regarding claim 20, Zamft teaches a system comprising: one or more computers; and one or more storage devices communicatively coupled to the one or more computers, wherein the one or more storage devices store instructions that, when executed by the one or more computers, cause the one or more computers to perform operations(Zamft, para. 0080, see also fig. 9, “The processing depicted in FIG. 9 may be implemented in software (e.g., code, instructions, program) executed by one or more processing units ( e.g., processors,
cores) of the respective systems, hardware, or combinations thereof. The software may be stored on a non-transitory storage medium ( e.g., on a memory device).”) and for all other claim limitations they are rejected on the same basis as independent claim 1 since they are analogous claims.
Claim(s) 2-3 are rejected under 35 U.S.C. 103 as being unpatentable over Zamft et al. US 2022/0301658 Al (“Zamft”) in view of Chen et al., GenePT: a simple but effective foundation model for genes and cells built from ChatGPT. bioRxiv. 2024 Mar 5 (“Chen”) and in view of Serber et al., US 2017 /0159045 Al(“Serber”) and in view of Hauck US 2023/0170040 Al(“Hauck”) and further in view of Yu T et al., Machine learning-enabled retrobiosynthesis of molecules. Nature Catalysis. 2023 Feb (“Yu”).
Regarding claim 2, Zamft in view of Liu, Serber and Hauck teaches the method of claim 1, wherein the one or more embedding models include two or more embedding models selected from: a GenePT model the method further comprising aggregating corresponding multi-dimensional vector representation generated by the two or more embedding models to create the set of genetic embeddings(Liu, pgs., 14-16, see also fig. 1, “For a typical single-cell dataset
X
n
×
m
... with n cells and m features, our target is to utilize the text description from a mapping function
M
(
)
for feature-level metadata information
f
m
×
1
and cell-level metadata information
c
n
×
1
to learn the embeddings of cells. If we define the embedding generation layer of
M
(
)
as
M
e
(
)
, our cell embeddings can be represented as...
e
c
e
l
l
s
=
A
V
G
X
e
f
+
e
c
where Prompt is a mapping function that can transfer the name of input data to the prompt space. The prompts can be used as the input of language models. The function
A
V
G
(
)
represents the method we used to average the embeddings of all genes for each cell[aggregating corresponding multi-dimensional vector representation generated by the two or more embedding models to create the set of genetic embeddings]... [o]ur default setting of the mapping function is a LLM. GenePT can be treated as a special case...that is, replacing the LLM with a known database and using aa mode[a GenePT model].”).6
Zamft in view of Liu, Serber and Hauck does not teach: a Proteinfer model, a pFBA-PCA model, or a GO-PCA model.
However, Yu teaches:
a Proteinfer model, a pFBA-PCA model, or a GO-PCA model(Yu, pg., 4, “ProteInfer, a state-of-the-art model[a Proteinfer model], is a dilated convolutional neural network (CNN) that infers functional annotations from protein sequences.”).7
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft in view of Liu, Serber and Hauck with the teachings of Yu the motivation to do so would be to develop a machine learning model that is able to predict enzymes from sequence regions to complete the mapping of metabolic pathways(Yu, pg., 4, “After a retrobiosynthesis route has been proposed, enzymes need to be identified and selected to fill the missing links between each step of the pathway. In the case where such enzymes are missing from the enzymatic databases, ML models have been developed to predict enzyme functions, reactivities and other properties, aiming to accelerate the identification and selection process.”).
Regarding claim 3, Zamft in view of Liu, Serber, Hauck and Yu teaches the method of claim 2, wherein each token of the set of genetic embeddings corresponds to a genetic edit of the plurality of genetic edits(Liu, pgs., 8-10, see also fig. 4, “[T]he gene embeddings we extracted from GPT 3.5 to model human diseases and reveal potential therapeutic targets[wherein each token of the set of genetic embeddings]... [w]e identified genes whose in-silico deletion in the disease conditions could significantly shift the cell embeddings from the disease conditions to the non-failing (NF) or control condition[corresponds to a genetic edit of the plurality of genetic edits].”).8
Allowable Subject Matter
Claim 21 is objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ADAM C STANDKE whose telephone number is (571)270-1806. The examiner can normally be reached Gen. M-F 9-9PM EST.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Michael J Huntley can be reached at (303) 297-4307. 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.
/Adam C Standke/
Primary Examiner
Art Unit 2129
1 Examiner Remarks: The claim limitations that are not bolded and contained with square brackets i.e., [] are claim limitations that are not taught by the prior art of Zamft
2 Examiner Remarks: The claim limitations that are not bolded and contained with square brackets i.e., [] are claim limitations that are not taught by the prior art of Zamft
3 Examiner Remarks: According to the broadest reasonable interpretation (BRI), the use of alternative language amounts to the claim requiring one or more elements but not all.
4 Examiner Remarks: According to the broadest reasonable interpretation (BRI), the use of alternative language amounts to the claim requiring one or more elements but not all.
5 Examiner Remarks: According to the broadest reasonable interpretation (BRI), the use of alternative language amounts to the claim requiring one or more elements but not all.
6 Examiner Remarks: It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
7 Examiner Remarks: According to the broadest reasonable interpretation (BRI), the use of alternative language amounts to the claim requiring one or more elements but not all.
8 Examiner Remarks: It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Zamft with the above teachings of Liu for the same rationale stated at Claim 1.
Prosecution Insights