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 .
Response to Arguments
Applicant’s arguments with respect to claims 1-9 and 21-31 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 in the argument.
The joint end-to-end training of the first model and the ensemble model, while maintaining first model viability, amounts to significantly more than the abstract idea; and overcomes the 101 rejection.
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 1-9 and 21-31 are rejected under 35 U.S.C. 103 as being unpatentable over EnGRaiN: a supervised ensemble learning method for recovery of large-scale gene regulatory networks by Maneesha et al (from IDS filed 10/13/2023), uGLAD: Sparse graph recovery by optimizing deep unrolled networks by Shrivastava et al (the attached 5/23/2022 version, not the version filed with the IDS filed 10/13/2023) and Holistically-Nested Edge Detection by Xie et al.
EnGRaiN teaches claims 1, 21 and 27. A method for visualizing complex data relationships comprising:
receiving expression data; (EnGRaiN sec. 1 p. 1313 “Using EnGRaiN, we report the construction and analysis of a whole-genome ensemble network of the plant Arabidopsis thali ana, created from painstaking curation of heterogeneous microarray datasets from multiple public repositories.” The microarrays are the claimed expression data.)
providing the expression data to a first generator model that generates a first graph representing relationships in the expression data, wherein the first generator model is a trainable generator model that is and was previously trained with a generator model loss function; (EnGRaiN sec. 2.1.1 “Consider ‘M’ GRN predictions generated by as many distinct GRN recovery methods, each run independently of the others. GRNs may have edge weights, denoting the confidence level in each predicted edge.” The Gene Regulatory Network (GRN) recovery methods are each a generator model that generates a graph GRN.)
providing the expression data to a second generator model that, in parallel with the first generator model, generates a second graph representing relationships in the expression data; (EnGRaiN sec. 2.1.1 “Consider ‘M’ GRN predictions generated by as many distinct GRN recovery methods, each run independently of the others. GRNs may have edge weights, denoting the confidence level in each predicted edge.” The second model in the set of M models generates a second graph. See also EnGRaiN fig. 1 where there are several GRNs, below.)
PNG
media_image1.png
564
1032
media_image1.png
Greyscale
providing the first graph and the second graph to an ensemble model that is a machine learning model and creates consensus relationship data from the first graph and the second graph, wherein the ensemble model and the first generator model (EnGRaiN fig. 1 “Gene networks for each tissue and condition using 10 different network inference methods. These were then used as input to generate genome-scale ensemble networks using both unsupervised and supervised ensemble learning methods.” The ensemble methods is the ensemble model and the ensemble network is the consensus graph, see below.)
PNG
media_image2.png
564
1032
media_image2.png
Greyscale
generating a consensus graph having nodes and edges from the consensus relationship data; and (EnGRaiN fig. 1 “Gene networks for each tissue and condition using 10 different network inference methods. These were then used as input to generate genome-scale ensemble networks using both unsupervised and supervised ensemble learning methods.” The ensemble methods is the ensemble model and the ensemble network is the consensus graph, see below.)
PNG
media_image3.png
564
1032
media_image3.png
Greyscale
causing a display to render a visual representation of the consensus graph. (this claim element is missing from claim 27. EnGRaiN fig. 3 teaches this display below.)
PNG
media_image4.png
524
534
media_image4.png
Greyscale
EnGRaiN doesn’t teach end-to-end differentiable models nor joint training.
However, uGLAD teaches a trainable generator model that is end-to-end differentiable… (uGLAD sec. 3.2 “They leveraged the interpretable nature of the GLAD’s deep architecture to define the loss for training.” The uGLAD model is end-to-end differentiable, according to Applicant’s spec. and claim 4.) the ensemble model and the first generator model are jointly trained (uGLAD abs “we apply a deep model on X that outputs the precision matrix ˆ Θ, which can also be interpreted as the adjacency matrix.” uGLAD sec. 3.2 “each iteration of the model will output a valid precision matrix estimation and this allowed them to add auxiliary losses to regularize the intermediate results of GLAD…” uGLAD p. 7 point 3 “we are jointly optimizing over a batch input XK.”)
uGLAD, EnGRaiN and the claims all generate GRNs with gene expression data. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use and end-to-end differentiable model and joint training with auxiliary loss as a regularization term so that “the similarity among the tasks is automatically learned from data.” uGLAD sec. 4
uGLAD and EnGRaiN don’t teach the claimed end-to-end ensemble model loss function.
However, Xie teaches the ensemble model and the first generator model are jointly trained end-to-end with (Xie sec. 2 “By “holistically-nested”, we intend to emphasize that we are producing an end-to-end edge detection system, a strategy inspired by fully convolutional neural networks [26], but with additional deep supervision…”) an ensemble model loss function (Xie sec. 2.2 p. 4 “To directly utilize side-output predictions, we add a “weighted-fusion” layer to the network and (simultaneously) learn the fusion weight during training. Our loss function at the fusion layer Lfuse becomes…
PNG
media_image5.png
28
260
media_image5.png
Greyscale
…” Lfuse is the ensemble loss, the distance between side predictions (first model) and fused predictions (ensemble model)used to learn fusion weight and the deep supervision is the joint training. Xie sec. 2.2 p. 4 “Deep supervision is imposed at each side-output layer, guiding the side-outputs towards edge predictions with the characteristics we desire.”) to which the generator model loss function of the first generator model is added as a regularization term, Xie sec. 2.2 p. 4 “To directly utilize side-output predictions, we add a “weighted-fusion” layer to the network and (simultaneously) learn the fusion weight during training. Our loss function at the fusion layer Lfuse becomes…
PNG
media_image5.png
28
260
media_image5.png
Greyscale
…” Lfuse is the ensemble loss, the distance between side predictions (first model) and fused predictions (ensemble model)used to learn fusion weight and the deep supervision is the joint training. Xie sec. 2.2 p. 4 “Deep supervision is imposed at each side-output layer, guiding the side-outputs towards edge predictions with the characteristics we desire.” Equation 4 in sec. 2.2. p. 4 shows how side prediction loss is added to ensemble loss as a regularization term “
PNG
media_image6.png
42
320
media_image6.png
Greyscale
…”) and wherein after joint training, the first generator model retains its functionality as a stand-alone model and is able to generate a valid (Xie sec. 2.2 p. 4 “During testing, given image X, we obtain edge map predictions from both the side output layers and the weighted-fusion layer:…” Each side prediction from the side output layers is a valid output independent of the fused/ensemble output.)
Xie, EnGRaiN, uGLAD, and the claims are all ensemble networks. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to add predictor loss of the first model to regularize the ensemble loss while jointly training the first model and the ensemble model in order to keep the keep the first model on track and relevant and meaningful to the overall goal of the ensemble model. Xie accomplishes this because it “automatically learns rich hierarchical representations (guided by deep supervision on side responses)…” Xie abs. That guided deep supervision on the side/first model is important to the ensemble “improved speed…” Xie abs.
EnGRaiN teaches claims 2, 22 and 28. The method of claim 1, wherein the expression data is microarray data generated by single cell RNA sequencing or bulk sequencing. (EnGRaiN sec. 1 p. 1313 “Using EnGRaiN, we report the construction and analysis of a whole-genome ensemble network of the plant Arabidopsis thali ana, created from painstaking curation of heterogeneous microarray datasets from multiple public repositories.” The microarrays are the claimed expression data generated by bulk sequencing.)
EnGRaiN teaches claim 3. The method of claim 2, further comprising generating expression data with a microarray. (EnGRaiN sec. 1 p. 1313 “Using EnGRaiN, we report the construction and analysis of a whole-genome ensemble network of the plant Arabidopsis thali ana, created from painstaking curation of heterogeneous microarray datasets from multiple public repositories.”)
uGLAD teaches claims 4 and 23. The method of claim 1, wherein the first generator model is one of GLAD, uGLAD, Neural Graph Revealers (NGR), or GRNUlar. (uGLAD abs “uGLAD1 , builds upon and extends the state-of the-art model GLAD [42] to the unsupervised setting.”)
EnGRaiN teaches claims 5 and 24. The method of claim 1, wherein the second generator model is a fixed model that is not trainable. (EnGRaiN sec. 2.1.4 “In EnGRaiN, we do not modify the individual methods participating in the ensemble as done in a ‘joint training’ process (Cheng et al., 2016). Instead, all the networks in the input are created independently by the respective methods without the knowledge of each other.”)
EnGRaiN teaches claim 6. The method of claim 5, wherein the second generator model is one of GENIE3 or GRNBoost2. (EnGRaiN sec. 2.2.1 “all parallel methods… GENIE3, GRNBoost…”)
EnGRaiN teaches claims 7, 25 and 30. The method of claim 1, wherein the ensemble model learns a function for each edge in the consensus graph over the edges present in the first graph and the second graph. (EnGRaiN sec. 2.1.4 “Our ensemble model is able to learn a weighing function over the input methods by using the data of a few thousand edges. Second, this allows us to scale to millions of edges in an efficient manner, as each of these edge-wise predictions can be executed in parallel. Last, the edge-wise prediction approach also facilitates fast inclusion of additional GRN recovery methods into the EnGRaiN framework.”)
EnGRaiN teaches claims 8, 26 and 31. The method of claim 7, wherein the ensemble model is an edge-selector neural network. (EnGRaiN teaches an edge-selector neural network because Spec. para 61 says, “Thus, the ensemble model may be implemented as an edge-selector neural network. One example of a suitable ensemble model is EnGRaiN.”)
EnGRaiN teaches claim 9. The method of claim 1, wherein, in the consensus graph, the nodes represent genes and the edges represents a regulatory relationship between a first one of the genes and a second one of the genes, wherein the first one of the genes is a transcription factor gene. (EnGRaiN fig. 1 shows the graphs where nodes are genes, edges are regulatory relationships between genes, see below. EnGRaiN sec. 2.5 “It includes a total of 1359 non-redundant regulatory interactions between 388 transcription factors and target genes.”)
PNG
media_image7.png
320
1244
media_image7.png
Greyscale
uGLAD and EnGRaiN teach claim 29. (New) The computer-readable storage media of claim 27, wherein the first generator model is one of GLAD, uGLAD, Neural Graph Revealers (NGR), or GRNUlar (uGLAD abs “uGLAD1 , builds upon and extends the state-of the-art model GLAD [42] to the unsupervised setting.”) and wherein the second generator model is one of GENIE3 or GRNBoost2. (EnGRaiN sec. 2.2.1 “all parallel methods… GENIE3, GRNBoost…”)
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Austin Hicks whose telephone number is (571)270-3377. The examiner can normally be reached Monday - Thursday 8-4 PST.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/AUSTIN HICKS/Primary Examiner, Art Unit 2142
Prosecution Insights