System and method for deconvolution of expression data

Abstract

To provide a system and a method for deconvolution of expression data.SOLUTION: A method for determining one or more cell composition percentages from expression data includes obtaining expression data for a biological sample. The biological sample has previously been obtained from a subject, and the expression data includes first expression data associated with a first set of genes associated with a first cell type. The method also includes determining a first cell composition percentage for the first cell type using the expression data and one or more non-linear regression models including a first non-linear regression model. The first cell composition percentage indicates an estimated percentage of cells of the first cell type in the biological sample, and in the determining step, the first cell composition percentage for the first cell type is determined by processing the first expression data with the first non-linear regression model. The method further includes outputting the first cell composition percentage.SELECTED DRAWING: Figure 2A

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application is a joint venture between the US Provisional Patent Application No. 202002222222 and the Japanese Patent Application No. 202002222222. No. 63 / 108,262, filed on March 12, 2020, entitled "GENE EXPRESSION DATA" MACHINE LEARNING SYSTEMS AND METHODS FOR DECONVOLUTION OF GENE E 35 U.S.C. § 119(e) of U.S. Provisional Patent Application No. 62 / 988,700 entitled "XPRESSION DATA" No. 6,393,636, each of which is incorporated herein by reference in its entirety. Be absorbed. [Background technology]

[0002] Generally, a tumor mass (or other diseased tissue) contains a population of malignant cells (e.g., cancer cells) and a population of endothelial cells (e.g., endothelial cells) e.g. The microenvironment may include immune cells, fibroblasts, and extracellular matrix proteins. It is done. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] Newman et al., “Robust enumeration of cell subsets from tissue expression profiles,” Nat. Methods 12, pp. 453–457 (2015) [Non-Patent Document 2] Newman et al., “Determining cell type abundance and expression from bulk tissues with digital cytometry,” Nat Biotechnol 37, pp. 773–782 (2019) [Non-Patent Document 3] Finotello et al., “Molecular and pharmacological modulators of the tumor immune contexture revealed by deconvolution of RNA-seq data,” Genome Med 11, p. 34 (2019) [Non-Patent Document 4] Hao et al., “Fast and Robust Deconvolution of Tumor Infiltrating Lymphocytes from Expression Profiles using Least Trimmed Squares”, bioRxiv p. 358366; doi: https: / / doi.org / 10.1101 / 358366 [Non-Patent Document 5] Aran et al., “xCell: digitally portraying the tissue cellular heterogeneity landscape,” Genome Biol. 18, p. 220 (2017) [Non-Patent Document 6] Monaco et al., “RNA-Seq signatures normalized by mRNA abundance allow absolute deconvolution of human immune cell types,” Cell Rep. 26, pp. 1627–1640, e1627 (2019) [Non-Patent Document 7] Vaught et al., “Biospecimens and biorepositories: from afterthought to science,” Cancer Epidemiol Biomarkers Prev. 2012 Feb;21(2):253–5. [Non-Patent Document 8] Vaught and Henderson, “Biological sample collection, processing, storage and information management”, IARC Sci Publ. 2011;(163):pp. 23-42. [Non-Patent Document 9] Li et al., JCO Precis Oncol. 2018; 2: PO.17.00091 [Non-Patent Document 10] Consea et al., "A survey of best practices for RNA-seq data analysis," Genome Biology 201617:13 [Non-Patent Document 11] Pereira and Rueda (bioinformatics-core-shared-training.github.io / cruk-bioinf-sschool / Day2 / rnaSeq_align.pdf) [Non-Patent Document 12] Wagner et al., Theory Biosci. (2012) 131:281-285 Summary of the Invention [Means for solving the problem]

[0004] Some embodiments use at least one computer hardware processor obtaining expression data for a biological sample, the biological sample having been previously obtained from the subject; and the expression data comprises a first expression data set associated with a first set of genes associated with the first cell type. and performing one or more nonlinear regressions including the expression data and a first nonlinear regression model. determining a first cellular composition ratio for a first cell type using the regression model; The first cellular proportion indicates an estimated proportion of cells of a first cell type in the biological sample, The step of determining a first cellular constituent ratio for the type includes subjecting the first expression data to a first nonlinear regression model. determining a first cellular composition ratio for a first cell type; and The present invention provides a method for detecting a cell composition ratio of a subject, the method comprising the steps of:

[0005] Some embodiments include at least one hardware processor and at least one hardware When executed by a hardware processor, it causes at least one hardware processor obtaining expression data for the biological sample, the biological sample having been previously obtained from the subject; a first expression data set associated with a first set of genes associated with a first cell type; and performing one or more nonlinear regressions including the expression data and a first nonlinear regression model. determining a first cellular composition ratio for a first cell type using the regression model; The first cellular proportion indicates an estimated proportion of cells of a first cell type in the biological sample, The step of determining a first cell composition ratio for a cell type includes subjecting the first expression data to a first nonlinear regression. processing the model to determine a first cellular composition ratio for the first cell type; and and outputting the first cellular constituent ratio. and at least one non-transitory computer readable storage medium storing the Provide the system.

[0006] Some embodiments, when executed by at least one hardware processor, comprise: obtaining expression data for the biological sample on at least one hardware processor; wherein the biological sample has previously been obtained from the subject and the expression data relates to the first cell type. a step of obtaining first expression data relating to a first set of genes; and and one or more nonlinear regression models, including a linear regression model, for the first cell type. A step of determining a first cellular constituent ratio, the first cellular constituent ratio being a first cellular constituent ratio in a biological sample. Showing an estimated proportion of cells of one cell type and determining a first cellular composition ratio for a first cell type The step of processing the first expression data through a first nonlinear regression model to obtain an expression profile for the first cell type. determining a first cellular constituent ratio based on the determined cellular constituent ratio; and outputting the first cellular constituent ratio. at least one non-transitory computer storing processor-executable instructions for performing the steps of A computer readable storage medium is provided.

[0007] In some embodiments, the subject has, is suspected of having, or is a patient with cancer. There is a risk of this happening.

[0008] In some embodiments, the expression data is RNA expression data.

[0009] In some embodiments, processing the first expression data with a first nonlinear regression model providing first expression data as input to a first nonlinear regression model to generate a first expression vector from a first cell type; obtaining a corresponding output representing an estimated proportion of RNA from the first cell type; and determining a first cellular proportion for the first cell type based on the first cellular proportion.

[0010] In some embodiments, the expression data relates to a first set of genes associated with a first cell type. and a first nonlinear regression model using the first expression data as an input. and configured to generate a first value for the estimated proportion of RNA from the first cell type using the The first submodel was compared with the second expression data and the estimated proportion of RNA from the first cell type. Using the first value and as input, generate a second value for the estimated proportion of RNA from the first cell type. and a second sub-model configured to generate a

[0011] In some embodiments, the expression data is a second cell type associated with a second cell type that is different from the first cell type. and second expression data relating to the set of two genes, and comprises a second nonlinear regression model. Some embodiments are at least in part based on the second cell The second expression data is subjected to a second nonlinear regression model to determine a second cellular fraction for the type. determining a second cell composition ratio for a second cell type by treating the cells with a It further includes a process.

[0012] In some embodiments, the first type of cells is a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibronectin, or a fibronectin. A group consisting of fibroblasts, lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells are selected.

[0013] In some embodiments, the first expression data is for a first cell type in Table 2. The present invention includes expression data for at least 10 genes selected from the group of genes of

[0014] In some embodiments, the expression data comprises a set of genes associated with each of the multiple cell types. The plurality of gene sets includes expression data associated with a first gene set and a first cell. and a plurality of cell types, including a plurality of cell types, Some embodiments include a model using expression data associated with a plurality of gene sets. The method further comprises determining a plurality of cellular composition ratios for a plurality of cell types, In some embodiments, the step of determining a plurality of cellular constituent ratios comprises a first cellular constituent ratio. The process includes, for each cell type of the plurality of cell types, determining, at least in part, a cellular composition for the cell type. Using each of the nonlinear regression models to determine a ratio, By processing expression data related to sets of genes associated with cell types, The method includes a step of determining the composition ratio of each cell.

[0015] In some embodiments, the genes in the set of genes are selected from the set of genes in Table 2. The gene includes at least 25 genes selected from a group of genes, and determines a plurality of cell constituent ratios. The steps include processing expression data for at least 25 genes.

[0016] In some embodiments, the genes in the set of genes are selected from the set of genes in Table 2. The gene comprises at least 35 genes selected from a group of genes, and determines the composition ratio of multiple cells. The steps include processing expression data for at least 35 genes.

[0017] In some embodiments, the genes in the set of genes are selected from the set of genes in Table 2. The gene includes at least 50 genes selected from a group of genes, and determines a plurality of cell constituent ratios. The steps include processing expression data for at least 50 genes.

[0018] In some embodiments, the genes in the set of genes are selected from the set of genes in Table 2. The gene comprises at least 75 genes selected from a group of genes, and determines a plurality of cell constituent ratios. The steps include processing expression data for at least 75 genes.

[0019] In some embodiments, the genes in the set of genes are selected from the set of genes in Table 2. The method includes determining a plurality of cell constituent ratios, the method comprising: The step of processing comprises processing expression data for at least 100 genes.

[0020] In some embodiments, the one or more nonlinear regression models include one or more random forks. Includes rest regression models.

[0021] In some embodiments, the one or more nonlinear regression models include one or more neural networks. Includes network regression models.

[0022] In some embodiments, the one or more nonlinear regression models include one or more support vectors. Includes terminology regression models.

[0023] In some embodiments, the first nonlinear regression model simulates, at least in part, obtaining simulated expression data; and performing a first nonlinear and training a regression model.

[0024] Some embodiments include obtaining simulated expression data and and training a first nonlinear regression model using the expression data.

[0025] In some embodiments, obtaining the simulated expression data comprises generating simulated expression data, the generating the simulated expression data comprising: Obtaining a set of RNA expression data from one or more biological samples, The set includes microenvironment cell expression data and malignant cell expression data. using the cell expression data to generate simulated microenvironment cell expression data; and generating simulated malignant cell expression data using the malignant cell expression data. The process was performed by comparing the simulated microenvironment cell expression data with the simulated malignant cell expression data. and combining the expression data to generate at least a portion of the simulated expression data. Includes:

[0026] Some embodiments provide an expression profile for a first cell type and a determining a malignant tumor expression profile using the first cellular composition ratio; .

[0027] In some embodiments, the first nonlinear regression model is obtaining training data comprising: a first cell type; a first RNA expression data set for a first set of genes associated with the first cell; training a first nonlinear regression model to estimate the proportion of RNA from a cell type, The training step comprises training a first cell using a first nonlinear regression model and the first RNA expression data. generating an estimated proportion of RNA from the first cell type, and using the estimated proportion of RNA from the first cell type; updating parameters of the first nonlinear regression model based on the calculated values; So they are trained.

[0028] Some embodiments use at least one computer hardware processor obtaining RNA expression data for a biological sample, the biological sample having cancer; Previously obtained from subjects suspected of having or at risk of having cancer and A. Expression data includes first RNA expression data associated with a first set of genes associated with a first cell type. The first RNA expression data includes the gene expression data for the first cell type in Table 2. a first cell comprising expression data for at least 10 genes selected from the group of cells; The types are B cells, CD4+ T cells, CD8+ T cells, endothelial cells, fibroblasts, lymphocytes, and macrophages. a first RNAi gene, a first RNAi gene, a second RNAi gene, a third RNAi gene, a fourth RNAi gene, a fifth RNAi gene, a sixth ... determining a first cellular proportion for a first cell type using the expression data; The first cellular composition ratio indicates an estimated ratio of cells of a first cell type in the biological sample, and the first Determining a first cellular composition ratio for a cell type includes inputting the first cellular composition ratio into a first nonlinear regression model. to provide first RNA expression data representing an estimated proportion of RNA from the first cell type, obtaining a titer for the first cell type based on the estimated proportion of RNA from the first cell type; determining a first cellular composition ratio; and .

[0029] Some embodiments include at least one hardware processor and at least one hardware When executed by a hardware processor, it causes at least one hardware processor obtaining RNA expression data for a biological sample, the biological sample having cancer; Previously obtained from subjects suspected of having or at risk of having cancer and A. Expression data includes first RNA expression data associated with a first set of genes associated with a first cell type. The first RNA expression data includes the gene expression data for the first cell type in Table 2. a first cell comprising expression data for at least 10 genes selected from the group of cells; The types are B cells, CD4+ T cells, CD8+ T cells, endothelial cells, fibroblasts, lymphocytes, and macrophages. a first RNAi gene, a first RNAi gene, a second RNAi gene, a third RNAi gene, a fourth RNAi gene, a fifth RNAi gene, a sixth ... determining a first cellular proportion for a first cell type using the expression data; The first cellular composition ratio indicates an estimated ratio of cells of a first cell type in the biological sample, and the first Determining a first cellular composition ratio for a cell type includes inputting the first cellular composition ratio into a first nonlinear regression model. to provide first RNA expression data representing an estimated proportion of RNA from the first cell type, obtaining a titer for the first cell type based on the estimated proportion of RNA from the first cell type; determining a first cellular composition ratio. and at least one non-transitory computer readable storage medium storing the Provide the system.

[0030] Some embodiments, when executed by at least one hardware processor, comprise: obtaining RNA expression data for the biological sample on at least one hardware processor; wherein the biological sample is from a subject having, suspected of having, or at risk of having cancer. The RNA expression data is previously obtained from a subject having a first cell type, and the RNA expression data is associated with a first cell type. The method includes: providing a first RNA expression data set associated with a set of genes; the first RNA expression data set is shown in Table 2 ( At least 10 genes selected from the group of genes for the first cell type in Table 2 The first cell type includes expression data for B cells, CD4+ T cells, CD8+ T cells, endothelial cells, It consists of cells, fibroblasts, lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells. and using the first RNA expression data to generate a first RNA expression profile for the first cell type. The first cellular composition ratio is a first cellular composition ratio in a biological sample. Determining a first cellular composition ratio for a first cell type, indicating an estimated proportion of cells of the cell type. providing the first RNA expression data as input to a first nonlinear regression model to estimate the expression of a first cell type; obtaining a corresponding output representing an estimated proportion of RNA from the first cell type; and determining a first cellular composition ratio for the first cell type based on the ratio; at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium is provided.

[0031] In some embodiments, the RNA expression data comprises a first set of genes associated with a first cell type. and a first nonlinear regression model for the first RNA expression data. as input to generate a first value for the estimated proportion of RNA from a first cell type. The first submodel was constructed as follows: Using the first values ​​for and as inputs, and a second sub-model configured to generate a second value.

[0032] In some embodiments, the RNA expression data comprises a second set of genes associated with a second cell type. The second RNA expression data includes a second RNA expression data related to the first RNA expression data in Table 2. Expression of at least 10 genes selected from a group of genes for two cell types. The second cell type includes B cells, CD4+ T cells, CD8+ T cells, endothelial cells, and fibroblasts. , lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells. In some embodiments, determining the second cellular proportion for the second cell type comprises: The second RNA expression data is then analyzed to determine a second cellular fraction for the second cell type. Processing with a non-linear regression model is included.

[0033] In some embodiments, the RNA expression data comprises a plurality of genes associated with each of the plurality of cell types. The plurality of gene sets includes RNA expression data associated with the plurality of gene sets, the plurality of gene sets including a first gene set and a second gene set. In some embodiments, the method includes the step of: Using RNA expression data to determine multiple cellular constituent ratios for multiple cell types. In some embodiments, the method further comprises the steps of: The step of determining the plurality of cellular composition ratios includes determining at least one cell type for each of the plurality of cell types. In the first section, we used each nonlinear regression model to determine the cellular composition ratio for each cell type. by processing RNA expression data related to sets of genes associated with cell types. The method includes determining the respective cellular composition ratios for each cell type.

[0034] In some embodiments, the first nonlinear regression model comprises a random forest regression model. nothing.

[0035] In some embodiments, the first nonlinear regression model is a neural network regression model. Includes.

[0036] In some embodiments, the first nonlinear regression model is a support vector machine regression model. Includes.

[0037] In some embodiments, the first nonlinear regression model simulates, at least in part, The method is trained by generating training data including RNA expression data. In embodiments, generating training data comprises collecting RNA expression data from one or more biological samples. obtaining a set of RNA expression data, the set of RNA expression data being comprised of microenvironment cell RNA expression data and and a step of detecting the RNA expression of malignant cells using the RNA expression data of microenvironment cells. generating simulated microenvironment cell RNA expression data; and generating simulated malignant cell RNA expression data using the simulated We combined the RNA expression data of simulated microenvironment cells with the RNA expression data of simulated malignant cells. and generating at least a portion of the simulated RNA expression data. .

[0038] Some embodiments provide an RNA expression profile for the first cell type and a method for detecting the presence of a marker in the first cell type. determining a malignant tumor expression profile using the first cellular proportion in the include.

[0039] In some embodiments, the first RNA expression data is selected from the group of genes in Table 2. The present invention includes expression data for at least 25 genes selected from the above-mentioned genome-wide database.

[0040] In some embodiments, the first RNA expression data is selected from the group of genes in Table 2. The present invention includes expression data for at least 50 genes selected from the above-mentioned genome-wide database.

[0041] In some embodiments, the first RNA expression data is selected from the group of genes in Table 2. The present invention includes expression data for at least 100 genes selected from the above-mentioned genome-wide database.

[0042] In some embodiments, the first nonlinear regression model is obtaining training data comprising: a first cell type; and a second RNA expression data for a first set of genes associated with the first cell. training a first nonlinear regression model to estimate the proportion of RNA from a cell type, The training step comprises training a first nonlinear regression model using the first nonlinear regression model and the second RNA expression data. generating an estimated proportion of RNA from the first cell type, and using the estimated proportion of RNA from the first cell type; updating parameters of the first nonlinear regression model based on the first nonlinear regression model. So they are trained.

[0043] Some embodiments use at least one computer hardware processor obtaining training data comprising simulated RNA expression data, The collected RNA expression data includes a first RNA expression vector for a first gene associated with a first cell type. The data and a second RNA for a second gene associated with a second cell type different from the first cell type. A method for estimating the proportion of RNA from each of one or more cell types, comprising: training a plurality of nonlinear regression models to determine the first A nonlinear regression model to estimate the proportion of RNA from one cell type and from a second cell type A second nonlinear regression model was used to estimate the ratio of RNA to RNA. The training step includes, at least in part, training a first nonlinear regression model and a first RNA expression data. generating an estimated proportion of RNA from the first cell type using the Update the parameters of the first nonlinear regression model using the estimated proportion of RNA in training a first nonlinear regression model by a step of: and outputting the trained multiple nonlinear regression models including the first nonlinear regression model and the second nonlinear regression model. The present invention provides a method comprising the steps of:

[0044] Some embodiments include at least one computer hardware processor and at least and when executed by a computer hardware processor, Training a computer hardware processor with simulated RNA expression data Obtaining data, the simulated RNA expression data being associated with the first cell type. and a second cell type different from the first cell type. and further comprising: a step of: receiving second RNA expression data for a second gene of interest; and training a plurality of nonlinear regression models to estimate the proportions of RNA from the cell types; Thus, the multiple nonlinear regression models are used to estimate the proportion of RNA from a first cell type. A nonlinear regression model and a second nonlinear regression model to estimate the proportion of RNA from a second cell type. and training a plurality of nonlinear regression models includes, at least in part, Using the linear regression model and the first RNA expression data, estimate the proportion of RNA from the first cell type. generating a first nonlinear regression model using the estimated proportion of RNA from the first cell type; Training a first nonlinear regression model by updating the parameters of the model. a step of: and outputting the plurality of nonlinear regression models. and at least one non-transitory computer-readable storage medium for provide.

[0045] Some embodiments are implemented by at least one computer hardware processor. When executed, the simulated obtaining training data including simulated RNA expression data, The data includes first RNA expression data for a first gene associated with a first cell type and a first cell type. and second RNA expression data for a second gene associated with a second cell type distinct from the cell type. and performing a plurality of nonlinear iterative processes to estimate the proportion of RNA from each of one or more cell types. training a regression model, the multiple nonlinear regression models being adapted to estimate the RNA sequence from the first cell type; A first nonlinear regression model for estimating the proportion of RNA from a second cell type. and training a plurality of nonlinear regression models includes a second nonlinear regression model for training the plurality of nonlinear regression models. A first cell is analyzed using, at least in part, the first nonlinear regression model and the first RNA expression data. generating an estimated proportion of RNA from the first cell type, and using the estimated proportion of RNA from the first cell type; updating parameters of the first nonlinear regression model using the training a nonlinear regression model; and and outputting a plurality of trained nonlinear regression models, including a linear regression model. at least one non-transitory computer readable memory device storing processor executable instructions; Provide a storage medium.

[0046] In some embodiments, obtaining training data comprises extracting RNA expression data from one or more biological samples. obtaining a set of RNA expression data, the set of RNA expression data being microenvironment cell RNA expression data; and based on the microenvironment cell RNA expression data, Obtaining simulated microenvironment cell RNA expression data and analyzing the malignant cell RNA expression data. obtaining simulated malignant cell RNA expression data based on the simulated We combined the RNA expression data of the microenvironment cells with the RNA expression data of the simulated malignant cells. and generating at least a portion of the simulated RNA expression data. Generating at least a portion of the simulated RNA expression data.

[0047] Some embodiments may further comprise: The method further comprises adding noise to the expression data.

[0048] In some embodiments, the noise is at least one of Poisson noise or Gaussian noise. Includes 1.

[0049] In some embodiments, generating simulated microenvironment cellular RNA expression data using a first portion of the microenvironment cell RNA expression data for a first microenvironment cell type , generating a first RNA expression profile.

[0050] In some embodiments, the first portion of the microenvironment cell RNA expression data comprises a first microenvironment cell It contains RNA expression data from multiple subtypes of HIV-1.

[0051] In some embodiments, generating a first RNA expression profile comprises: Resample the first portion of the microenvironment cell RNA expression data using multiple subtypes of cell types The method includes a ringing step.

[0052] In some embodiments, the first portion of the microenvironment cellular RNA expression data comprises RNA expression data from a plurality of samples. A expression data included.

[0053] In some embodiments, generating a first RNA expression profile comprises multiple samples. We take as input several samples to generate the first part of the microenvironment cell RNA expression data. The method includes the step of resampling.

[0054] In some embodiments, generating simulated microenvironment cellular RNA expression data using a second portion of the microenvironment cell RNA expression data for a second microenvironment cell type generating a second RNA expression profile; and comparing the first RNA expression profile with the second RNA expression profile. The small amount of simulated microenvironment cellular RNA expression data was combined with the current profiles. and generating at least some of the

[0055] In some embodiments, the first RNA expression profile and the second RNA expression profile are combined. Generate at least some of the simulated microenvironment cellular RNA expression data together determining a weighted sum of the first RNA expression profile and the second RNA expression profile. The process includes:

[0056] In some embodiments, the malignant cell RNA expression data comprises RNA expression data from a plurality of malignant cell samples. Includes data.

[0057] In some embodiments, generating the simulated malignant cell RNA expression data comprises: The method includes combining RNA expression data from multiple malignant cell samples.

[0058] In some embodiments, generating the simulated malignant cell RNA expression data comprises: Adding noise to the simulated malignant cell RNA expression data.

[0059] In some embodiments, the coefficients of the weighted sum are determined using the output of a previously trained nonlinear regression model. It is determined using

[0060] In some embodiments, the first RNA expression data is for a first cell type in Table 2. The present invention includes expression data for at least 10 genes selected from the group of genes in

[0061] In some embodiments, the second RNA expression data is for a second cell type in Table 2. The present invention includes expression data for at least 10 genes selected from the group of genes in

[0062] In some embodiments, the first cell type and the second cell type are B cells, CD4+ T cells, CD8+ T cells, alveoli, endothelial cells, fibroblasts, lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T The cell is selected from the group consisting of:

[0063] In some embodiments, the simulated RNA expression data is a first RNA expression data set associated with a first cell type. The second RNA expression data for the first gene is included.

[0064] In some embodiments, a first nonlinear regression model of the one or more nonlinear regression models is a Using the RNA expression data of one as input, a first analysis is performed on the estimated proportion of RNA from the first cell type. a first sub-model configured to generate values ​​of , using as input the first value for the estimated proportion of RNA from the first cell type, R A second sub-model configured to generate a second value for the estimated proportion of NA.

[0065] In some embodiments, the second sub-model comprises a plurality of cell types other than the first cell type, each of the plurality of cell types. for the estimated proportion of RNA from the first cell type, using the estimated proportion of RNA from The second value is further configured to generate a second value of

[0066] Some embodiments use at least one computer hardware processor obtaining expression data for a biological sample, the biological sample comprising a cancer sample; previously obtained from a subject suspected of having or at risk of having cancer; obtaining a plurality of expression profiles for a corresponding plurality of cell types, Each of the files contains one or more genes associated with each cell type from the plurality of cell types. and, at least in part, combining the expression data and the plurality of expression data from the children. A piecewise continuous error function between the current profile and the plurality of cells is optimized. and determining a plurality of cellular constituent ratios for the type.

[0067] Some embodiments include at least one computer hardware processor and at least and when executed by a computer hardware processor, obtaining expression data for the biological sample on a computer hardware processor; Thus, the biological sample may be collected from a subject having, suspected of having, or at risk of having cancer. The method includes the steps of: obtaining a file, each of the expression profiles being from each of a plurality of cell types; and including at least one step of collecting expression data from one or more genes associated with the cell type. At least in part, a piecewise continuous error function between the expression data and the multiple expression profiles. By optimizing the ratio of multiple cell types, multiple cell composition ratios are determined. A process of determining at least one computer readable memory storing processor executable instructions for causing the and a storage medium capable of storing the stored information.

[0068] Some embodiments are implemented by at least one computer hardware processor. When the biological sample is processed, the biological sample is processed by the at least one computer hardware processor. obtaining expression data from a biological sample having or suspected of having cancer; or a corresponding plurality of samples previously obtained from a subject at risk of having cancer. Obtaining a plurality of expression profiles for the cell type, each of the expression profiles comprising This involves the identification of individual expression vectors from one or more genes associated with each cell type from a plurality of cell types. and, at least in part, the expression data and a plurality of expression profiles. A piecewise continuous error function between determining a cellular composition ratio; Another computer-readable storage medium is provided.

[0069] In some embodiments, the expression data is RNA expression data, and the plurality of expression profiles is NA expression profile.

[0070] In some embodiments, determining a plurality of cellular constituent ratios for a plurality of cell types comprises determining a plurality of cellular constituent ratios for a plurality of cell types. determining a weighted sum of the difference values, the error values ​​being determined using a piecewise continuous error function. can be.

[0071] In some embodiments, determining a plurality of cellular constituent ratios for a plurality of cell types comprises determining a plurality of cellular constituent ratios for a plurality of cell types. Minimizing a weighted sum of the difference values.

[0072] In some embodiments, the one or more genes are selected from the group consisting of less than 5000 genes in Table 2. It is a gene that consists of at least two genes.

[0073] Some embodiments provide a plurality of expression profiles for a corresponding plurality of cell types and The method further comprises using the cellular constituent ratios of said cells to determine a malignant tumor expression profile. [Brief description of the drawings]

[0074] [Figure 1A] FIG. 1A is a diagram depicting a system for determining cellular constituent ratios based on expression data, according to some embodiments of the technology described herein. [Figure 1B] FIG. 1B is a diagrammatic example for determining various cellular constituent ratios of various cell types and cell subtypes using a nonlinear regression model of each respective cell type and cell subtype according to some embodiments of the technology described herein. [Figure 1C] FIG. 1C shows a t-SNE visualization depicting an exemplary cell population including malignant and microenvironmental cells, according to some embodiments of the techniques described herein. [Figure 1D] FIG. 1D illustrates a t-SNE visualization depicting an exemplary malignant cell population in accordance with some embodiments of the techniques described herein. [Figure 1E] FIG. 1E is a chart depicting exemplary gene expression of various cells according to some embodiments of the technology described herein. [Figure 1F] FIG. IF is a chart depicting exemplary correlations between genes and selected cell percentages in a sample mixture of diverse cell types, according to some embodiments of the technology described herein. [Figure 1G-1] FIG. 1G is a chart depicting exemplary gene expression of tumor cell lines, according to some embodiments of the technology described herein. [Figure 1G-2] FIG. 1G is a chart depicting exemplary gene expression of tumor cell lines, according to some embodiments of the technology described herein. [Figure 1G-3]FIG. 1G is a chart depicting exemplary gene expression of tumor cell lines, according to some embodiments of the technology described herein. [Figure 1G-4] FIG. 1G is a chart depicting exemplary gene expression of tumor cell lines, according to some embodiments of the technology described herein. [Figure 2A] FIG. 2A is a flow chart depicting an exemplary non-linear method for determining cellular constituent ratios based on expression data, according to some embodiments of the technology described herein. [Figure 2B] FIG. 2B is a flowchart illustrating an example implementation of a method 200 for determining cellular composition ratios based on expression data, according to some embodiments of the technology described herein. [Figure 2C] FIG. 2C is a flowchart illustrating an example implementation of act 216a of method 200 according to some of the embodiments of the technology described herein. [Figure 3A] FIG. 3A depicts the use of machine learning methods to determine RNA ratios based on RNA expression data, in accordance with some embodiments of the technology described herein. [Figure 3B] FIG. 3B is a diagram depicting the use of a nonlinear regression model including sub-models to determine RNA ratios based on RNA expression data, according to some embodiments of the technology described herein. [Figure 3C] FIG. 3C depicts a method for determining cellular constituent ratios based on RNA ratios according to some embodiments of the technology described herein. [Figure 3D] FIG. 3D depicts an example method for determining a malignant tumor expression profile based on cellular composition ratios, according to some embodiments of the technology described herein. [Figure 4] FIG. 4 is a flowchart depicting an exemplary method for training one or more nonlinear regression models for determining cellular constituent proportions based on RNA expression data, according to some embodiments of the technology described herein. [Figure 5A]FIG. 5A depicts an example method for training one or more machine learning models including validation and multi-stage training in accordance with some embodiments of the techniques described herein. [Figure 5B-1] FIG. 5B is a diagram depicting an example method for training one or more machine learning models including validation and multi-stage training in accordance with some embodiments of the techniques described herein. [Figure 5B-2] FIG. 5B is a diagram depicting an example method for training one or more machine learning models including validation and multi-stage training in accordance with some embodiments of the techniques described herein. [Figure 6A] FIG. 6A depicts an exemplary method for training one or more nonlinear regression models that includes generating simulated RNA expression data, according to some embodiments of the technology described herein. [Figure 6B] FIG. 6B is an exemplary diagram for generating artificial mixtures of RNA expression data to mimic real tissues, according to some embodiments of the techniques described herein. [Figure 6C] FIG. 6C is an exemplary diagram for generating and using artificial mixtures to train cell-type models, according to some embodiments of the techniques described herein. [Figure 6D] FIG. 6D is an exemplary illustration for generating specific artificial mixtures for training particular cell type / subtype models, according to some embodiments of the techniques described herein. [Figure 6E] FIG. 6E is an exemplary illustration for generating specific artificial mixtures for training particular cell type / subtype models, according to some embodiments of the techniques described herein. [Figure 6F] FIG. 6F is an example diagram illustrating a technique for processing a dataset and generating an artificial mixture according to some embodiments of the techniques described herein. [Figure 7A]FIG. 7A is a chart comparing simulated RNA expression data with RNA expression data from a biological sample, according to some embodiments of the technology described herein. [Figure 7B] FIG. 7B is a chart depicting exemplary cellular constituent ratios predicted according to a deconvolution technique developed by the inventors and the corresponding true cellular constituent ratios, according to some embodiments of the techniques described herein. [Figure 7C] FIG. 7C is a chart comparing an exemplary predictive accuracy of a deconvolution technique developed by the inventors against the predictive accuracy of alternative algorithms, in accordance with some embodiments of the techniques described herein. [Figure 7D] FIG. 7D is a chart comparing an exemplary predictive accuracy of a deconvolution technique developed by the inventors against the predictive accuracy of alternative algorithms, in accordance with some embodiments of the techniques described herein. [Figure 7E] FIG. 7E is a diagram depicting the expression of four selected genes in normal tissues, immune cell types, and cancerous tissues, according to some embodiments of the technology described herein. [Figure 7F] FIG. 7F is a chart depicting an exemplary prediction specificity of a deconvolution technique developed by the present inventors, in accordance with some embodiments of the techniques described herein. [Figure 7G] FIG. 7G is a chart comparing an exemplary non-specificity score of a deconvolution technique developed by the inventors against the non-specificity scores of alternative algorithms, according to some embodiments of the techniques described herein. [Figure 8] FIG. 8 is a flow chart depicting an exemplary linear method for determining cellular constituent ratios based on RNA expression data, according to some embodiments of the technology described herein. [Figure 9A-1] FIG. 9A depicts an exemplary RNA expression profile and global RNA expression data according to some embodiments of the technology described herein. [Figure 9A-2] FIG. 9A depicts an exemplary RNA expression profile and global RNA expression data according to some embodiments of the technology described herein. [Figure 9A-3] FIG. 9A depicts an exemplary RNA expression profile and global RNA expression data according to some embodiments of the technology described herein. [Figure 9A-4] FIG. 9A depicts an exemplary RNA expression profile and global RNA expression data according to some embodiments of the technology described herein. [Figure 9B] FIG. 9B is a diagram illustrating an example piecewise continuous error function in accordance with some embodiments of the techniques described herein. [Figure 10] FIG. 10 depicts an illustrative implementation of a computer system that may be used in connection with some embodiments of the technology described herein. [Figure 11] FIG. 11 is a block diagram of an illustrative environment in which one or more embodiments of the techniques described herein may be implemented. [Figure 12A] FIG. 12A is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12B] FIG. 12B is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12C] FIG. 12C is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12D] FIG. 12D is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12E]FIG. 12E is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12F] FIG. 12F is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12G] FIG. 12G is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12H] FIG. 12H is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12I] FIG. 12I is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12J] FIG. 12J is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 12K] FIG. 12K is a chart and graphs representing the analysis and results from an experiment to establish RNA transcript normalization and analyze sequencing technical noise, as described in connection with Example 1. [Figure 13A] FIG. 13A is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13B] FIG. 13B is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13C]FIG. 13C is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13D-1] FIG. 13D is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13D-2] FIG. 13D is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13E-1] FIG. 13E is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13E-2] FIG. 13E is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13F] FIG. 13F is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13G] FIG. 13G is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13H] FIG. 13H is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13I] FIG. 13I is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 13J]FIG. 13J is a chart and graphs representing the analysis and results from an experiment to deconvolute RNA-seq of multiple normal and cancer tissues, as described in connection with Example 2. [Figure 14A] FIG. 14A is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14B] FIG. 14B is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14C] FIG. 14C is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14D] FIG. 14D is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14E] FIG. 14E is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14F] FIG. 14F is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 14G] FIG. 14G is a chart and graphs representing the analysis and results from an experiment to deconvolute single cell RNA-seq data and bulk RNA-seq of blood, as described in connection with Example 3. [Figure 15A]FIG. 15A is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15B] FIG. 15B is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15C] FIG. 15C is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15D] FIG. 15D is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15E] FIG. 15E is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15F] FIG. 15F is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15G] FIG. 15G is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15H] FIG. 15H is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. [Figure 15I] FIG. 15I is a chart and graphs representing the analysis and results from an experiment deconvolving several different cancer tissues, as described in connection with Example 4. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0075] The present inventors have demonstrated that RNA expression data (e.g., biological samples subjected to sequencing techniques, e.g., Based on the data collected by RNA sequencing, The cellular composition (e.g., a specific proportion) of a sample (e.g., a sample from a tumor or other diseased tissue) In some embodiments, the researchers developed a machine learning approach to determine the proportion of each type of cell in the population. The step of determining the cellular composition ratio for one or more cell types may include one or more nonlinear iterative processes. The method may include using a regression model to estimate the respective cellular constituent proportions for the cell types. Linear regression models can be used to assess the efficacy and safety of the methods described herein, for example, for various malignant and / or micromal lesions. Combining RNA expression data for environmental cell types and / or The method includes the steps of: The method can be trained using simulated RNA expression data, which can be generated by .

[0076] The inventors have demonstrated that the tumor microenvironment (TME) plays a key role in disease progression (e.g., whether the tumor is eradicated or metastasized). Recognize and understand that the effects of steroids may play an important role in determining whether or not a drug is effective and in determining treatment response / resistance. For example, as recognized and understood by the inventors, the immune and non-immune components of the TME The components are involved in cell-cell contact and a variety of different molecular signals, such as growth factors and cytokines. Furthermore, the present inventors have used TME to investigate the mechanisms of tumor survival, maintenance, growth and development. mediates tumor survival by controlling the host immune system and provides tumor immunosurveillance. For this reason, the inventors have sought to understand the quantity and functionality of TME components. is essential to cancer research and is important for understanding treatment and its clinical impact. However, despite the importance of understanding TME components, existing cancer research The study was limited to a limited set of cellular constituents of the TME due to limitations of conventional methods for analyzing TME components. We have focused only on the elements of immunohistochemistry, flow cytometry, and CyTO. Techniques such as F rely on the availability of target-specific antibodies and unique tags, e.g., fluorescent dyes. There are limits to what can be done.

[0077] The inventors further demonstrated that the method can simultaneously provide information on tens of thousands of genes in a biological sample. Bulk RNA sequencing (RNA-seq) provides signals that represent the combined contributions of multiple cell types However, the inventors recognize and understand that this method enables the detection of all RNA of this kind. AExpression data do not provide information about the origin of individual RNA molecules and therefore the bulk RNA- Many challenges remain to determine the cellular composition (e.g., cellular composition ratios) of the TME from seq. We recognize that the process of determining cellular constituent ratios from RNA expression data is described herein. This is sometimes called "deconvolution".

[0078] The inventors believe that one of the key issues in cellular deconvolution is the identification and characterization of tumors and their microenvironments. Many genes can be expressed simultaneously by several types of cells present in the environment This is true for closely related cell types (e.g., T cells). specific cell types that may be considered subtypes (e.g., CD4+ and CD8+ T cell subtypes) This poses a particular challenge in identifying closely related cell types, as genetic markers between these types are often lacking. In some embodiments, the cells may be the same or similar. A type can be thought of as a population of cells with a distinct expression profile. For example, CD4+ T cells CD8+ T cells, and NK cells were analyzed using metabolic, signaling, and surface markers. The expression of a significant number of structural and regulatory genes, including It appears that globulin is specifically expressed by mature dendritic cells and macrophages. These cells express various differentiation genes at low levels. Therefore, we believe that the RNA expression data are consistent. Recognize and understand that the present invention may include both specific marker genes and genes associated with cell lineage. In addition, the ratio of markers to lineage-specific gene expression provides insight into cell subtypes. The inventors recognize that information may or may not be available. (For example, the CD4 / CD3D gene ratio may be a marker for CD4+ T cells, but CD3D is not a marker for hematopoietic (It is not a specific marker for any subtype of primary T cell.) Different types of cells may Even if these factors are closely related, their effects on tumor pathogenesis may differ greatly. However, the inventors have found that it is not possible to distinguish cell populations, even between closely related cell types. I still believe it is important.

[0079] Another challenge of cellular deconvolution that we recognize is the number of cells and their Difficulty in distinguishing between conditions such as cell type specific or semi-specific genetic Expression of the progeny may vary depending on the activation state of the cells of that type, or may be dependent on the subtype of that type. Multiple searches can determine the sequence of similar cell subtypes. However, they may be captured in different biological states. The authors conclude that variation in biological conditions is important in deriving accurate estimates of cellular composition. recognize and understand that

[0080] Furthermore, the inventors have demonstrated that the tumor microenvironment represents a relatively small proportion of the tumor as a whole. We recognize and understand that it may be difficult to extract small cell populations from bulk RNA-seq data. Identification of groups can be particularly challenging due to low signal-to-noise ratios. They found that even small cell populations can have a large impact on response to therapy. We acknowledge that it remains important to identify changes in small cell populations (e.g., NK cells). Furthermore, the value of RNA expression of a gene depends on the specific measurement technique, library, and - Preparation protocols and RNA enrichment methods (e.g., total RNA-seq (REF), polyA enrichment (REF), exon enrichment (EXO) The inventors recognize and understand that the use of scRNA sequences in a multi-target assay can be highly dependent on 3' scRNA capture or 3' scRNA-seq (REF). Even with the use of techniques such as single-cell RNA-seq (scRNA-seq), the coverage of such techniques is limited. In this case, it is generally not possible to extract marker genes useful for identifying cell types.

[0081] Therefore, the present inventors have developed an accurate and robust cellular We recognize the need for deconvolution techniques. We are applying machine learning techniques to estimate cellular constituent ratios based on data (e.g., RNA expression data). In some embodiments, novel systems and methods for using a biological sample from a subject have been developed. obtaining expression data (e.g., bulk RNA-seq data) for one or more cells; Type (e.g., B cells, CD4+ T cells, CD8+ T cells, endothelial cells, fibroblasts, lymphocytes, macrophages, etc.) The method includes a step of determining the cell composition ratios of the following cells: phage, monocytes, NK cells, neutrophils, and T cells. The cellular composition ratio is determined by determining the ratio of each specific cellular component in a biological sample. According to some embodiments, the estimated proportion of cells of a particular cell type may be shown. The step of determining the proportion of members may involve determining a set of genes associated with the cell type (e.g., one or more marker genes, which may be type-specific or semi-specific genes obtaining expression data for the specific cell types and determining the cellular composition of the specific cell types; Processing the current data with a non-linear regression model. This process can be repeated multiple times to achieve deconvolution across multiple cell types. Each of the cell types (which may include subtypes of the cell types, as described herein) The present invention can be implemented in an iterative or parallel manner with respect to at least FIG. As described herein, these techniques represent a significant improvement over the prior art.

[0082] In some embodiments, the machine learning approach used to determine the cellular composition is Multiple nonlinear models, each trained to determine the cellular composition of each specific cell type, were used. In some embodiments, the non-linear regression model may include using a multi-linear regression model. of parameters (e.g., thousands, tens of thousands, hundreds of thousands, at least a million, millions, tens of millions, or The nonlinear regression model may have a number of parameters (hundreds of millions of parameters), and the step of training the nonlinear regression model may include Such parameters can be calculated by computer from simulated expression data. In some embodiments, the simulated training data may include estimating values ​​of The generating step may include generating multiple training sets (e.g., For example, at least 25,000, at least 50,000, at least 100,000, at least 150 ,000, at least 200,000, at least 500,000, etc. In some embodiments, multiple nonlinear regression models are performed on multiple cell types (e.g., at least five, at least 10, at least 20, at least 30, at least 40, etc.) This can be trained.

[0083] The techniques described herein and developed by the inventors use machine learning techniques. This allows for a more robust computational approach to determine cell composition than conventional methods. This approach also provides significant improvements in performance, accuracy, and efficiency. For example, Figures 7C and 7D show Compared with the method, the nonlinear deconvolution method developed by the present inventors (e.g. In the study, called "Kassandra," the researchers were able to detect the presence of cancer cell overexpression noise. , which has been shown to provide more accurate predictions of cellular composition for different cell types ( For example, as shown in FIG. 7D. As a result, the techniques described herein allow Specifically, the techniques described herein improve the overall format of cell structure. The present invention provides an improved method for determining the composition ratio (e.g., for cell populations specifically within the tumor microenvironment). This provides improved insight into clinical decision making and understanding of tumor pathogenesis.

[0084] For example, unlike conventional approaches, the machine learning approaches described herein can Expression data relating to genes associated with (e.g., specific and / or semi-specific) types of By using them as inputs into a nonlinear regression model trained specifically for that subtype This allows successful identification of dependencies and interrelationships between genes in phenotypically closely related cell types. This allows accurate detection of cell subtypes even with similar expression patterns. We use training data that mimics the cellular complexity and diversity of tumor biopsies (Figures 7A and 7B). By using the expression profiles and unique characteristics of cell population markers, Thus, the nonlinear deconvolution techniques described herein are more efficient than previous algorithms. It is more robust and shows more consistent accuracy across different cell types / subtypes, provides significantly more accurate results than conventional methods on challenging and noisy data (Figure 7C, 7D, 13F, 15G). In the context of the tumor microenvironment (e.g., in the clinical setting of patients) These more accurate results could lead to improved cancer diagnosis and prognosis, as well as personalized treatment for patients. enabling more tailored treatment options.

[0085] One of the factors that contribute to the accuracy and robustness of the approach developed by the present inventors is The above aspects are particularly relevant for each respective cell type to determine the corresponding cellular composition ratios. One approach is to use expression data. For example, for a given cell type, the expression data may be: The expression data may include data relating to specific genes associated with that given cell type. In the form of at least as described herein with respect to Figures 1D-1E and Table 2 The expression data may include expression data relating to genes for a given cell type. As described herein, the step of identifying genes associated with a particular cell type can include identifying a particular Identifying genes that are expressed only or predominantly in a cell type or subtype To achieve this, the sequences are obtained from multiple databases and / or using a variety of sequencing techniques. This may include processing expression data from multiple samples, such that any particular associated with a particular cell type, regardless of how the gene is determined for that cell type The use of expression data associated with specific genes is a novel method for detecting cellular markers developed by the present inventors. Cellular deconvolution techniques allow researchers to determine which genes are expressed by which cell types This allows for the exploitation of domain-specific knowledge about Contribute to success.

[0086] Another aspect of the approach developed by the inventors that contributes to its performance is the The architecture employed for both the training described in the paper and the use of nonlinear deconvolution techniques is For example, as described herein, in some embodiments, Estimate the cellular composition for each cell type and / or subtype analyzed in the sample To do this, a separate nonlinear regression model is trained and used (e.g., at least FIG. 3A (including as described herein with respect to the detection of cellular components in biological samples). This may allow for more accurate identification of cell types and / or subtypes (e.g., FIG. 7A 7G). Further, in some embodiments, the model architecture Hierarchical structures that may be used as part of the training and / or use of the machine learning techniques described herein. The structure may include, for example, as described herein, including at least with respect to FIG. 5A. For example, the model architecture includes multiple sub-models corresponding to multiple stages. In this case, the output of one or more previous sub-models (e.g., one or more (which may include an initial prediction of one or more cellular constituent proportions for a cell type) is used to generate subsequent sub-models. This allows the model to be used as part of the input for a model (e.g., in the second, third, etc. stages of training and / or use of By improving initial predictions (e.g., from the first stage of training and / or using the model), According to some embodiments, multiple cell types and / or or the output from the first sub-model across multiple models for the subtype is The hierarchical structure of the model can be used to provide input to subsequent submodels. For example, the first submodel prediction of cellular composition ratios for all cell types (e.g., provide as input to a second sub-model (for another cell type and / or subtype) This allows subsequent sub-models (e.g., the second sub-model) to be modeled according to cell type and / or or subtype interdependencies can be taken into account, thereby allowing the analysis of various cell types and The present invention can provide a more accurate prediction of cellular composition across cellular and / or subtypes.

[0087] Another advantage of the approach developed by the inventors is that, in some embodiments, The models presented here are trained on data representing artificial mixtures of cell types, The training process is carried out by physically sampling and analyzing tumor samples. Malignant cells and microenvironmental cells across a much larger range of diverse sample compositions than was previously possible. Take into account the diverse and tissue-specific expression of This makes it possible to perform cellular deconvolution. The effort and computational resources associated with training a nonlinear regression model for the purpose of The artificial mixtures described in the specification are intended to mimic technical noise and the wide biological variability. The data is obtained in such a way that it captures the This improves our ability to identify biologically meaningful signals in the presence of such noise and variability. For example, quantification of technical noise as described herein may be performed. A practical noise model has been developed and may be applied to artificial mixtures. The RNA expression data used to develop these artificial mixtures represent different biological states. These artificial mixtures are derived from multiple different samples across multiple cell populations. The objective of this study is to demonstrate that a nonlinear regression model can predict the cellular composition across different cell types in real tumor samples. Improve our ability to effectively estimate

[0088] As described herein below, including with respect to Figures 8 and 9A-9B, the present invention The method developed by the authors includes improved linearity for cellular deconvolution. As described herein, one aspect that contributes to the success of linear approaches is , the use of an error function developed by the inventors. As described herein, including, the error function may be a piecewise continuous error function. Compared to traditional methods, such as squared distance, a piecewise continuous error function is Consider genes that are strongly expressed in tumor cells. This allows for the identification of genes that are highly expressed in tumor cells. The accuracy of the deconvolution can be improved. The method developed by the present inventors reduces the error associated with the predicted cell composition ratios. can be accurately modeled (e.g., as described herein, including with respect to Figures 8 and 9A). The method provides improved results over conventional techniques (as described in US Pat. No. 6,333,625).

[0089] The following describes the cellular deconvolution system and method developed by the present inventors. A more detailed description of the various concepts and embodiments of the method described herein is provided. It should be appreciated that various aspects can be implemented in any of numerous ways. Examples of typical implementations are provided herein for illustrative purposes only. In addition, the following embodiments The various aspects described in the embodiments can be used alone or in any combination and are described herein. It is not limited to the combinations explicitly stated.

[0090] FIG. 1A depicts a system 100 for determining a cellular composition 110. The illustrated system may be used in clinical or field trials as described herein, including with respect to 11. It can be implemented in a laboratory setting.

[0091] As shown, the system 100 includes a biological sample 102, which may be, for example, a sample from a subject (e.g., , those who have, are suspected of having, or are at risk of having cancer) The subject may have a tumor biopsy sample obtained by screening for a genetic predisposition to cancer (e.g., a known genetic predisposition to cancer). If you have one or more genetic mutations, or if you may have been exposed to a cancer-causing agent, In some cases, there may be a risk of having cancer. The biological sample 102 may be obtained by performing a biopsy on a patient. The antibody can be obtained by obtaining a blood sample, saliva sample, or any other suitable biological sample from The biological sample 102 may have been previously obtained from the subject. Any step applied to (e.g., obtaining expression data from a biological sample) may be performed in vitro. The biological sample 102 may include diseased tissue (e.g., tumor) and / or healthy tissue. In some embodiments, the biological sample may include a sample of a physician, hospital, clinic, or other healthcare provider. In some embodiments, the source or preparation method of the biological sample may include, for example, The embodiments may include any of the embodiments described in the "Biological Sample" section. In an embodiment, the subject matter may include any of the embodiments described in the "Subject Matter" section.

[0092] The system 100 further includes a sequencing platform 104 capable of generating sequence information 106. In some embodiments, the sequencing platform 104 may include next-generation sequencing. Scanning platforms (e.g., Illumina™, Roche™, Ion Torrent™) etc.), or any high-throughput or massively parallel sequencing platform. In some embodiments, the sequencing platform 104 may be any suitable sequence Sequencing device and / or any sequencing system that includes one or more devices In some embodiments, these methods may be automated, and some implementations may include In some embodiments, there may be manual intervention. It may also be the result of next generation sequencing (e.g., Sanger sequencing). In embodiments, sample preparation may follow the manufacturer's protocol. Sample preparation may be performed according to custom protocols or for research, diagnostic, prognostic, and / or clinical purposes. It may be any other protocol for clinical purposes. In some embodiments, the protocol is an experimental In some embodiments, the origin or preparation of the sequence information may be unknown. good.

[0093] The sequence information 106 includes sequence data (e.g., , next-generation sequencing, Sanger sequencing, etc. In addition to the sequence of nucleotides, information contained therein (e.g., information indicating origin, tissue type, etc.) and consider them as information that can be inferred or determined from sequence data. For example, in some embodiments, the nucleic acid is primarily polyadenylated. In some embodiments, the RNA sequence information can be analyzed to determine whether the sequence is The column information 106 may be information contained in a FASTA file, a description contained in a FASTQ file, and / or The quality score, the aligned position contained in the BAM file, and / or any preferred The information may include any other suitable information available from a suitable file.

[0094] In some embodiments, the sequence information 106 is generated using nucleic acid from a sample from a subject. A reference to a nucleic acid can refer to one or more nucleic acid molecules (e.g., multiple nucleic acid molecules). In some embodiments, the sequence information can be used to identify individuals who have or are suspected of having a disease. DNA and / or DNA fragments from previously obtained biological samples of subjects with or at risk of having the disease may be sequence data representing the nucleotide sequence of the RNA. In some embodiments, the nucleic acid is a deoxyribonucleic acid (DNA). In some embodiments, the nucleic acid is prepared to be present in a protein-coding region of the genome. The genome is processed so that only the region (e.g., exome) remains. When nucleic acids are prepared for sequencing, this is called whole exome sequencing (WES). Various methods for isolating exomes for sequencing are known in the art. For example, in solution-based isolation, tagged probes are used to probe target regions (e.g., exons) ) is hybridized to the other region (e.g., a non-binding oligonucleotide) These tagged fragments can then be prepared and sequenced. It is possible.

[0095] In some embodiments, the nucleic acid is a ribonucleic acid (RNA). The RNA sampled includes both coding and non-coding transcribed RNA found in the sample. When such RNA is used for sequencing, sequencing is performed using "total RNA". This is said to be done by sequencing the entire genome, which can also be called whole transcriptome sequencing. The nucleic acid is then isolated so that the coding RNA (e.g., mRNA) can be used for sequencing. This can be prepared by any means known in the art, for example, by using a poly By isolating or screening RNA for adenylated sequences This is sometimes called mRNA-Seq.

[0096] In some embodiments, the sequence information 106 may include raw DNA or RNA sequence data, DNA exome sequences, Data (e.g., whole exome sequencing (WES), DNA genome sequence data (e.g., whole genome sequencing (WES) (from genome sequencing (WGS)), RNA expression data, gene expression data, bias-corrected Gene expression data obtained from a sequencing platform 104 or and / or derived from data obtained from the sequencing platform 104. In some embodiments, the sequence data may include any other suitable type of sequence data, including sequence data. The origin or preparation of information 106 can be “expression data,” “obtaining RNA expression data,” “alignment,” "Genotyping and annotation", "Removal of non-coding transcripts", and "Conversion to TPM and gene assembly". " section.

[0097] Regardless of the sequence data obtained, the sequence information 106 is used to determine the cellular composition ratio 110. The sequence information 106 can be processed using a computer device 108. 1. A computer system 108 (e.g., as described herein with respect to FIG. 10) For example, sequence information can be processed by one or more software programs. 106 using the machine learning based approach of FIG. 2A to FIG. 2C or Any other method described herein (e.g., at least the method shown in FIG. 2A-2C and FIG. 3A-3C) The nonlinear deconvolution method described with respect to FIG. 8 and FIG. 9A-FIG. 9B The signal can be processed according to the linear deconvolution method described in In some embodiments, the computing device 108 may be a physician, clinician, researcher, patient, or other For example, the user may input sequence information 106 to a computer. The data can be provided as input to the computer device 108 (e.g., by uploading a file). by sequencing, and / or using the sequence information to specify the processing or other procedures to be performed. A user input can be provided to

[0098] Regardless of how the sequence information 106 is processed, the result is one or more cellular constituents. As described herein, each cellular component ratio can be a ratio 110 of the total number of cellular components in the biological sample 102. In some embodiments, the biological sample may represent an estimated proportion of each particular type of cell present in the biological sample. The cell composition ratios are normalized to represent 100%. Cell types include, for example, B cells, plasma plasma B cells, non-plasma B cells, T cells, CD4+ T cells, CD8+ T cells, regulatory T cells, helper T cells Cells, CD8+PD1-high, CD8+PD1-low, NK cells, monocytes, macrophages, dormant tumor associated macrophages phages (TAM), M1-like or activated macrophages, neutrophils, endothelial cells, fibroblasts, and and / or any other suitable cell type. According to some embodiments, the cell type is one or T cells may include multiple subtypes. For example, T cells may be CD4+ T cells, CD8+ T cells, regulatory T cells, etc. The cell composition ratio 110 can have subtypes including the following: In addition to the ratios for all cell types, the ratios may also include ratios for cell types that are not subtypes of any other cell type. According to some embodiments, the cellular composition ratio includes a ratio for "other" cell types. This allows for the analysis of cells that are not accounted for in other cellular proportions (e.g., cells that are not explicitly included in the analysis). The estimated proportion of one or more types of cells not included in the total number of cells in the nucleus can be expressed.

[0099] FIG. 1B illustrates a schematic diagram of a method for detecting each of the cell types and cellular components according to some embodiments of the technology described herein. A nonlinear regression model for cell subtypes was used to identify the different cell types and cell subtypes. FIG. 1 is an exemplary diagram for determining different cellular constituent ratios for

[0100] As shown in this example, the first nonlinear regression model, Model A 126, is used to The cell composition ratio 128 for cell type A 122 is calculated using sequence information 124 related to cell type A 122. A second nonlinear regression model, Model B136, can be used to estimate the cell type The cellular composition ratio 138 for B 132 is estimated using sequence information 134 associated with cell type B 136 It is possible.

[0101] For this example, cell type A 122 and cell type B 132 are different cell types. 122 may include B cells, whereas cell type B 132 may include T cells. As the described embodiment of the method is not limited in that respect, cell type A and / or cell type B may be It may be of any suitable cell type.

[0102] In some embodiments, sequence information 124 and sequence information 134 are from cell type A 122 and cell type B 134, respectively. In some embodiments, sequence information can be obtained for cell type B 132. For example, sequence information 1 24 can be linked to a first set of genes specific to cell type A122, while sequence information The information 134 can be linked to a second set of genes specific to cell type B 132. Methods for identifying cell type and / or subtype specific and / or semi-specific genes includes any of the embodiments described in the "Gene Selection and Specificity" section. obtain.

[0103] As shown in Figure 1B, different nonlinear regression models were used to estimate the cellular structure for different cell types. For example, model A126 is used to determine the cell type A122. Model B 136 was used to estimate the composition ratio 128, while model B 136 was used to estimate the composition ratio 128 for cell type B 132. In some embodiments, the method according to at least FIG. Each of the models, including those described herein, is used for a particular cell type. The method can be trained to estimate the cellular composition of a cell.

[0104] In some embodiments, the different cell types can include cell subtypes. As such, cell subtypes of close origin may be differentiated (e.g., from each other and / or from the cells into which they differentiate) As shown in Figure 1B, cell type B 1 32 includes subtype A 142 and subtype B 162. For example, cell type B 132 is a T cell , and subtype A 142 and subtype B 162 are T cell subtypes (e.g. , CD4+ T cells and CD8+ T cells).

[0105] In some embodiments, a third nonlinear regression model, model C146, is used to determine the subtypes. The cellular composition 148 for group A 142 can be estimated using the sequence information 144. A fourth nonlinear regression model, model D156, was used to estimate the cellular structure for subtype B162. The proportion of copies 158 can be estimated using sequence information 164 .

[0106] In some embodiments, sequence information 144 and sequence information 164 are identified as subtype A 142 and subtype B 164, respectively. For subtype B 162, in some embodiments, this may be obtained Obtain sequence information related to gene sets that contain type-specific and / or semi-specific genes For example, the sequence information 144 may include a set of first genes specific to subtype A 142. sequence information 164 is specific for a second subtype B 144 May be associated with a set of genes specific to a particular cell type and / or subtype Methods for identifying specific and / or semi-specific genes include "gene selection and specificity" Any of the embodiments described in relation to the section may be included.

[0107] FIG. 1C shows the expression of multiple genes for exemplary cell populations, including malignant cells and microenvironment cells. A t-SNE plot depicting the current data. As indicated in the legend, the t-SNE plot The cell types and / or subtypes depicted in include macrophages, M1 macrophages, , M2 macrophages, B cells, B cells (non-plasma), plasma B cells, T cells, CD8+ T cells, PD1+ CD8+ T cells, PD1- CD8+ T cells, CD4+ T cells, regulatory T cells, helper T cells, endothelial cells These include cells, monocytes, NK cells, fibroblasts, neutrophils and tumor cells (e.g., cancer cells). The term "cells" includes tumor cells or any other cells associated with a disease and / or affected tissue. Microenvironmental cells may include, for example, immune cells, skin cells, or tumor cells. Any non-tumor cell may be included, including any other cell.

[0108] The t-SNE plot in FIG. 1C shows the t-SNE results for A large number (e.g., at least 1000, at least 5000, at least 10000) of the antibodies that can be collected from a biological sample. Delineating cell types / subtypes across RNA-seq samples (at least 10,000). In this study, we combined RNA-seq datasets, uniformly annotated them, and analyzed the bioinformatics data. Recalculate the expression values ​​using a formatting method (e.g., convert the expression values ​​into By using the method, accurate and comparable measurements of transcript expression can be obtained. For our example, we analyzed RNA-seq data from 12,450 selected samples (e.g., flow cytometry). (selected by magnetic-assisted sorting of cells using beads) This could be subdivided into 19 cell populations of interest. After removal and quality inspection, the selected samples were classified into 10 major subgroups as shown in Table 1 below. They were distributed across cell types and 19 cell subpopulations.

[0109] In the illustrated example, the t-SNE plot 140 shows the cell types / subtypes listed before quality control. The data shows the RNA-seq samples (n=12450) from each type, while the t-SNE plot150 shows the quality control RNA-seq samples from the listed cell types / subtypes were collected after removing samples that did not pass the analysis. The quality control methodology includes "data collection, analysis, and preprocessing" Any of the embodiments described in the section or any other suitable quality control techniques may be included. For example, in some embodiments, data from cells having an abnormal physiological condition may be collected simultaneously. can be excluded (e.g., based on annotations provided with the data). For example, in some embodiments, phorbol myristate acetate / ionomycin All T cell samples with IL-1 activation and / or induced pluripotent stem cell derived samples were excluded. In some embodiments, low isolation purity, sequencing quality parameters, other organisms (e.g. Also remove samples with high contamination (e.g., organisms other than the primary organism under study) and / or low coverage. It was done.

[0110] [Table 1]

[0111] As shown in plot 150, the cell population can include tumor cells 152. 152 is a t-SNE plot of cancer cell lines (n=2166) color-coded by cancer type in Fig. 1D. As shown, the types of cancer include breast, colorectal, and Cancer, head and neck cancer, kidney cancer, lung cancer, melanoma, pancreatic cancer, prostate cancer, gastric cancer, and / or or any other type of cancer.

[0112] According to some embodiments, a portion of the sample of RNA expression data plotted in FIG. 1C and FIG. 1D or all of the above, as described herein, including at least with respect to FIG. 1E. Use cell type / subtype specific and / or semi-specific genes as part of the selection In some embodiments, the method may be performed at least as described herein with respect to FIG. 2. Substitute some or all of the depicted samples of RNA expression data for artificial mixtures of RNA expression data. In some embodiments, the process shown in FIG. 1C and FIG. 1D can be used as part of generating the The RNA expression data included in the collected data and the RNA expression data plotted in Figure 1C and Figure 1D are Data similar to the present data are derived from public datasets and are included in the Gene Expression Omnibus (GE O) and open source databases such as ArrayExpress. In an embodiment, RNA expression data similar to the RNA expression data plotted in FIG. 1C and FIG. 1D are For example, a dataset containing The cells represented in Table 1 are represented by multiple samples from multiple datasets. A similar data set that includes some or all of the types can be used.

[0113] FIG. 1E is a heat map depicting the expression 170 of exemplary genes for cell types 160. As shown, the vertical axis represents cell types160 and the horizontal axis represents gene expression170 per million. Each row in the heatmap represents one RNA-seq sample. As described in the literature, some genes appear to be specific to certain cell types. For example, as shown in the heat map in FIG. 1F, the selected genes 190 The ratio of RNA in the corresponding sorted cell population 180 may be correlated to the ratio of RNA in the corresponding sorted cell population 180. For example, As shown in the heatmap in Figure 1G, the 192 selected genes were significantly associated with 182 tumor cell lines. expression may be limited or absent.

[0114] As shown below, Table 2 shows the results for each of several cell types. Decombinations that are believed to be specific or semi-specific and / or that are described herein The paper specifies a set of genes that can be used in the solution approach.

[0115] Gene Selection and Specificity In some embodiments, the cellular deconvolution method developed by the present inventors To determine the cellular composition of a particular cell type, certain gene expression data are used. For example, in some embodiments, the method may include at least one step of using the method shown in FIG. As described herein, including those relating to specific and / or semi-specific for particular cell types, In some embodiments, only expression data of genes that are specific to a particular cell can be used. Cell type (e.g., non-malignant cell type) specific and / or semi-specific genes are expressed uniquely. In particular, genes that are highly expressed in malignant cells (e.g., cancer cell lines) (e.g., tumor cell specific genes) In some embodiments, specific and / or specific to a particular cell type may be excluded. The step of selecting semi-specific genes comprises carrying out any or all of the following procedures: May include: literature analysis, statistical Kruskal-Wallis test (similar to non-parametric ANOVA) Number change analysis, Conover-Iman test (nonparametric pairwise test for multiple comparisons), and / or correlation analysis using RNA-seq data from Figures 1C-1D.

[0116] In some embodiments, a set of genes (e.g., for a particular cell type) is selected from various sources. In some embodiments, only genes with known functions are used. Some genes may be similar to the labels used in CYTOF. Some may be derived from literature data (indicating specificity of certain genes) and / or some genes are not seen in existing RNA-seq samples of selected cells. (e.g., experimental conditions, sequencing quality, and expression quality) Searching for genes in samples can be done in several ways. Correlate gene expression with the proportion of cells in an artificial mixture using differential gene expression (e.g., as described herein, including at least with respect to FIG. 6A ) , gene expression and TCGA (The Cancer Genome Atlas) samples or samples of selected cells were mixed. Correlation with several marker cell genes (such as CD3 for T cells) in pooled TCGA samples (e.g., increase the number of read counts to add a larger proportion of cells to the sample) and to reduce the correlation between the presence of various cells in the tumor), artificial mixtures Use linear regression (e.g., with L1 regularization) on We use some important metrics (e.g., SHAP or the gain of gradient boosting trees) or by using some genetic algorithms to create artificial cells with known cell composition. and / or the legacy that gives the highest quality of predictions of the machine learning method on real, independent data. Any combination or linkage of these described methods may be used to select a combination of genes. Use.

[0117] If a gene is expressed only in a particular cell type or cell subtype, then that gene is Genes can be considered to be "specific" for a particular cell type or subtype. A compound can be considered "semi-specific" for a particular cell type or subtype if: (1) It is a cell type or subtype that is involved in the differentiation of both a particular cell type or subtype and one or more other cell types or subtypes. (2) if it is expressed in a particular cell type or subtype but not in other cell types or subtypes; For example, a gene is expressed more abundantly in a particular cell type or subtype than in another. The average expression of a gene is less than the average expression of the same gene in other cell types or subtypes. At least a threshold percentage (e.g., 50%, 100%, 200%, 500%, 1000%, etc.) or a threshold factor (e.g., 2 If the coefficient (coefficients, 1, 5, 10, 15, 20, etc.) is high, the gene is semi-specific to a particular cell type or subtype. In one embodiment, the cell type or subtype of a gene may be considered to be heterogeneous. The average expression in one cell type is less than the average expression of the gene in other cell types or subtypes. If both are 10 times larger, the gene is semi-specific for a particular cell type or subtype. For example, between macrophages and monocytes, between CD4+ T cells and CD8+ T cells, There may be common genes between NK cells and CD8+ T cells. In this context, common genes can be considered semi-specific to cell type and / or subtype. In some embodiments, the antibody can be semi-specific for both CD4+ T cells and CD8+ T cells. The present invention identifies genes whose expression is significantly lower or absent in malignant cell (e.g., tumor) lines. In some embodiments, as described above, multiple data Specificity criteria can be evaluated when evaluating against the combined expression data from the In some embodiments, when several types of cells are present in the same dataset, For each such dataset, we consider similar idiosyncratic features in the dataset to reduce batch effects. Gender analysis can also be performed.

[0118] In some embodiments, for each set of genes, the genes are To determine how the gene is expressed in the Genome Atlas for a tumor type of interest For example, for a given cell type, the average occurrence of TCGA expression can be calculated. In other words, it may be desirable for the ratio of the actual Specific or semi-specific genes (e.g., in a set of specific or semi-specific genes) The mean expression of was 70% of the mean expression in the sorted cell sample, while If the expression of other genes in the tumor is around 5%, the specific or semi-specific gene is considered to be a tumor-specific gene. It is likely that this gene is expressed by other cells or that cells within the tumor The manifestations of the children are very different.

[0119] Additionally or alternatively, expression of genes from the same set may be associated with tumors of this type (e.g., If it is desired to correlate between TCGA samples for the desired tumor type described above For this, one can analyze the average correlation with other genes from the set. The expression signatures of genes considered in TCGA LUAD may be low (e.g., 10 TPM Therefore, the correlation between these genes may also be low (e.g., In some cases, NK cell and neutrophil genes Sometimes expression is particularly low.

[0120] The inventors have demonstrated that cells with a common origin and function may often express the same genes. For example, hematopoietic immune cells express CD45 (PTPRC) and HCLS1. Due to their development, immune cells can be divided into lymphoid and myeloid cells. In the current study, lymphocytes can be divided into T cells, B cells, and NK cells, and among T cells, CD4+ T cells However, some of these cells are involved in tumor progression and There are also subtypes that may play an important role in both the treatment and treatment process. As described in the literature, determining the cellular composition of a particular cell subtype may be performed. However, the present inventors have found that the specific There may be few specific and / or semi-specific genes, and such genes in the tumor microenvironment may be important. Because the number of cells that can be expressed may be smaller than the number of cells in the combined cell population, It is recognized and understood that isolating cell subtypes in vivo can be difficult.

[0121] The present inventors have proposed a method to improve the accuracy of determining both cell type and subtype. The method includes, when determining the cell composition ratio for a cell subtype, a combination group of cells (e.g., Specific and / or semi-specific for cell types and subtypes that share common genes We have discovered that using information about gene expression can be a common Genes can be used, for example, to determine the cellular composition of individual cell types and subtypes. Another method using genes common to a group of cell subtypes is described elsewhere herein. As described in, first, the cell composition ratios for the combination groups were calculated, and then the individual It may be possible to refine the calculations to determine the cellular composition ratios for different cell types. do.

[0122] [Table 2A]

[0123] [Table 2B]

[0124] [Table 2C]

[0125] [Table 2D]

[0126] [Table 2E]

[0127] [Table 2F]

[0128] [Table 2G]

[0129] [Table 2H]

[0130] [Table 2I]

[0131] FIG. 2A depicts a method 200 for determining cellular composition for at least one cell type. In some embodiments, the method 200 includes a computer device. 10)。 For example, a computing device may include at least one processor and at least and a processor and processor-executable instructions that, when executed, perform the operations of method 200. and at least one non-transitory storage medium for storing the data. In a system such as system 100 (which may include, for example, a clinical setting or a laboratory setting) , by one or more computing devices, such as by computing device 108 This can be done.

[0132] At operation 202, the method 200 begins with obtaining expression data for a biological sample from a subject. In some embodiments, obtaining expression data comprises obtaining expression data from a subject using any suitable technique, In some embodiments, the expression data may include obtaining expression data from a previously obtained biological sample. Obtaining the data can include obtaining previously obtained expression data from a biological sample (e.g., In some embodiments, the expression data may include accessing a database to obtain expression data. The data is RNA expression data. Examples of RNA expression data are provided herein. In some embodiments, the subject has, is suspected of having, or is diagnosed with cancer. As described herein, including with respect to FIG. 1A, Body samples include biopsy samples (e.g., from a tumor or other diseased tissue of a subject), also referred to as "biological samples." Any of the embodiments described herein, including those relating to In some embodiments, the origin or preparation of the expression data may include any type of biological sample. Any of the embodiments described in the sections "Expression Data" and "Obtaining RNA Expression Data" For example, the expression data may include RNA expression vectors extracted using any suitable technique. As another example, the expression data obtained in act 202 may include expression data measured by TPM. The data may include RNA expression data collected.

[0133] In some embodiments, the expression data is stored in at least one storage medium and For example, expression data may be stored in one or more files, or accessed as part of a The sequences can be stored in a database and then retrieved. At least one storage medium storing the current data is local to the computing device. (e.g., stored on the same at least one non-transitory storage medium), It may be external to the computing device (e.g., a remote database or cloud storage). The expression data may be stored in a single storage medium or in multiple The information may be distributed across multiple storage media.

[0134] In some embodiments, the expression data of operation 202 is a first cell type (e.g., in a biological sample). For a first set of genes associated with the cell type and / or subtype being analyzed, In some embodiments, the first set of genes may include first expression data associated with at least one As described herein with respect to at least FIG. 1E, For example, for endothelial cell types, the set of genes may include ANGPT2, AP LN, CDH5, CLEC14A, ECSCR, EMCN, ENG, ESAM, ESM1, FLT1, HHIP, KDR, MMRN1, MMRN2, It may include NOS3, PECAM1, PTPRB, RASIP1, ROBO4, SELE, TEK, TIE1, and / or VWF. In some embodiments, the first set of genes includes at least those described in Figures 4-6. Train a corresponding nonlinear regression model for that cell type as described herein. It is the same as the set of genes or a subset of the set of genes used as part of the process. It is also possible.

[0135] At operation 204, the method 200 determines a first cellular composition ratio for at least a first cell type. Proceed to step 1002 of the procedure shown, which includes determining a first cell composition ratio for a first cell type. The process comprises: generating first expression data relating to a first set of genes for a first cell type; A first nonlinear regression model (e.g., one or more nonlinear regression models) is processed to obtain a first The method may include determining a first cellular composition ratio for a cell type. For example, the first expression data may be provided as an input to the first nonlinear regression model. In some embodiments, the nonlinear Other information may be provided as part of the input to the regression model, for example, the median expression data. may be included as part of the input to the non-linear regression model. Any other suitable information may additionally or alternatively be provided as part of the input (e.g., The mean of the expression data, the median or mean of a subset of the expression data, or any combination of the expression data or any other suitable statistics derived or otherwise related to the expression data.

[0136] In some embodiments, a portion of operation 204 may be performed for each cell type and / or subtype being analyzed. For example, the expression data may be analyzed using a number of different methods. The subsets were then compared to each nonlinear regression model for each cell type and / or subtype. may be provided as an input.

[0137] In some embodiments, the output of the nonlinear regression model represents an RNA sequence from a first cell type in the sample. As described herein, including at least those relating to Figures 2C and 3C, The estimated proportion of RNA from the first cell type is used to determine the proportion of RNA from the first cell type, as described in the literature. In some embodiments, the corresponding cellular composition can be calculated using at least one of the following: The techniques described herein, including those relating to 3C, include processing a nonlinear regression model. In this case, the output of the nonlinear regression model is the estimated ratio of RNA The percentage may be an estimated cellular composition ratio for the first cell type, rather than a percentage.

[0138] In some embodiments, the process 200 then performs a step of generating a first cellular composition ratio. Proceed to act 206. Applying the nonlinear regression model to the first cell type, including the nonlinear regression model. Regardless of the architecture or input, the output of one or more nonlinear regression models may be processed according to the method 200. They can be combined as part of, stored, or otherwise post-processed. For example, the cellular composition for each cell type may be determined by the computer used to perform method 200. Some of the information may be stored locally on the computer device (e.g., on a non-transitory storage medium). In embodiments, the cellular composition ratios are stored in one or more external storage media (e.g., a remote database). Or cloud storage environment, etc.

[0139] FIG. 2B is an example implementation of a method 200 for determining cellular composition based on expression data. In some embodiments, steps for implementing method 200 may be performed in accordance with the exemplary flowchart of FIG. In some embodiments, the steps of implementing method 200 may include any suitable combination of steps that may be included in the method. The steps may include additional or alternative steps not shown in FIG. 2B. For example, method 200 may be The steps of performing the process may include all the operations included in the exemplary flowchart. Method 200 may include only a subset of the acts included in the exemplary flowchart (e.g., , operation 212 and operation 216, operations 212, 214, 216 and 218, operations 212, 216 and 220, etc.).

[0140] In some embodiments, implementation 220 begins with act 212, where a biological sample from a subject is The step of obtaining expression data for a biological sample from a subject is shown in FIG. 2A, as described herein above, including in relation to operation 202.

[0141] In some embodiments, operation 212 includes obtaining first expression data and second expression data. The first expression data may relate to a first set of genes associated with the first cell type. and the second expression data is associated with a second set of genes associated with the second cell type. For example, the first expression data can be associated with a first set of genes associated with B cells. and the second expression data can be associated with a second set of genes associated with T cells. Additionally or alternatively, the first expression data can be associated with a first cell subtype. The second expression data can be associated with a first set of genes associated with a second type. The gene can be associated with a second set of genes associated with a cell subtype of, for example, The first expression data can be associated with a first set of genes associated with CD4+ cells; The second expression data can relate to a second set of genes associated with CD8+ cells. Techniques for identifying genes associated with different cell types and / or subtypes are known as "genetic This disclosure is incorporated herein, including with respect to the "Child Selection and Specificity" section.

[0142] In some embodiments, the exemplary method 220 proceeds to operation 214, where the expression data is preprocessed. In some embodiments, preprocessing involves using one or more nonlinear regression models. For example, expression data may be generated that is suitable for processing by filtering the expression data. , combining, batching, filtering, or any other In some embodiments, the expression data can be pre-processed by any suitable method. The methods used include "alignment and annotation," "removal of non-coding transcripts," and Any of the embodiments described in the section "Conversion to TPM and Gene Assembly" may be included. It is possible.

[0143] After the expression data is pre-processed, the exemplary method 220 proceeds to operation 216, where the expression data and one or more nonlinear regression models (e.g., at least 5, at least 10, at least Using the 15 models, multiple cellular composition ratios can be determined for multiple cell types. In some embodiments, the nonlinear regression models include at least those relating to FIGS. The neural network may be trained for the first time according to the techniques described herein.

[0144] In some embodiments, a separate nonlinear regression model is used to determine the phenotype of each cell type and / or subtype. For example, operation 216 may include operations 216a and 216b. and operation 216b, which may include detecting the first and second cell types and / or subtypes, respectively. A separate nonlinear regression model was used to determine the cellular composition of the Act 216a includes a step of: Operation 216b includes determining a first cellular composition ratio for the first cell type by performing a second A second cellular composition is derived for the second cell type using the expression data and a second nonlinear regression model. In some embodiments, operation 216 includes determining a ratio. In some embodiments, operation 216 may include only one of the following: To determine the cellular composition of a cell type or subtype (e.g., a third cell type or subtype), one or more An example implementation of operation 216a may include using an additional nonlinear regression model in conjunction with FIG. These and other related art objects are described herein.

[0145] In some embodiments, the exemplary method 220 further comprises an operation 2 for outputting a plurality of cellular constituent ratios. In some embodiments, the plurality of cellular constituent ratios are displayed on a graphical user interface. The data is output via an interface, stored in memory, and transmitted to one or more other computer devices. The signal may be transmitted to a device and / or output in any other suitable manner.

[0146] In some embodiments, the output of the plurality of cellular composition ratios in operation 218 and / or the cellular composition ratios obtained in operation 212 are Techniques for post-processing the collected expression data can be used. Thus, the post-processing method includes using the cellular composition and expression data to perform a post-processing of the biological sample in operation 220. The method may include determining a malignant tumor expression profile for the sample. The file may contain information indicative of the expression of malignant cells contained in the biological sample. In some embodiments, the expression of a plurality of different genes associated with malignant cells is The step of determining a malignant tumor expression profile includes: (a) detecting expression profiles for TME cells in the biological sample; (b) estimating a current profile; and (b) estimating the total expression of the biological sample (e.g., bulk expression data, The method may include subtracting expression of TME cells from the expression data (e.g., expression data obtained in act 212). Exemplary methods for determining a tumor expression profile are described herein, including those relating to FIG. It is stated in the book.

[0147] FIG. 2C shows a graph of the expression pattern of a first cell type using the first expression data and a first nonlinear regression model. 1 illustrates an example implementation of operation 216a for determining a first cellular composition ratio using the Thus, in some embodiments, the first nonlinear regression model is a regression model that is based on the first expression data (e.g., The method may include a first sub-model and / or a second sub-model for processing the first and second sub-models (as shown in

[0148] In some embodiments, the first expression data comprises a set of a first gene associated with a first cell type. In addition to the first expression data associated with the first cell type, a second set of genes associated with the first cell type is also The expression data may include associated second expression data.

[0149] In some embodiments, the implementation uses the first sub-model to analyze RNA from a first cell type. In some embodiments, the method begins with operation 232 for predicting a first value for the estimated ratio of a first expression data relating to a first set of genes and / or any other input information, It can be provided as an input to the first sub-model of the linear regression model, the output of which is the first It can be a predicted ratio of RNA from a cell type.

[0150] In some embodiments, after predicting the first value, the implementation uses a second sub-model to The method then proceeds to operation 234 for predicting a second value for the estimated proportion of RNA from the first cell type. In some embodiments, second expression data relating to a second set of genes is generated for the first sub-model. In addition to the predictions from the model and / or any other input information provided in the first sub-model, The gene expression model may be provided as an input to a second sub-model of the gene expression model. In particular, the first expression data associated with the first set of genes is used as input to the second sub-model. According to some embodiments, predictions from multiple nonlinear regression models may be provided. The measurements (e.g., the output of the first sub-model of each nonlinear regression model for each cell type) are It can be provided as input to the second sub-model of the nonlinear regression model for cell type. The output of the second submodel of a nonlinear regression model is always the same regardless of the inputs to the second submodel. The output of the second sub-model can be the estimated proportion of RNA from the first cell type in the sample. The power, in some embodiments, may include the output of a nonlinear regression model for the first cell type. .

[0151] In some embodiments, the nonlinear regression model may include more than two sub-models. For example, the second submodel can be repeated any number of times, each time adding one or more of the previous submodels. The predictions from the submodel are included as input.

[0152] In some embodiments, the implementation then performs a first analysis of the estimated proportion of RNA from the first cell type. Using the value of 2, proceed to operation 236 for determining the cell composition ratio for the first cell type. In some embodiments, determining the estimated proportion of RNA from a first cell type comprises: (a) detecting a biological sample; and (b) estimating the number of cells of the first type contained in the biological sample. The method may include a step of estimating (e.g., using formula 350). is the estimated ratio of RNA (e.g., R in Equation 350) cell ) into the RNA factor per cell (e.g., A in Equation 350) c ell The step of estimating the total number of cells may include a step of comparing the number of cells of each cell type to the total number of cells of each cell type. The method for estimating the cellular composition ratio may include the steps of: These are described herein, including those relating to 3C.

[0153] FIG. 3A shows an exemplary use of a machine learning method for determining RNA ratios based on RNA expression data. In the illustrated example, primary tumours available in the TCGA database RNA expression data from tumor sample 302 was compared for T cells, CD4+ T cells, and CD8+ T cells. To arrive at the estimated RNA ratio 306, the present specification, including at least those relating to Figures 2A-2C, The data is processed according to the machine learning techniques described in the document.

[0154] In the illustrated example, the RNA expression data for tumor sample 302 is From a line database (e.g., in this example, The Cancer Genome Atlas (TCGA) database) In some embodiments, the RNA expression data is obtained from one or more databases, such as TCGA. The nucleic acid sequence may be obtained from any suitable source, including a database, or directly from a biological sample. (eg, as described herein, including at least with respect to FIG. 1A).

[0155] Regardless of how the RNA expression data is obtained from the tumor sample 302, can be processed using a non-linear regression model 304. According to some embodiments, The nonlinear regression model 304 may be any of the models described herein, including those with respect to at least FIGS. Implement it using a gradient boosting technique (e.g., as implemented in XGBoost) that According to some embodiments, the method described herein, including with respect to Figures 2A-2C, As described above, the nonlinear regression model 304 generates a separate nonlinear regression model for each of the multiple cell types. In the illustrated example, the nonlinear regression model 304 is a T cell A nonlinear regression model for CD4+ T cells, and a nonlinear regression model for CD8+ T cells As shown, in some embodiments, one or provides additional nonlinear regression models for multiple additional cell types and / or subtypes It is possible.

[0156] In some embodiments, the input to the nonlinear regression models 304 is for each nonlinear regression model: For example, the RNA expression data of Figures 2A-2C may include a selected subset of the RNA expression data of As described herein, including the inputs to the nonlinear regression model for a particular cell type, The input may include RNA expression data for genes specific and / or semi-specific to that cell type. For example, in the illustrated example, the nonlinear regression model for T cells is The input RNA expression data included CAMK4, CBLB, CD2, CD226, CD3D, CD3E, CD3G, CD48, and CD 5, CD6, CD7, FLT3LG, ITK, KCNA3, KLRB1, LAG3, LAT, LCK, LTA, SIRPG, SIT1, SLA2, TBX21, TCF7, TESPA1, TRAC, TRAF3IP3, TRAT1, TRBC2, TRDC, TRGC1, TRGC2, UBASH3A, In some embodiments, other information related to the RNA expression data, such as For example, the median of the RNA expression data, or any other suitable statistical value, is used as input to the nonlinear regression model. may additionally or alternatively be provided as a force.

[0157] In some embodiments, the output of the nonlinear regression model 304 is a set of data for each cell type and / or subtype. For example, the nonlinear regression model for T cells may be As an output, we generate a predicted proportion of RNA from T cells in the input RNA expression data. Similarly, the nonlinear regression model for CD4 T cells can be expressed as The nonlinear regression model for CD8 T cells produced predicted proportions as output, and A predicted proportion of RNA in the genomic DNA described herein with respect to FIG. The predicted proportions of RNA are used to identify a portion or subset of the cell types and / or subtypes being analyzed, such as The corresponding cellular proportions for all can be calculated.

[0158] In the illustrated example, we compare the predictions for T cells with the predictions for CD4 T cells + CD8 T cells. In some embodiments, a plot comparing the predicted totals for the subtypes is shown. may or may not equal the prediction for the type, including its subtypes. For example, the sum of the predictions for CD4 T cells and CD8 T cells gives the prediction for T cells. Sometimes the sum of the predictions for CD4 T cells and CD8 T cells is greater than the prediction for T cells. In some embodiments, the sum of the subtype predictions may be less than the total type prediction. may be equal and / or the subtype predictions are normalized to be equal to the total of type predictions Or it can be adjusted.

[0159] FIG. 3B illustrates a first sub-model 326, 327 for determining RNA ratios based on RNA expression data. 8, 330 and the use of nonlinear regression models 320, 322, 324 including second sub-models 338, 340, 342 This is a diagram depicting the above.

[0160] As shown in the exemplary embodiment of FIG. 3B, different nonlinear regression models 320, 322, 32 4 was used to identify genes associated with cell types A 308, B 310, and C 312. In some embodiments, the expression data 314, 316, 318 for each of the exemplary nonlinear The shape regression model generates first values ​​332, 334, 336 for the estimated proportion of RNA from each cell type. the first submodel for estimating the proportion of RNA from each cell type; 3, and includes second sub-models 338, 340, 342 for generating 344, 346, 348.

[0161] Non-limiting examples for using a non-linear regression model that includes one or more sub-models include: 322, a nonlinear regression model trained to estimate RNA ratios for cell type B 310 In some embodiments, expression data 316 from a set of genes associated with cell type B 310 is generated. can be obtained and used as input to the nonlinear regression model 322. For example, 10 can include immune cells, and the expression data 316 can include genes ADAP2, ADGRE3, ADGRG3, C1QA , C1QC, and C3AR1 (e.g., immune cell-related gene sets listed in Table 2) In some embodiments, the expression data 316 may include expression data from at least one of Some (e.g., expression data related to a subset of genes, some related to all genes) The expression data of gene A is used as an input to the first sub-model 328. We input a subset of expression data 316, including expression data for DAP2, ADGRE3, and ADGRG3. The first submodel then uses the input expression data Processing can be performed to determine a first value 334 of an estimated proportion of RNA from cell type B 310.

[0162] In some embodiments, the exemplary nonlinear regression model 322 estimates RNA from cell type B 310. A second sub-model 340 may be included to generate a second value 346 of the ratio. The second sub-model 340 may use one or more inputs to generate a second value 340. For example, in some embodiments, at least a portion of the expression data 316 can be used as input. In some embodiments, the expression data can be used in the same manner to the first sub-model 328. expression data input (e.g., expression data for genes ADAP2, ADGRE3, and ADGRG3) In some embodiments, the expression data is input to the first sub-model using the same expression data input. In addition to the above, additional expression data may be included (e.g., genes ADAP2, ADGRE3, ADGRG3, C1QA, and In some embodiments, the expression data is based on a first sub-model. The expression data input to the gene expression database may include expression data different from that input to the gene expression database (e.g., genes C1QA, C1QC, and C Expression data for 3AR1).

[0163] Additionally or alternatively, in some embodiments, the second sub-model 340 may include other cell types 308, 3 Estimation of RNA output by the first submodel 326, 330 of the nonlinear regression model 320, 324 for 12 The ratio can be taken as an input. As shown, the second ratio for cell type B 310 The sub-model 340 generates a first value 332 for the estimated proportion of RNA from cell type A 308 and a second value 332 for the estimated proportion of RNA from cell type C 3 12. This type of input is taken as input 336 for the estimated proportion of RNA from another Determine the ratio of RNA from cell types that are associated with the same gene or set of genes as the cell types. For example, cell type B 310 may be informative when trying to differentiate between cell type C 312 and In the case of a gene X that is the same gene, the expression data obtained for gene X may be Which of the two cell types is present in a biological sample may not be highly informative. This is because it may be unclear which cell type generated the expression data. However, the first sub-model 330 determines the first estimated proportion of RNAs determined for cell type C. Consider a scenario where the output is 0% as value 336. This is because there are no cells of cell type C 312 in the biological sample. As a result, all expression data obtained for gene X The second subunit must be expressed by cell type B 310. The delta 340 may use the first values ​​332, 336 to make such an inference.

[0164] In some embodiments, the output of the second sub-model 340 is an estimated proportion of RNA from cell type B 310 As described herein, including with respect to FIG. The estimated RNA ratios can be processed to determine the cellular constituent ratios for each cell type. do.

[0165] FIG. 3C depicts a method for determining cellular constituent ratios 370 based on RNA ratios 360. For example, the method of FIG. 3C may involve the analysis of a portion or portions of the cell types and / or subtypes. To arrive at a prediction of the cellular composition for all, including those for Figures 2 and 3A, and can be applied to predicted RNA ratios according to the techniques described herein.

[0166] As shown in the figure, the process of obtaining cellular composition based on RNA ratios is In some embodiments, formula 350 can be applied to the ratio of RNA for may be applied to each RNA ratio individually (e.g., sequentially), or in some embodiments may be applied to some or all of the RNA ratios together (e.g., in parallel) In some embodiments, formula 350 relates to RNA ratios for cell types that are not subsets of each other. In some embodiments, the formula 350 can then be applied to the first The present invention can be applied to RNA ratios for cell types that are subtypes of one or more cell types. In some embodiments, the calculation of the cellular composition ratio for a cell subtype is performed first. For example, in some embodiments, the cellular composition ratios can be modified based on the The cell composition ratios for the cell subtypes calculated in the first step were summed to obtain the total cell type. So that the cellular composition ratio of the body (i.e., the first calculated cell type that is a subtype) It may be normalized or otherwise adjusted.

[0167] For a given cell type, cell, formula 350 is:

[0168]

number

[0169] Wherein, C cell is the cell composition ratio for that cell type, and R cell The cell type is the RNA ratio for A cell is the RNA factor per cell. As shown in Equation 350, The denominator may include the sum of all cell types and / or subtypes analyzed. Therefore, the formula

[0170]

number

[0171] is first calculated for all cell types and / or subtypes, and then or individual C for subtypes cell This can be used to calculate the value.

[0172] According to some embodiments, the RNA ratio for a cell type can be expressed as a fraction or decimal. (e.g., for purposes of calculation by Equation 350). In some embodiments, RNA ratios can sum to 1 (e.g., Σ cells R cell = 1). In some embodiments, RN If the sum of the A ratios is less than 1, then R other We can introduce the formula, which is 1-Σ cell s R cell In some embodiments, when the sum of the RNA ratios is greater than 1, , R other = 0, and the RNA ratios can be normalized so that they sum to 1.

[0173] In some embodiments, formula 350 represents the RNA factor A per cell. cell per cell The inventors have demonstrated that the abundance of RNA per cell is a function of cell size and / or We recognize and understand that the results may depend on other factors. This can lead to different amounts of RNA compared to the bulk sample. RNA factor per cell can be used to convert RNA ratios to corresponding cellular constituent ratios. So, the RNA factor per cell, A cell can be determined as part of the model training process. can be used (e.g., simulated or artificial models in which the ratios of different cell types are known). In some embodiments, the RNA factor A per cell cell Some or all of the details For example, the RNA factor per cell can be determined experimentally for each cell type. Access data on RNA expression for the type (e.g., available scientific literature, e.g., From PMID:29130882, PMID:30726743, or average or nonlinearly transformed UM by cell type (estimated from single-cell data using I counts) and using that data, Determine the corresponding RNA factor per cell (e.g., purity and / or histological TCGA phosphorylation). This can be obtained by analyzing cell count data.

[0174] In some embodiments, the RNA factor per cell may be tissue specific and is analyzed. It may vary based on disease (e.g., cancer). In some embodiments, the RN per cell The A coefficient may be tissue independent and may not vary depending on the disease being analyzed (see For example, non-malignant microenvironment cells are the same across different cancers, tissues, or diseases. or a substantially similar cellular phenotype). We combine data from multiple cancer types, tissues, diseases, etc. to calculate the RNA factor for each For example, in some embodiments, the RNA factor per cell for a cell type can be calculated. As part of the process to determine the The present inventors have demonstrated that the non-malignant cell composition is determined by histology and WES analysis. We recognize and understand that certain embodiments may be In one embodiment, the step of determining the RNA factor per cell includes deriving a factor for RNA per cell type. To obtain this, we compared the non-malignant cell composition ratio obtained from RNA with the cell composition ratio obtained from DNA. An aligning step may be included.

[0175] The techniques described herein are not limited to application solely to RNA-seq data. For example, some embodiments of the techniques described herein may be implemented in a computer-based system. For this purpose, expression values ​​can be calculated using the 1 for RNA-seq. Normalize the expression levels (e.g., total expression) to be in a similar range to the transcripts per million (TPM) values. is 1 million), optionally a linear scale can be used.

[0176] FIG. 3D illustrates a method for dividing a cell into 3 groups based on cellular composition in accordance with some embodiments of the technology described herein. 3 is a diagram depicting an exemplary method 380 for determining a malignant tumor expression profile using This involves obtaining a biological sample (e.g., a biopsy sample) and identifying malignant cells contained in the biological sample. In some embodiments, the method may include determining the expression of a gene (e.g., the expression of an individual gene). This allows for the derivation of expression in TME cells from overall expression in a biological sample (e.g., expression in a bulk biopsy sample). The method may include a step of removing the

[0177] As shown, this exemplary method includes three steps. The first step 382 is to In some embodiments, this includes determining the average expression profile of non-malignant cell types. may include using expression data from the sorted cell types. For example, this may include cells, B cells, macrophages, fibroblasts, and any other cells that may be present in the TME. In some embodiments, the method may include obtaining and using RNA-seq data from a suitable cell type. The cell types can exclude tumor (e.g., malignant) cells. The average expression profile includes , the average expression of the set of genes for each cell type.

[0178] This exemplary method then uses cellular deconvolution techniques to classify cellular components. The cellular composition may be determined by analyzing a biological sample (e.g., a biopsy sample) to determine whether the cellular composition is in the biological sample or not. As shown, this includes the proportion of each cell type in the cellular composition. A step of generating a ratio vector may be included. The process may include any of the embodiments described herein, including those relating to Figures 1-3C. It may include.

[0179] Average expression profiles of multiple different cell types contained in the TME (e.g., first step 382) and and the respective proportions of those cell types in the biological sample (e.g., second step 384). Thus, the expression of each cell type in a biological sample can be estimated. Step 386 includes determining the product of the expression profile matrix and the vector of cell proportions. The resulting vector can include a step of expressing a putative expression profile of TME cells in the biological sample. It is a file.

[0180] In some embodiments, determining the tumor expression profile comprises determining a bulk expression profile of the biological sample. (eg, expression from a bulk biopsy sample) to obtain a TME expression profile. As shown, this includes expression profiles of TME cells from bulk expression vectors. This involves subtracting the vector generated for

[0181] FIG. 4 illustrates the use of one or more nonlinear 4 is a flow chart illustrating a method 400 for training a shape regression model. As such, the method 400 may include one or more nonlinear regression models (e.g., at least five, A set of nonlinear regression models (at least 10, but at least 15) is trained to measure the corresponding In some embodiments, the method may include estimating the cellular composition ratio for one or more cell types. In this case, each nonlinear regression model estimates the cellular composition of a particular cell type in a biological sample. A separate nonlinear equation for each cell type and / or subtype is trained to estimate A regression model can be trained.

[0182] In some embodiments, the method 400 may be performed on a computing device (e.g., , at least as described herein, including with respect to FIG. 10). The computer device includes at least one processor and, when executed, performs the operations of the method 400. and at least one non-transitory storage medium storing processor-executable instructions for .

[0183] At operation 402, the method 400 includes obtaining training data including simulated RNA expression data. In some embodiments, one can start by using "simulated" RNA expression data. The data may include RNA expression data that is partially generated in silico. The RNA expression data collected includes data from multiple expression datasets from purified cell type samples. In some embodiments, the data may include data obtained by sampling the reads. In the illustrated example, the RNA expression data may include expression data measured with TPM. the RNA expression data includes first RNA expression data for a first gene associated with a first cell type; and second RNA expression data for a second gene associated with the second cell type. The gene may be, for example, a first cell type specific and / or semi-specific gene, while Alternatively, the second gene may be a second cell type specific and / or semi-specific gene. In some embodiments, the training data is for each cell type and / or subtype to be analyzed, and The data may include RNA expression data for genes associated with the endothelial cell type or endothelial cells.

[0184] In some embodiments, the training data may be generated as part of operation 402. As described herein, including at least with respect to FIG. 6A, in some embodiments, RNA expression data from malignant cells (e.g., cancer cells) and microenvironment cells (e.g., immune cells, skin We combine RNA expression data from human skin cells with multiple simulated A process for generating an RNA mixture (which may be referred to herein as an "artificial mixture" or "mixture") The process can generate simulated RNA expression data. In some embodiments, at least one thousand, at least ten thousand, at least one hundred thousand, or at least one million The mixture of may be generated and / or accessed as part of operation 402.

[0185] The training data may be obtained in any suitable manner at operation 402. For example, the training data may be The data is stored on at least one storage medium (e.g., in one or more files or in a database). In some embodiments, the training data may be stored in at least one Another storage medium may be local to the computing device (e.g., stored on at least one non-transitory storage medium (including the same It may be external (e.g. stored in a remote database or cloud storage environment) The training data may be stored on one storage medium or distributed across multiple storage media. It may be scattered.

[0186] In some embodiments, operation 402 includes pre-processing the training data in any suitable manner. For example, filtering, combining, and organizing the training data into batches. The image may be pre-processed by filtering, filtering, or any other suitable method. Pre-processing can be done, for example, by processing the data using one or more non-linear regression models. In some embodiments, training data suitable for performing the method according to at least one embodiment of FIG. As described herein, including by dividing the training data into separate training, validation, and holdout The data set can be divided into

[0187] In operations 404 through 408, the method 400 uses the training data to generate one or more The process may proceed to train the nonlinear regression model. In particular, acts 404 through 408 , a first model of the nonlinear regression model is trained to obtain a cellular composition for the corresponding first cell type. The process of estimating the ratio is described. Operations 404 and 406 are referred to herein as the training process. According to some embodiments, each model of the nonlinear regression model may include at least one Alternatively, each nonlinear regression model can be trained separately for each cell type (e.g., (with corresponding different input data and different learned parameters). In some embodiments, each nonlinear regression model of the one or more nonlinear regression models is 406 to 407, as modified as necessary, in accordance with the techniques described herein. , and then trained and / or saved according to operation 408.

[0188] In operation 404, training a first model of the nonlinear regression model includes: Proceed to generate an estimated proportion of RNA for the first cell type using the RNA expression data. As described herein, the first RNA expression data can be associated with a first cell type. The first gene (e.g., only the first cell type specific and / or semi-specific genes) In some embodiments, the first RNA expression data is provided as an input to the first model. In some embodiments, other inputs may additionally or alternatively be provided to the first model. For example, medians, means, or other arbitrary values ​​relating to some or all of the RNA expression data may be used. Any suitable information may be provided as part of the input to the first model.

[0189] In operation 406, training a first one of the nonlinear regression models includes training a first model of the nonlinear regression model based on RNA from a first cell type. We can then proceed to update the parameters using the estimated ratio of In an embodiment, as part of operation 406, the estimated proportion of RNA from the first cell type is The results can be compared to known values ​​for the ratio of RNA from A loss function can be applied to the estimated and known values ​​to determine the loss. In this embodiment, the loss can be used to update the parameters of the model. For example, to update the model parameters to minimize the loss Gradient descent, or any other suitable optimization technique, can be applied to .

[0190] The first model may be generated using any suitable method, including non-linear regression techniques, as described herein. In some embodiments, the first model can be ,Gradient boosting machine learning techniques can be used. For example, the first model is Use an ensemble of weak predictive models such as decision trees or gradient boosting algorithms The present invention may include any other suitable predictive models that can be combined in an iterative manner. In one embodiment, a gradient boosting framework such as XGBoost or LightGBM is used as the first It can be used as part of the process of training a model. The forest model can be used as part of the process of training the first model. .

[0191] In some embodiments, for a given nonlinear regression model, operations 404 through 406 are performed multiple times. (For example, at least 100 times, at least 1,000 times, at least 10,000 times, at least 100,000 times, In some embodiments, operations 404 through 406 may be repeated (at least 1 million times). , may be repeated for a set number of iterations, or until a threshold is exceeded (e.g. For example, until the loss falls below a threshold. In some embodiments, including at least those relating to FIG. 5A. As described herein, a nonlinear regression model is trained in two or more stages. It is possible.

[0192] At operation 408, the method 400 continues by training a regression model including the first nonlinear regression model and the second nonlinear regression model. In some embodiments, the method may proceed to output the plurality of nonlinear regression models generated. The step of outputting the trained nonlinear regression models may include: and storing one or more of the above-mentioned information on at least one non-transient computer for later access. storing the model in a readable storage medium (e.g., a memory); providing the model to a recipient. (e.g., using any suitable communication network or other means to communicate with the model) transmitting the associated data to a recipient; Graphical user interface and / or any other method that outputs the trained model. displaying the information to the user in a suitable manner, which is an embodiment of the technology described herein The method is not limited in this respect.

[0193] FIG. 5A shows a method for calculating one or more nonlinear regression models according to the method developed by the present inventors. 5 shows an example method 500 for training a model. The illustrated technique includes at least In combination with any of the other techniques described herein, including those relating to FIGS. can be used.

[0194] As shown, the method 500 includes preparing one or more datasets for training. In some embodiments, the data set may be , may be generated as part of operation 502 (e.g., the present invention, including at least those relating to FIG. 6A ). according to the techniques described herein) and / or may be accessed (e.g., one or more database), as described in more detail herein, including with respect to FIG. The dataset may include multiple artificial mixtures of RNA expression data, including species RNA expression data from various malignant (e.g., tumor) and / or microenvironment cells may be included. In an embodiment, the data set is at least 1000, at least 10,000, at least 10 It may contain tens of thousands, or at least a million artificial mixtures.

[0195] In some embodiments, the dataset may be split into a training dataset and a holdout dataset. For example, in some embodiments, the data sets can be used for training and holding, respectively. A set proportion of the data set to be used for training and holdout is randomly For example, in the illustrated example, 80% of the data set is the training data set. The remaining 20% ​​is left as a reserve dataset.

[0196] As shown in the figure, the holdout data set is used to develop quality metrics. 7B). In some embodiments, the reserved data set is used so that the entire data set can be used for training. As shown in the diagram of operation 502, the training data set is Further, one or more (e.g., 10) sets each including a respective training set and validation set. According to some embodiments, the training data set may be divided into lamellar folds. In some embodiments, the crossfolds are used as part of the training. It is possible to perform field verification.

[0197] Regardless of how the data set is prepared in act 502, method 500 Training a plurality of nonlinear regression models using the dataset in acts 510 and 520. As described herein, including at least with respect to FIG. Each nonlinear regression model is then applied to the corresponding RNA expression data from a particular cell type. As shown in the illustrated example, First, the nonlinear regression model is trained in a first stage corresponding to training a first sub-model of the nonlinear regression model. The training step corresponds to the first stage of training the second submodel of the nonlinear regression model. You can train it.

[0198] In the first stage, in operation 510, a first sub-model of each nonlinear regression model is trained to obtain the An initial prediction of the proportion of RNA from each cell type can be generated. For each first sub-model, the inputs are corresponding cell type specific and / or semi-specific genetic In some embodiments, the RNA expression data may include cell type specific and / or semi-specific. In some embodiments, only RNA expression data for genes that are not required may be provided as input. can provide other information, e.g., median expression data. Regardless of the inputs used, the output of the first stage is an initial estimate of the proportion of RNA from each cell type. Each first sub-model of each nonlinear regression model may be predictive, and each first sub-model of each nonlinear regression model may be predictive for its respective cell type. Provides predictions about.

[0199] In a second stage, in operation 520, a second sub-model of each nonlinear regression model is trained to A second prediction can be generated for the proportion of RNA from each cell type. For each second submodel of the model, the inputs are specific and / or semi-specific for the corresponding cell type. In some embodiments, the method may include the RNA expression data of genes in the target gene, as well as the predictions from the first stage. The RNA expression data used in the second stage are different from the RNA expression data used in the first stage. For example, in some embodiments, a nonlinear regression model is trained in the second stage. For purposes of revising the above, some or all of the training data may be regenerated (e.g., see FIG. 5B and In some embodiments, the first The training data for the first and second stages are parallelized so that the training data for each stage is different. In addition to the RNA expression data, the first The predictions from the stage may be provided as input to a second stage. The initial predictions for all cell types may be provided as input to a second stage. This allows the second stage to effectively refine the predictions from the first stage, resulting in a final model. This may improve the consistency and / or accuracy of the results.

[0200] Regardless of the input provided to the second stage, the output of the second stage is the RNA from each cell type. and a second sub-model of each nonlinear regression model may be: A prediction for each cell type may be provided. In some embodiments, the second prediction is , which may be the final output of a nonlinear regression model (e.g., the present invention, including those relating to Figs. 2 and 4). As described herein. In some embodiments, additional training stages (e.g., additional sub A step model can be implemented (e.g., a third step, a fourth step, etc.), with each step taking as input It takes new training data (e.g., RNA expression data) and the predictions from the previous stage.

[0201] Providing predictions from a previous step as part of the input to a next step may be useful for identifying cell types that are specific to a cell type. The model for uses information about the estimated proportions of other cell types to This allows for adaptation (e.g., when the total number of T cells is equal to 10 and the number of CD4+ T cells is equal to 8 (Knowing that the number of CD8+ T cells is 1, the number of CD8+ T cells cannot exceed 2.) As shown, a multi-stage training procedure allows the model to take this into account. This procedure allows for the integration of information from multiple different cell types and subtypes into each individual cell. It may be possible to use for the mold model.

[0202] FIG. 5B illustrates a block diagram of a machine learning model being trained in accordance with some embodiments of the techniques described herein. FIG. 2 is an exemplary, non-limiting diagram of a method for detecting a smeared area of ​​a smeared area of ​​a smeared area. and in combination with any of the other techniques described herein, including those relating to FIG. It can be used.

[0203] As shown in the figure, scheme 530 is a representation of the method described herein, including with respect to FIG. 4 illustrates the division of a dataset into one or more folds, for example: Randomly split the dataset into three folds and train each of the three folds. In some embodiments, the data set may be further divided into a validation data set and a validation data set. Using the dataset, as described herein, including with respect to FIG. A target mixture can be produced.

[0204] In some embodiments, folds are then used to: One or more models for a given set of parameters (e.g., parameters 550) The parameters are set in the predefined ranges shown in Table 3. In some embodiments, the number of occurrences may be generated (e.g., randomly) based on the number of occurrences. At least some (e.g., all) of the classes are used to train each cell type model individually. In some embodiments, a validation mix can then be used to test each parameter. A set of metrics can be evaluated to generate associated evaluation data. In an embodiment, the parameters are as described herein, including with respect to FIG. It can be updated at each stage of training and / or used as input to subsequent training stages. For example, the first fold can be used as input for the first stage of training. A first set of parameters can then be generated. A second fold is then trained. The updated set of parameters is then used as the input for the second stage of Table 4 and Table 5 show the first and second stages of training, respectively. FIG. 1 shows exemplary parameters for one or more cell type models after the first and second steps. There are.

[0205] [Table 3]

[0206] [Table 4]

[0207] [Table 5]

[0208] FIG. 6A illustrates a method for generating simulated RNA expression data comprising one or more nonlinear FIG. 6 depicts an example method 600 for training a regression model (e.g., at least These data are used as training data as described herein, including with respect to FIGS. In some embodiments, the simulated RNA expression data is generated according to branch 6 of method 600. As shown in Fig. 10 and 620, malignant cells (e.g., cancer cells) and microenvironmental cells (e.g., The RNA expression data from multiple cell lines (e.g., immune cells, stromal cells, etc.) is generated by combining samples of RNA expression data. An exemplary process for generating an artificial mixture of RNA expression data can be described as follows: As described below with respect to FIG. 6A.

[0209] FIG. 6B illustrates a method for mimicking real tissue in accordance with some embodiments of the techniques described herein. FIG. 1 is a diagram depicting an example of a process for generating an artificial mixture of RNA expression data for In some embodiments, the RNA expression data is collected from one or more live samples, as shown at branch 630. One or more selected genes that represent a biological state (e.g., positive gene regulation, negative gene regulation, etc.) In some embodiments, the chromosomes are derived from the same cell type / subtype as shown at branches 640 and 650. As such, one or more cell types / subtypes can be mixed in various ratios to generate artificial mixtures. do.

[0210] FIG. 6C illustrates a method for training a cell type model according to some embodiments of the technology described herein. 1 is an exemplary diagram for generating and using an artificial mixture to In the present specification, the data set is a fall as described herein, including with respect to FIG. In some embodiments, the resulting dataset is used to Then, in some embodiments, the artificial mixture is used to create a or train one or more nonlinear regression models specific for each of the multiple cell types / subtypes. In some embodiments, the folds are mixed and verified as described with respect to FIG. The models resulting from each of these can be considered together or independently.

[0211] 6D and 6E show the results of a method for detecting a specific cell type according to some embodiments of the technology described herein. An exemplary method for generating specific artificial mixtures for training models of / subtypes FIG. 1 is an explanatory diagram. In some embodiments, the methods described herein, including those related to Table 6, To train a particular cell type / subtype model, one or more data Sets can be excluded.

[0212] FIG. 6F illustrates a process for processing a data set in accordance with some embodiments of the techniques described herein. 1 is an exemplary diagram showing an approach for generating artificial mixtures using

[0213] As shown, operation 602 includes rebalancing (e.g., to avoid overtraining the model). Before the step of resampling the large dataset to obtain the In some embodiments, the present specification, including those related to FIG. The data sets are rebalanced 604 and combined as described below: It may be the entire set of samples for a particular cell type. The samples are then randomly selected in operation 608 and averaged in operation 612. In some embodiments, in accordance with the techniques described herein, As mentioned above, overexpression noise can be added to the expression in a cell type.

[0214] Data collection, analysis and pre-processing According to some embodiments, the sample of RNA expression data includes at least one of those shown in FIG. 1C-FIG. 1D. For example, selected malignant cells can be obtained as described herein, including By using multiple samples of cells and microenvironment cells to construct an artificial mixture of RNA expression data, In some embodiments, the number of samples may be as large as the number of samples included in Table 1. In some embodiments, the number of samples may be at least 5,000, at least 1 0,000 pieces, at least 15,000 pieces, at least 20,000 pieces, at least 30,000 pieces, at least 5 The number of samples may be 0,000, at least 100,000, or any suitable number of samples. We will introduce open source datasets, such as Gene Expression Omnibus (GEO) and Arra. In some embodiments, the dataset used is: Selected to meet the following criteria: human (homo sapiens) only, read length > 31bp Standard RNA-seq (no polyA depletion, targeted panels, etc.). In some embodiments, artificial mixtures are constructed. To construct a genomic DNA sequence, only cell types that are relevant to the particular disease being analyzed (e.g., a particular type of tumor) are selected. In contrast, the method described herein, including at least with respect to FIG. As described in the literature, the analysis of gene expression specificity involves data for all cell types. It can also be used for.

[0215] In some embodiments, the selection of the data set includes determining the biological parameters and bioinformatics parameters. The parameters can be based on both pharmacokinetic and ophthalmic parameters, e.g., normal physiological conditions. A dataset of samples cultured under similar conditions may be used. CD4+ overstimulated by ball 12-myristate 13-acetate and ionomycin activation Data sets of 4+ T cells or macrophages co-cultured with excess bacterial cultures As with the previous dataset, datasets with abnormal stimulation were excluded. Only samples with coding read counts of at least 4 million were used.

[0216] In some embodiments, quality control is performed on the RNA expression data prior to construction of the artificial mixture. can be implemented (e.g. to exclude strange or unreliable data sets). For example, some samples of CD4+ T cells lack expression of the CD45, CD4, or CD3 genes. In some embodiments, other The same can be done for cell types. For example, samples for some cell types can be , if they significantly express genes that are not typical for that type of cell (e.g., T cells In the samples with cells, CD19, CD33, MS4A1, etc. were expressed in significant amounts, whereas most of the other In some embodiments, these T cell samples may be excluded if their expression is low. In this condition, if a sample of CD4+ T cells expresses significant amounts of the CD8 gene, they are depleted. In some embodiments, a method such as t-SNE or PCA with different sets of genes can be used. Several methods of expression analysis can be used to depict similarities and differences between datasets. (e.g., as shown in Figures 1C and 1D). If a particular cell type in a dataset does not cluster with the same cell type in other datasets, If so (e.g., in t-SNE, PCA, or other plots), quality control the dataset. and exclude some or all of the data from that data set. This can be done.

[0217] Building the mixture According to some embodiments, a sample prepared as described herein above is used. To generate various artificial mixtures of RNA expression data (e.g., representing simulated tumor tissues), An artificial mixture can be constructed using sample expression to calculate TPM (transcripts per million). The gene expression for an entire sample is generated on a per-sample basis, and the gene expression for an entire sample is generated on a per-sample basis. In some embodiments, the quantization coefficients may be formed as a linear combination. RNA expression data from samples of different cell types were mixed in a defined ratio as described below. As shown in Figure 6A, the simulated RNA for malignant cells A expression data (e.g., as generated as shown in branch 610) is used to generate stains for microenvironment cells. and combine it with the simulated RNA expression data (e.g., as shown in branch 620). It is possible.

[0218] Now referring to branch 620, a method for generating simulated microenvironment cellular RNA expression data is provided. In the illustrated example, a sample of each cell type (e.g., For example, samples of RNA expression data for genes GSE1, GSE2, GSE3, or GSE4, as indicated. , per dataset (e.g., deweighting datasets with a large number of samples) and sub- You can rebalance by type (e.g., to change the proportion of a subtype in a sample). The methods for rebalancing are "rebalancing by data set" and As described herein, including in the section "Subtype Rebalancing" For each cell type, multiple samples can then be randomly selected and averaged. Then, rebalanced / balanced samples were generated for some or all of the cell types used. The normalized samples can be mixed together in a specific ratio (e.g., actual tumor (to simulate the tumor microenvironment).

[0219] Now referring to branch 610, a method for generating simulated malignant cell RNA expression data is provided. In the illustrated example, cancer cells (e.g., NSCLC, A random sample of 1000 sera (CCRCC, Mel, HNCK, etc.) can then be selected. We add high expression noise to the RNA expression data to take into account the aberrant expression of genes in malignant cells. For example, tumor cells may express genes that are not normally present in the parent cell type. Specific, semi-specific, or inflammatory responses associated with immune or stromal cells within the TME may be observed. Where this is the case for a marker gene, the highly expressed gene is one that is Contains high expression noise that may interfere with the deconvolution methods described in Regardless of whether or not the results of branch 610 are true, the results are simulated malignant cell RNA expression data. It is possible.

[0220] As shown in the figure, simulated RNA expression data for malignant cells (e.g. For example, a simulated RNase for the microenvironment cells is generated as shown in branch 610. A expression data (e.g., as shown in branch 620) to generate an artificial mixture (referred to as an "expression mixture" in FIG. 6A). Simulated RNA expression data for cells and simulated RNA expression data for microenvironment cells The RNA expression data collected was randomly split based on a given distribution for the cancer cells. In some embodiments, due to technical noise and biological variability, Noise can then be added to the mixture to mimic the noise that Each type can be designated according to one or more suitable distributions. For example, in FIG. As shown, technical noise is specified by a Poisson distribution and biological variability is The noise due to the technique may be specified to follow a normal distribution. The experimental noise may have multiple components that may be specified by other distributions. For example, Another component of technical noise may be specified by a non-Poisson distribution. Regardless of how the product is produced, in some embodiments, the artificial mixture is It is possible to represent an artificial tumor containing a tumor microenvironment (TME).

[0221] When creating artificial mixtures, the inventors have found that different cells of the same type from different samples I recognize and understand that it may be desirable to use a few By using only one sample per cell type, It is expected that poorer performance will be obtained for tumor samples with This is due to the limited number of differentially expressed read counts as well as the variability of the results. noise, alignment errors, and other sources of technical noise). In creating the artificial mixture, the inventors intend to use as many available cell samples as possible. We recognize that it may be desirable to use

[0222] Therefore, for this example, a large number of RNA-seq samples from various cell types (e.g., at least At least 100, at least 500, at least 1000, at least 2000, or at least 5000 In some embodiments, malignant cells (e.g., pure samples for various diagnostic purposes) were collected. We have developed several datasets of cancer cells (cancer cells, cancer cell lines, or cancer cells selected from tumors) For each cell type, several corresponding Table 7 shows the remaining samples after quality control for some cell types. The quantities of samples are listed.

[0223] In some embodiments, as described herein, including with respect to FIG. The mixture is used as a training data set for training one or more nonlinear regression models. In some embodiments, a non-linear regression model can be used to model the cell type / subtype. Thus, in some embodiments, for each specific cell type model To train the model, we generate many (e.g., 150,000) artificial mixtures. As shown in Figures 6D and 6E, the set of mixtures used for each model contains specific data sets that allow for differentiation between specific cell types / subtypes. For example, to train a model of CD4+ T cells, To avoid uncertainty about the proportion of CD4+ T cells in the dataset, we used unspecified T cell As an example, Table 6 shows that a dataset containing one or The mixture of multiple corresponding cell types / subtypes used to train the model is specified. is.

[0224] [Table 6]

[0225] Sample averaging In some embodiments, multiple samples for each cell type are averaged in any suitable manner. It is possible to improve the quality of the sample before adding artificial noise, for example. In some embodiments, the averaging step can be performed in groups of two, e.g., 400 An averaged sample of 10,000 reads may contain information from 8 million reads. In this embodiment, the step of averaging across multiple samples may be affected by technical factors during sequencing. This can reduce expression noise caused by

[0226] In some embodiments, for each cell type, num av Select samples and average their expression do(num av The values ​​of are shown in the parameter table, Table 9. As a type sample, any subtype sample can be used at this stage. For example, in some embodiments, regulatory T cells can be treated together with T cells. Although the approach allows for greater subtype diversity in artificial samples, it is not possible to Averaging multiple samples reduces biological variability in gene expression within cell types or subtypes. the degree of averaging employed may affect the outcome of the study. For this reason, we can set the number of samples for averaging as a parameter and use it to train other can be selected together with other parameters (e.g., to enhance or maximize quality). (To do so).

[0227] Sample Rebalancing The number of cell samples available varies widely for different data sets and cell subtypes. As such, in some embodiments the number of samples may be rebalanced. As described herein below, in one example, samples are rebalanced for each data set. The cells can then be sorted and then rebalanced by subtype. From the rebalanced number of samples, num av A sample can be selected.

[0228] Per-dataset rebalancing In some embodiments, the number of sorted cell samples in a dataset can range from one to several hundred (e.g., For example, at least 5, at least 10, at least 50, or at least 100 samples. Typically, each dataset is selected and sequenced in the same way. The cell samples in the same dataset may include samples of one or two cell types in which A specific marker set for selection or a specific condition, such as a specific disease of the patient from whom the cells were collected Datasets with a large number of samples may also have the following conditions: To reduce the weights in a dataset with a large number of samples, To achieve this, samples from all datasets were rebalanced for each dataset. It is sampled.

[0229] For example, in some embodiments, for each data set, the number of samples is set to a number N dataset,new To Replace and resample.

[0230]

number

[0231] In the formula, N max is the number of samples in the largest data set (e.g., for a particular cell type) N dataset,old is the number of original samples in the data set. The parameter is a value in the range [0,1], where 0 means no change in the number of samples. Where 1 means that there are the same number of samples for each data set. In this case, the rebalancing parameters can be selected during training.

[0232] Rebalancing by cell subtype For some cell types, samples of this type as well as more specific subtypes were included. The number of available subtype samples may vary depending on the type of subtype. The ratio of the mixture with the glycerol may not match that specified for the cell type. When creating a mixture of two or more samples of a given subtype, the samples can be rebalanced. do.

[0233] For example, in some embodiments, a CD4+ T cell (and helper T cells along with regulatory T cells) sample may be available in significantly greater numbers than CD8+ T cells. In this case, the average T cell The proportion of CD4+ and CD8+ T cell samples was determined prior to random selection of samples to form a cell sample. For example, the percentages can be changed in the TCGA or PBMC samples for these cell types. The ratio of RNA fractions can be chosen to be similar to the average ratio of RNA fractions expected for some experiments. In an embodiment, prediction is performed using one or more linear models trained with a mixture of equal cell proportions. This can be obtained using Dell.

[0234] A subtype rebalancing algorithm can be as follows: For a given type To rebalance each subtype, P subtype *msize / min P +1 Resampling is performed with replacement for a number of samples equal to In the formula, P subtype is a number reflecting the proportion of a given subtype (e.g., The proportion of this subtype among all subtypes in the (This can be expressed as the number of samples in each type divided by the total number of samples of this type) , msize is the maximum number of samples among all subtypes of a given type, and min P All P between all subtypes subtypeAccording to some embodiments, rebalance is The operations on the tag are handled by all nested subtypes (e.g. types that themselves have subtypes). The method can be performed recursively on all subtypes (including all subtypes).

[0235] Generation of microenvironmental cell proportions According to some embodiments, the resulting samples of the different cell types are subjected to a simulated To generate input microenvironment cell RNA expression data, For example, a first aliquot of an artificial mixture can be generated using a random proportion of each cell type. You can generate a set,

[0236]

number

[0237] In the formula, R cell is a random number uniformly distributed between 0 and 1, and K cell The associations for specific cell types It is a number.

[0238] According to some embodiments, the coefficient K in the above formula cell is the most likely site of cellular mRNA The ratios are close to those observed in TCGA or PBMC samples. You can choose to do this. These approximate ratios were then used to estimate the mean mean of the TCGA or PBMC samples using models trained without such ratios. For example, the approximate percentage for a given type of tissue can be calculated from the You can use a vector of numbers that reflects the vector. Multiply each number in the vector by a random number between 0 and 1. The resulting coefficients are normalized to the sum and used in a linear combination. In the embodiment, K cellFor each of the multiple cell types, The most likely proportion of cell types based on the phenotype can be selected from Table 7. do.

[0239] We believe that the deconvolution algorithm works across the full range of cells. We recognize and appreciate that in some cases it may be desirable to use a cell suspension from a tumor sample. Preparation of a suspension can lead to a dramatic increase in the proportion of lymphocytes. In some cases, it may be desirable to run algorithms on sequencing data. The authors report that the formation of a cell ratio by the described method involves the differentiation of certain cell types, e.g., NK cells. There is a possibility that samples with a high proportion (e.g., 70-100%) of It is recognized and understood that, in some embodiments, a parameter is An additional mixture is created in which the proportions are generated from a Dirichlet distribution with a parameter 1 / number_of_types. This parameter is selected together with other parameters to create the mixture. The number of samples in the data set formed in this way can be determined by the parameter This parameter can be controlled by the dilution sample ratio (Table 9). can also be selected as a parameter for creating a mixture. Thus, in the final data set, each cell type can be found in a proportion between 0 and 100%. Here, most of the characteristic quantities likely reflect cell populations that mimic real tumors. There is.

[0240] In some embodiments, expression of the artificial tissue is controlled by expression vectors for each cell type and the cells. For example, the present invention can be used to generate a gene encoding a nucleotide sequence based on a randomly selected proportion of RNA. Expression vectors were ligated to random vectors reflecting the RNA fraction of those cells, as described in Add to the coefficient.

[0241]

number

[0242] where α is a random coefficient reflecting the random proportion of cellular RNA for each cell type. can be,

[0243]

number

[0244] represents the RNA expression data of a particular gene for that cell,

[0245]

number

[0246] represents the RNA expression data of a particular gene for that mixture.

[0247] [Table 7]

[0248] Noise Generation As shown in Figure 6A, after the artificial mixture was generated, the RNA expression data showed no significant differences between the neutrophils and the control. Adding noise (e.g., technical noise, uniform noise, or any suitable type of noise) For example, noise can be generated and RNA expression data can be generated according to the process described below. The data is added to the

[0249]

number

[0250] In some embodiments, the expression of each gene can contribute noise to the overall tissue expression. For example, a single gene

[0251]

number

[0252] The expression of

[0253]

number

[0254] It can be expressed as In the formula, μ Ti represents the true expression of the gene,

[0255]

number

[0256] represents the Poisson technical noise, and N prepi is derived from sequencing library preparation represents normally distributed noise, N bioi represents the variable biological noise.

[0257] In some embodiments, the relative standard deviation of the Poisson technique noise (δ Pi ) and normally distributed noise Relative standard deviation (δ Ni ) is used to calculate the quantitative relative standard deviation.

[0258]

number

[0259] Technical variability includes differences in sample and library preparation (non-Poisson noise), as well as Random transcript selection in the sequencing pathway due to limited coverage (Poisson et al., 2013). Many cell types in the microenvironment are typically absent in tumor samples. Therefore, we have decided to use the It is important to take into account the different levels of variability or noise depending on the level of expression of For example, in some embodiments, technical noise (Poisson noise) may be present. A mathematical noise model based on the TPM is provided, taking into account the noise of the eigenvalues ​​(both Poissonian and non-Poissonian). In an embodiment of the present invention, a method for generating a nonlinear regression model as described herein is provided. A model of this variability can be added to the artificial mixture created. The technical non-Poisson noise is assumed to be normally distributed. Account for manufacturing variability, alignment, or human handling variability of different samples. In contrast, Poisson noise is a type of technical noise that is May be related to the coverage or read count of the target gene and may not be normally distributed. The resulting dependence of technical noise on coverage and gene expression is given by the formula :

[0260]

number

[0261] It can be expressed by In the formula, l i is the effective gene length, Tj is the average TPM in technical replicates, and R is the read where is the count and α is the estimated proportionality coefficient. According to this formula, the lower the coverage, the more dispersion there is. According to this formula, genes with low expression levels have high levels of Poisson noise. It is thought that this will become more common.

[0262] As described herein below with respect to Example 1, this model uses purified cells As a result of expression levels and coverage as shown using technical replicates of the population As shown, this can accurately represent the variability in gene expression (Figure 12I). In this case, the detection limit for gene expression was 1 TPM at a coverage of 20 million total reads, 100% of the sample. The coverage ranged from 12 TPM at 1 million reads per well to 12 TPM at 1 million reads per well. The ability to replicate may be affected by the amount of material available. Calculate the noise factor (α) for Poisson noise by plotting the values ​​of By calculating this coefficient, the technique for each sample and each gene can be calculated (Figure 12K). The technical noise can be estimated according to an estimation formula.

[0263] In addition to technical noise, biological noise that may be related to different activation states of cells is also involved in the RNA- In some embodiments, artificial mixtures may contribute to the overall variance of the genomic DNA sequencing samples. It may not be necessary to add biological noise because this noise is a function of the biological state. Already existing through the use of RNA-seq data derived from cell subsets that represent variability As described herein below with respect to Example 1, this Overall variance can be calculated by comparing data for the same cell type obtained in different experiments. The Poisson and non-Poisson data can be evaluated by plotting the Both technical noise, as well as biological variability, depend on the average sequencing coverage. An example of the existence of chromatin-dependent chromatin-dependent markers is provided in FIG. 12J. In this example, for a particular cell type, , there is an average increase of 10% to 26% in noise from technical to biological replicates ( Figure 12J, right).

[0264] In some embodiments, the noise contribution from single gene expression may be determined as described herein. The analysis is applied to simulate technical and biological noise in artificial mixtures. For example, noise can be added to the total gene expression as two summands: It is possible.

[0265]

number

[0266] In the formula, ξ P ,ξ N 〜N(0,1), β is the coefficient of the Poisson noise level coefficient, and γ is the uniform level are the coefficients of non-Poisson noise (Table 9).

[0267] As described herein below with respect to Example 1, the above approach can be used in a non-technical manner. The study was carried out by subtracting technical Poisson noise from the Poisson noise and biological noise. In the example of Figures 12L to 12M, an average variance of about 16% was obtained, which was then In this example, after technical correction, the noise was Technical non-Poisson and biological variability are independent of the measurement method. This is to be expected, since

[0268] The noise models described herein are used to model the artificial mixtures using technical (Poisson and This allows us to add variability (both Poisson and non-Poisson) to the actual tissue. A good imitation of the artificial mixture is obtained. The improved artificial mixture is then used to Train the deconvolution algorithm (e.g., see Figures 4 to 6). As described herein, the model is Stability can be ensured.

[0269] Hyperparameter Estimation As shown in FIG. 6A, a nonlinear regression model was developed according to the method developed by the present inventors. In some embodiments, training the model includes estimating parameters for the model and As described herein, the model may include a step of updating the model. In addition to the weights learned for the model, the parameters of Several parameters, referred to as parameters, may be included (e.g., at least those related to FIG. 4). Such hyperparameters and their An exemplary list of these values ​​is shown in Table 9.

[0270] In some embodiments, each time a nonlinear regression model is trained, the hyperparameters For example, some or all of the hyperparameters can be estimated using the training data. It can be updated based on one or more validation sets of data (e.g., the model In some embodiments, the hyperparameters are calculated using the TCGA data. For example, the Results are checked for consistency against TCGA data to ensure that TCGA model agreement is achieved. For example, in the illustrated example, for a given cell type (e.g., lymphocytes), The sum of the results across cell subtypes (e.g., T cells, B cells, and NK cells) is used to determine the cell It has been confirmed that the results are equal (or close to) the overall results for the cell types.

[0271] In some embodiments, as part of the estimation of hyperparameters, a parameter search can be implemented using a random search, a grid search, or a genetic algorithm. Any suitable parameter search technique may be used. For example, parameter testing can be performed using Bayesian optimization, gradient-based optimization, or evolutionary optimization. In some embodiments, a parametric search can be used to Select one or more hyperparameter values ​​from a given range associated with the parameter You can choose.

[0272] Table 8 and Table 9 list examples of hyperparameters: Averaging The number of samples for (Nav), uniform noise level (γ), Dirichlet sample proportion (Dp), and rebalancing The regulating parameter (r), high expression ratio (Hf), and maximum high expression level (Mhl).

[0273] For each cell type, the "Na v" samples are selected and their expression is averaged.

[0274] As described above, including with respect to the section "Generating Microenvironmental Cell Proportions," We can create several artificial mixtures "Dp" whose proportions are generated from the distribution.

[0275] As discussed above, including with respect to the "Per Dataset Rebalancing" section, The rebalancing parameter “r” is used in the formula to estimate the number of new samples in the data set. As stated, "r" is a value in the range [0,1], where 0 means there is no change in the number of samples, 1 means there are the same number of samples for each data set In some embodiments, the rebalancing parameters are selected during training. It is possible.

[0276] As described above, including with respect to the "Construction of the Mixture" section, To mimic the over-expression of genes, we added high-expression noise to each of the artificial mixtures. In some embodiments, the genome of a selected tumor sample can be used to generate each mixture. A random value is added to the expression of a gene with a small probability. For example, the probability of "Hf" is between 0 and A uniformly distributed random number ranging from "Mhl" to "Mhl" can be added to the expression of each gene.

[0277] [Table 8]

[0278] [Table 9]

[0279] Computational Complexity The machine learning models described herein may have tens of thousands, hundreds of thousands, or even millions of parameters. For example, the nonlinear regression model 304 may include at least As described herein, including with respect to FIGS. 2-6, at least 10,000 parameters A computer with at least 100,000 parameters, or at least 1 million parameters Therefore, the data may be processed by a machine learning model such as a nonlinear regression model 304. The process of creating them requires millions of calculations, even after they have been trained, and is However, it is practically impossible to do this mentally without a computer.

[0280] The algorithms for training machine learning models as described herein are further This can require a large amount of computational resources because such models can be This is because they are trained using hundreds of thousands or millions of artificial mixtures (e.g., at least (See also FIG. 6A and described herein). In one specific example, a two-stage Three million artificial mixtures were generated to train a nonlinear regression model over (eg, as described herein, including at least with respect to FIG. 5A). Without computational resources, neither the training algorithm nor the use of the trained model can be implemented. I can't come.

[0281] result The remainder of this specification describes various methods that have been achieved using the techniques developed by the present inventors. The results are described with respect to Figures 7A-7G. The method developed here substantially outperforms previous methods for cellular deconvolution. In the figure, the cellular deconvolution technique developed by the present inventors is shown. This technique is sometimes called "Cassandra."

[0282] FIG. 7A illustrates simulated RNA expression data 702 (e.g., generated according to the methodology of FIG. 6A). The researchers compared RNA expression data from multiple biological samples (e.g., tumors) with multiple artificial mixtures of genomic DNA. In the illustrated example, the RNA expression data 702 includes the The results were obtained from 500 artificial lung cancer samples derived using the techniques described herein. In comparison, RNA expression data 704 is from 500 RNA-seq data of non-small cell lung cancers from TCGA. As shown in the illustrated example, The gene expression patterns for the IL-1 and actual tumors are quite similar across all samples. As a result, the correlation between real and artificial tumors reached 0.9 (p=0.001).

[0283] FIG. 7B shows the predicted chromatin distribution according to the deconvolution method developed by the present inventors. 1 is a chart depicting exemplary cellular composition ratios and the corresponding true cellular composition ratios. In the illustrated example, the performance of the deconvolution method developed by the inventors is , measured as the Pearson correlation to the holdout artificial mixture (e.g., for Figure 5A As shown, the correlation is consistent across all cells. Correlation exceeds 0.94 for type and exceeds 0.98 for multiple cell types (p=0).

[0284] Figures 7C and 7D show the predicted artificial mixture values ​​versus the true artificial mixture values ​​for different cell types. 1 is an exemplary chart showing the Pearson correlation between mixture values ​​(e.g., prediction accuracy). These graphs are illustrative of the deconvolution method developed by the present inventors. The prediction accuracy of the proposed algorithm is compared with that of alternative algorithms. The prediction accuracy without high expression noise is presented. Figure 7D shows high expression in cancer cells. Prediction accuracy in some cases is presented.

[0285] As described herein, including at least with respect to FIG. 6A, random overexpression Noise can be added to the artificial mixture (e.g., the data set developed by the inventors). The convolution approach allows for ignoring aberrant expression from malignant cells in the sample. To generate accurate high-expression noise, we used TCGA data from four different cancers. Four exemplary gene markers in tumors: CD14 in bladder cancer, F in cutaneous melanoma We analyzed CRLA, STAP1 in clear cell renal cell carcinoma, and PADl2 in squamous cell carcinoma of the lung. Each of these markers was found to be highly expressed in the corresponding cancer types. These markers are not expressed in the corresponding normal tissues, but are expressed in immune cells. It has been found to be expressed (Figure 7E).

[0286] As a result, the deconvolution method developed by the inventors is able to extract the As shown in Figures 7C to 7D, the present inventors The method developed by this study allows for the detection of multiple expression patterns across multiple cell types, even in the presence of high expression noise. The algorithm produces accurate predictions (Figure 7D). Furthermore, Figure 7D shows that the performance of the alternative algorithms is significantly higher than that of the The presence of noise significantly reduces the detection accuracy. This shows that the correlation scores remained high on the evidence dataset.

[0287] Alternative algorithms include CIBERSORT, CIBERSORTx, QuanTISeq, FARDEEP, Xcell, and AB These include IS, EPIC, MCP-Counter, Scaden, and MuSiC. tion of cell subsets from tissue expression profiles,” Nat. Methods 12, 453–45. 7 (2015)) describes CIBERSORT. Newman et al. undance and expression from bulk tissues with digital cytometry”, Nat Biotechno l 37, pp. 773-782 (2019)) describes CIBERSORTx. ar and pharmacological modulators of the tumor immune context revealed by dec "The evolution of RNA-seq data," Genome Med 11, 34 (2019) describes QuanTIseq. Hao et al. ("Fast and Robust Deconvolution of Tumor Infiltrating Lymphocytopenia"). te from Expression Profiles using Least Trimmed Squares”, bioRxiv page 358366; doi : https: / / doi.org / 10.1101 / 358366) describes FARDEEP. Aran et al. : digitally portraying the tissue cellular heterogeneity landscape”, Genome Bio l. 18, 220 (2017)) describes the X cell. Monaco et al. ("RNA-Seq signatures normalized by mRNA abundance allow absolute deconvolution of human immune cells types”, Cell Rep. 26, pp. 1627-1640. e1627 (2019) describes ABIS.

[0288] FIG. 7F shows a different detailed analysis of the deconvolution method developed by the present inventors. Predicted artificial mixture values ​​versus true artificial mixture values ​​for cell types (e.g., prediction accuracy) 1 is a heatmap showing the Pearson correlation between For data from selected samples, predicted cell proportions for different cell types are shown. As shown, the deconvolution method developed by the present inventors , achieving high predictive accuracy scores across multiple cell types, including closely related cell types.

[0289] FIG. 7G shows an exemplary deconvolution method developed by the present inventors. A chart comparing the non-specificity scores with those of alternative algorithms In the illustrated example, the non-specificity scores of 11 alternative algorithms are shown. The values ​​in the chart in Figure 7G are the percentage of specific (true positive) predictions for several different cell types. The nonspecific score represents the proportion of nonspecific (false positive) predictions that A lower ratio of predictions (e.g., indicating a more specific model) is indicated. The detection of signals for each cell type in the population was assessed, with B cells, T cells, and myeloid cells being assessed. Further subdivision of the phages revealed that each subclass was clearly distinct from the others. It was done.

[0290] Linear methods for deconvolution According to some embodiments of the techniques developed by the present inventors, cellular deconvolution An exemplary linear deconvolution technique is shown in FIG. 8 and 9A-9C herein below.

[0291] FIG. 8 illustrates a method for determining cellular composition based on expression data (e.g., RNA expression data). 8 is a flow chart depicting an exemplary linear method 800 as described herein. For each cell type, method 800 generates an expression profile (e.g., RNA expression as shown in FIG. 9A, and / or expression profiles) for one or more cell types in the biological sample. The method may include estimating the proportions of all cells in the sample.

[0292] In some embodiments, the method 800 may be performed on a computing device (e.g., , at least as described herein, including with respect to FIG. 10). The computer device includes at least one processor and a processor that, when executed, performs the operations of method 800. and at least one non-transitory storage medium storing processor-executable instructions for The method 800 may be implemented, for example, in a system such as the system 100 (including, for example, in a clinical setting or The computer system may include a laboratory setting, and may be operated by one or more computing devices, e.g. This can be done by the computing device 108.

[0293] At operation 802, the method 800 begins with obtaining RNA expression data of a biological sample from a subject. In some embodiments, operation 802 may include processing previously obtained RNA expression data from a biological sample. As described herein, including with respect to FIG. The biological sample may include a biopsy sample (e.g., of a tumor or other diseased tissue of a subject) or any other Any suitable type of biological sample can be included, and expression data can be collected using any suitable technique. The expression obtained in operation 802 may include RNA expression data measured by TPM. In some embodiments, the source or preparation of the biological sample may be included in the "Biological Sample" section. In some embodiments, the expression data may include any of the embodiments described above. The origin or method of generation of the data is described in the sections "Expression Data" and "Obtaining RNA Expression Data". Any of the embodiments described may be included.

[0294] In some embodiments, the expression data is stored in at least one storage medium and For example, expression data may be stored in one or more files, or accessed as part of a can be stored in a database and then retrieved as part of operation 802. In an embodiment, the at least one storage medium for storing the expression data is a computer device. may be local to the device (e.g., stored on the same or at least one non-transitory storage medium) The information may be stored on the computer device (e.g., stored in a remote database) or may be external to the computer device (e.g., stored in a remote database). (The expression data is stored in a single storage medium or cloud storage environment.) The data may be stored in one storage medium or may be distributed across multiple storage media.

[0295] In some embodiments, operation 802 includes preprocessing the expression data in any suitable manner. For example, the expression data may be sorted, combined, organized into batches, 3. Preprocessing the image by filtering, filtering, or any other suitable method. By preprocessing, the expression data can be processed as described in the present specification, including those related to operations 804 to 806. The linear regression technique described in may be suitable for processing. In some embodiments, pre-processing of the RNA includes "alignment and annotation," "non-co-translational profiling," and "non-translational sequencing." "Removal of genomic transcripts" and "Conversion to TPM and gene assembly" are described in the sections It may include any of the embodiments.

[0296] As described herein with respect to operations 804 through 806, method 800 may include: 2. Analyze RNA expression using linear regression techniques to determine the corresponding cellular component ratio or ratios The data can then be processed.

[0297] At operation 804, the method 800 generates a plurality of expression profiles (e.g., For example, one may proceed to obtain a 3D image of ... When using method 800 to analyze CD4+ T cells, NK cells, and CD8+ T cells, operation 802 includes: Expression profiles for NK cells, and CD8+ T cells Expression profiles (e.g., RNA expression profiles) can be obtained for Each of the files (files) contains one or more genes associated with each cell type from the plurality of cell types. In some embodiments, each of the offspring may include expression data (e.g., RNA expression data). The genes associated with each cell type are specific and / or semi-specific genes for that cell type. For example, the genes associated with each cell type are listed in Table 2. In some embodiments, the corresponding genes include those included in Table 2. At least 2 genes, At least 4 genes, At least 6 genes, At least 8 genes, at least 10 genes, at least 12 genes, at least 14 genes In some embodiments, the corresponding genes may include at least 16 genes. is less than 10,000 units, less than 5,000 units, less than 2,000 units, less than 1,000 units, less than 500 units, less than 250 units, or Fewer than 100 genes may be included.

[0298] The expression profile can be obtained in any suitable manner. For example, the expression profile can be 8. Save the file in one or more files or in a database as part of operation 804. In some embodiments, the expression profile can be stored in a small At least one storage medium may be local to the computing device (e.g., , stored on the same at least one non-transitory storage medium), or on a computer device It may be external to the database (e.g., stored in a remote database or cloud storage environment). The expression profiles may be stored in a single storage medium or may be stored in multiple storage media. It may be dispersed throughout the medium.

[0299] At operation 806, the method 800 includes, at least in part, determining whether the expression data and the plurality of expression profiles are identical. A piecewise continuous error function between (e.g., the exemplary piecewise continuous error function described with respect to FIG. 9A ) A continuous error function is then used to optimize the plurality of cell composition ratios for the plurality of cell types. Operation 806 can proceed to determining the ratio across multiple cell types simultaneously or The method may be performed iteratively, and in some embodiments (e.g., for a set number of iterations). The method may be repeated until the error measurement is below a threshold.

[0300] According to some embodiments, operation 806 comprises: In some embodiments, the method may include performing a linear regression using a piecewise continuous error function. In some embodiments, this may involve optimizing a piecewise continuous error function. The step of optimizing the piecewise continuous error function is a global optimization of the piecewise continuous error function. It is not limited to finding maximums or minimums, but may be used to find local maximums or minimums within a threshold distance of a global maximum or minimum. For example, operation 806 may include performing a step of finding a specific maximum or minimum for the expression data. A minimum error or error below a threshold (e.g., measured using a piecewise continuous error function) determining a combination (e.g., a weighted sum) of expression profiles having a certain degree of error (for example, a certain degree of error) This may include the steps.

[0301] For a particular cell type, operation 806 performs a piecewise linear regression for each gene associated with that cell type. determining a corresponding output of a continuous error function (such as the error function of FIG. 9C) A piecewise continuous error function is a realistic measurement from real data (e.g., RNA-seq data). Using the expression values ​​obtained and the expression of genes in the expression profile for that cell type, It may be useful to compare the predicted expression values ​​calculated using the above method (e.g., obtained in operation 804) with the predicted expression values ​​calculated using the above method (e.g., obtained in operation 805). For example, the predicted expression value is a function of the expression of a gene in the expression profile and the It can be calculated as a product with the coefficient α.

[0302] For a given gene and cell type, the input to the error function is a coefficient α, the mean mean of the input expression data The expression of genes in the expression profile for that cell type, p The error function may be updated as part of operation 806, see FIG. The coefficients a, b, and k can be as described herein, including those related to the According to an embodiment, operation 806 is performed iteratively or in parallel for some or all of the genes. For example, a piecewise continuous error function can be minimized or below a threshold. Operation 806 is repeated across multiple cell types until a coefficient α that satisfies the above equation is found for each cell type. It can be performed iteratively. According to some embodiments, for a given cell type, the coefficient The value of α is the weighted sum of the errors across all genes (e.g., The piecewise error function calculated using the method described herein, including with respect to operation 806 and FIG. 9C, The coefficients can be determined by finding the coefficient values ​​that minimize

[0303] In some embodiments, the coefficient α may represent a cellular composition ratio for the corresponding cell type. (For example, α defines the weight of each expression profile in the weighted sum of expression data. For example, the step of determining a plurality of cellular composition ratios for a plurality of cell types may include the steps of: processing the coefficients to obtain corresponding cell composition ratios for each of the cell types, e.g. For example, it may include a normalization step.

[0304] FIG. 9A is a diagram depicting exemplary RNA expression profiles and global RNA expression data. The illustrated examples include known RNA expression profiles for CD4+ T cells, NK cells, and CD8+ T cells. The current profiles are shown. Each RNA expression profile is divided into two groups: the horizontal axis represents genes and the vertical axis represents are shown as bar graphs representing the expression of those genes. , each RNA expression profile can be unique for a given cell type.

[0305] As shown in the illustrated example, the overall expression observed for a biological sample is The expression profiles of the various cell types that make up the chromosome can be considered as the sum of the expression profiles of the various cell types that make up the chromosome. Although not included, each RNA expression profile can be weighted by a coefficient α, so that the biological samples are In some embodiments, the sum can be considered as a weighted sum of the NA expression profiles. The expression level can further include a term for unknown expression in other cell types. It can represent expression data that is not explained by the weighted sum of the current profile (e.g., As shown in grey in the expression observed for the samples).

[0306] FIG. 9B depicts an exemplary piecewise continuous error function for use in the method of FIG. As shown in the plot, the error function f is piecewise and the coefficients a and b are related. The coefficient k affects the shape of the rightmost interval of the error function. For the intervals, the error can be calculated according to the depicted expression.

[0307] Biological samples Any of the methods, systems, or other claimed elements may utilize a biological sample from a subject. In some embodiments, the method may be used to measure or analyze the trial The fee is for subjects who have cancer, are suspected of having cancer, or are at risk of having cancer. The biological sample may be obtained from a subject, for example, a biological sample of a bodily fluid (e.g., blood, urine, or brain). spinal fluid), one or more cells (e.g., from a buccal sample or tracheal brushing) (by rubbing or brushing, etc.), tissue fragments (cheek tissue, muscle tissue, lung tissue, heart tissue, brain tissue, or skin tissue), or organs (brain, lungs, liver, bladder, kidneys, pancreas, intestines, or Any type of biological sample containing all or part of a human body (e.g., human body tissue, muscle, etc.) or any other type of biological sample (e.g., human body tissue, muscle, etc.) For example, feces or hair).

[0308] In some embodiments, the biological sample is a tumor sample from the subject. In some embodiments, the biological sample is a blood sample from the subject. A tissue sample.

[0309] A tumor sample, in some embodiments, refers to a sample that contains cells from a tumor. In some embodiments, the tumor sample includes cells from a benign tumor, e.g., non-cancerous cells. In some embodiments, the tumor sample includes cells from a precancerous tumor, e.g., a precancerous cell. In an embodiment, the tumor sample comprises cells from a malignant tumor, e.g., cancer cells.

[0310] Examples of tumors include adenomas, fibromas, hemangiomas, lipomas, cervical dysplasia, pulmonary metaplasia, and leukoplakia. Cancers include, but are not limited to, cancers, carcinomas, sarcomas, germ cell tumors, and blastomas.

[0311] In some embodiments, the sample of blood is a sample containing cells, e.g., cells from a blood sample. In some embodiments, the blood sample comprises non-cancerous cells. In some embodiments, the sample of fluid contains precancerous cells. In some embodiments, the sample of blood contains cancer cells. In some embodiments, the sample of blood comprises blood cells. In some embodiments, the sample of blood comprises white blood cells. The fluid sample includes platelets. Examples of cancerous blood cells include leukemia, lymphoma, and myeloma. In some embodiments, the blood sample includes, but is not limited to, The sample is harvested to obtain cell-free nucleic acid (e.g., cell-free DNA).

[0312] The blood sample may be a whole blood sample or a fractionated blood sample. The sample of blood comprises whole blood. In some embodiments, the sample of blood comprises fractionated blood. In embodiments, the sample of blood comprises a buffy coat, hi some embodiments, the sample of blood comprises serum. In some embodiments, the sample of blood comprises plasma. In some embodiments, the sample of blood comprises a blood clot. Includes.

[0313] A tissue sample, in some embodiments, refers to a sample that contains cells from a tissue. In some embodiments, the tumor sample comprises non-cancerous cells from the tissue. The sample includes precancerous cells from the tissue.

[0314] The disclosed method can be used to treat muscle tissue, brain tissue, lung tissue, liver tissue, epithelial tissue, connective tissue, and nerve tissue. Various tissues, including organ and non-organ tissues, including, but not limited to, human tissues In some embodiments, the tissue may be normal tissue, diseased tissue, or In some embodiments, the tissue may be a tissue section or a complete tissue. In some embodiments, the tissue is an animal or human tissue. Animal tissues include those from rodents (e.g., rats or mice), primates (e.g., monkeys), Examples of tissues that may be used include, but are not limited to, tissues obtained from animals such as cats, dogs, and domestic animals.

[0315] A biological sample may include, but is not limited to, any bodily fluid, such as blood (e.g., whole blood, serum, or or plasma), saliva, tears, synovial fluid, cerebrospinal fluid, pleural fluid, pericardial fluid, peritoneal fluid, and / or urine], Hair, skin (including the epidermis, dermis, and / or hypodermis), oropharynx, larynx, esophagus, stomach , bronchi, salivary glands, tongue, oral cavity, nasal cavity, vaginal cavity, anal cavity, bones, bone marrow, brain, thymus, spleen, small intestine, Appendix, colon, rectum, anus, liver, biliary tract, pancreas, kidneys, ureters, bladder, urethra, uterus, vagina, vulva the cervix, ovaries, cervix, scrotum, penis, prostate, testicles, seminal vesicles, and / or any type of tissue ( Any tissue within the subject's body, including, for example, muscle tissue, epithelial tissue, connective tissue, or nervous tissue. It may be from a supply source.

[0316] Any of the biological samples described herein can be obtained from a subject using any known technique. For example, the collection, processing, and storage of biological samples can each be carried out in their entirety. See the following publications, which are incorporated herein in their entirety: Vaught et al., "Biological "ospecimens and biorepositories: from afterthought to science" (Cancer Epidemiol) Biomarkers Prev. 2012 Feb;21(2):253-5, and Vaught and Henderson, Biol ological sample collection, processing, storage and information management” (IARC Sci Publ. 2011;(163):23~42).

[0317] In some embodiments, the biological sample is obtained from a surgical procedure (e.g., laparoscopic surgery, microsurgical surgery, or endoscopic surgery), bone marrow biopsy, punch biopsy, endoscopic biopsy, or needle biopsy (e.g., fine needle aspiration) The biopsy can be obtained by a variety of techniques, including biopsy, core needle biopsy, vacuum assisted biopsy, or image guided biopsy.

[0318] In some embodiments, one or more cells (i.e., a cellular biological sample) are obtained by scraping or brushing. Cellular biological samples can be obtained from subjects using the sarcoma method. Cellular biological samples can be obtained from, for example, cervical, esophageal, stomach, Any tissue in or from the body of a subject, including one or more areas of the bronchus, respiratory tract, or oral cavity. In some embodiments, one or more tissue fragments (e.g., For example, a tissue biopsy sample can be used. In certain embodiments, the tissue biopsy sample can be , known to have cancerous cells or suspected to have cancerous cells, One or more (e.g., 2, 3, 4, 5, 6, 7) , 8, 9, 10, or more than 10 biological samples.

[0319] Any of the biological samples from a subject described herein may be prepared using methods that maintain the stability of the biological sample. In some embodiments, the biological sample may be stored using any method that maintains its stability. The components of a biological sample (e.g., DNA, RNA, protein, or tissue structure or morphology) , so that the measurements, when taken, represent the state of the sample when it was obtained from the subject. In some embodiments, the biological sample is The components of the biological sample (e.g., DNA, RNA, proteins, or tissue structures or The composition is stored in a manner that can protect the composition (or form) from deterioration. When used, degradation occurs at a rate such that the original form is no longer detectable at the same level as before degradation. It is the transformation of one component into another.

[0320] In some embodiments, the biological sample (e.g., a tissue sample) is fixed. When a sample is "fixed," it is considered that the sample is prevented or prevented from spoilage or deterioration, such as autolysis or putrefaction. It relates to a sample that has been treated with one or more agents or processes to reduce Examples of fixation processes include, but are not limited to, heat fixation, immersion fixation, and perfusion. In some embodiments, samples that are fixed are treated with one or more fixatives. Examples of fixing agents include crosslinking agents (e.g., formaldehyde, formalin, glutaraldehyde, etc.) aldehydes such as aldehydes), precipitants (e.g., ethanol, methanol, acetone, xylene, etc. Alcohol), mercury (e.g., B-5, Zenker's fixative, etc.), picric acid, Hepes-glutamine These include, but are not limited to, acid buffer-mediated organic solvent protective effect (HOPE) fixatives and the like. In some embodiments, the biological sample (e.g., a tissue sample) is treated with a crosslinking agent. In embodiments, the crosslinking agent comprises formalin. In some embodiments, formalin-fixed The biological sample is embedded in a solid substrate, for example, paraffin wax. The biological samples are formalin-fixed paraffin-embedded (FFPE) samples. Methods are known and are described, for example, by Li et al., JCO Precis Oncol. 2018; 2: PO.17.00091. It is listed.

[0321] In some embodiments, the biological sample is stored using cryopreservation. Typical examples include, but are not limited to, step-down freezing, quick freezing, direct plunge freezing, and s These include nap freezing, slow freezing using a programmable freezer, and vitrification. In some embodiments, the biological sample is stored using lyophilization. After collection of the biological sample from the subject, the biological sample may be treated with a preservative (e.g., a preservative for preserving RNA). The RNA is then placed in a container that already contains RNALater and then frozen (e.g., by snap freezing). In some embodiments, such storage in a frozen state occurs immediately after collection of the biological sample. In some embodiments, the biological sample is frozen in or without a preservative before being frozen. in a buffer solution free of The mixture can be kept at room temperature or at 4° C. for up to several days.

[0322] Non-limiting examples of preservatives include formalin solution, formaldehyde solution, RNALater or other equivalent solution, TriZol or other equivalent solution, DNA / RNA Shield or equivalent solution, EDTA (e.g. For example, Buffer AE (10 mM Tris-Cl, 0.5 mM EDTA, pH 9.0) and other coagulants, as well as Acids Citra te Dextronse (e.g., for blood samples).

[0323] In some embodiments, specialized containers are used to collect and / or store the biological samples. For example, a vacutainer may be used to store the blood. In some embodiments, the vacutainer may contain a preservative (e.g., a coagulant or anticoagulant). In some embodiments, the container in which the biological sample is stored is adapted to be cleaned for better preservation or to prevent contamination. In order to avoid

[0324] Any biological sample from a subject described herein may be subjected to any suitable method to maintain the stability of the biological sample. In some embodiments, the biological sample may be stored under any condition that preserves the stability of the biological sample. In some embodiments, the samples are stored at room temperature (e.g., 25° C.). In some embodiments, the samples are stored under refrigeration (e.g., at 4° C.). In some embodiments, the sample is stored under frozen conditions (e.g., at -20°C). and stored under cryogenic conditions (e.g., −50° C. to −800° C.). In some embodiments, the biological sample is stored at -60° C. to -1700° C. -80°C (e.g., -70°C) for up to 5 years (e.g., up to 1 month, up to 2 months, up to 3 months) , up to 4 months, up to 5 months, up to 6 months, up to 7 months, up to 8 months, up to 9 months, Up to 10 months, up to 11 months, up to 1 year, up to 2 years, up to 3 years, up to 4 years, or up to In some embodiments, the biological sample is stored for up to 5 years. Up to 20 years, as described by either (e.g., up to 5 years, up to 10 years, up to It is stored for up to 15 years, or up to 20 years.

[0325] The methods of the present disclosure include obtaining one or more biological samples from a subject for analysis. In some embodiments, a single biological sample is obtained from the subject for analysis. Now, multiple (e.g. 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 1, 14, 15, 16, 17, 18, 19, 20 or more biological samples are In some embodiments, a single biological sample from a subject is analyzed. In some embodiments, a plurality (e.g., 2, 3, 4, 5, 6, 7, 8, 9, 10 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 or more A biological sample is analyzed. If multiple biological samples from a subject are analyzed, the biological samples are analyzed simultaneously. The biological samples may be procured separately (e.g., multiple biological samples may be taken in the same procedure) or the biological samples may be At different time points (e.g., 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 days after the first procedure, 1, 2 , 3, 4, 5, 6, 7, 8, 9, 10 weeks procedure, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 months procedure , 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 years after the procedure, or 10, 20, 30, 40, 50, 60, 70, 80, 9 The samples can be taken at different times (including 0, 100 and 2000 years later).

[0326] A second or subsequent biological sample may be taken from the same area (e.g., from the same tumor or area of ​​tissue). Or may be taken or obtained from different regions (eg, including different tumors). A second or subsequent biological sample may be taken from the subject after one or more treatments, or may be obtained and may be taken from the same or different regions. The two or subsequent biological samples may be used to determine whether the cancer in each biological sample has different characteristics (e.g., , in the case of biological samples taken from two physically separate tumors in a patient), or whether the patient responded to one or more treatments (e.g., the same tumor or tumor before and after treatment) In the case of two or more biological samples taken from different tumors, In some embodiments, each of the at least one biological sample is a body fluid sample, a cell sample, , or a tissue biopsy sample.

[0327] In some embodiments, one or more biological specimens are combined (e.g., For example, the first tumor from the subject may be placed in the same container for storage. The sample can be combined with a second sample of a second tumor obtained from the subject, and the first and The second tumor may or may not be the same tumor. In some embodiments, the first tumor and the second tumor The tumors in a subject may be similar but not identical (e.g., two tumors in the brain of a subject). In an embodiment, the first biological sample and the second biological sample from the subject are samples from different types of tumors. (e.g., tumors in muscle tissue and tumors in brain tissue).

[0328] In some embodiments, the sample from which RNA and / or DNA is extracted (e.g., a tumor sample, or a blood sample) is A sample of at least 2 μg (e.g., at least 2 μg, at least 2.5 μg, The RNA should be large enough that at least 3 μg, at least 3.5 μg or more, of RNA can be extracted. In some embodiments, the sample from which RNA and / or DNA is extracted is peripheral blood mononuclear cells (PBMCs). In some embodiments, the sample from which RNA and / or DNA is extracted may be any type of In some embodiments, the sample from which RNA and / or DNA is extracted (e.g., For example, a tumor sample or a blood sample) should be prepared so that at least 1.8 μg of RNA can be extracted therefrom. In some embodiments, the amount is at least 50 mg (e.g., at least 1 mg). , at least 2 mg, at least 3 mg, at least 4 mg, at least 5 mg, at least 10 mg, At least 12 mg, at least 15 mg, at least 18 mg, at least 20 mg, at least 22 mg, at least 25 mg, at least 30 mg, at least 35 mg, at least 40 mg, at least 45 mg, or Tissue samples (at least 50 mg) are taken from which RNA and / or DNA are extracted. In this embodiment, a tissue sample of at least 20 mg is taken and RNA and / or DNA is extracted therefrom. In some embodiments, at least 30 mg of tissue sample is taken. In some cases, at least 10-50 mg (e.g., 10-50 mg, 10-15 mg, 10-30 mg, 10-40 mg, 20-30 mg, A tissue sample of 0 mg, 20-40 mg, 20-50 mg, or 30-50 mg is taken from which RNA and / or DNA is extracted. In some embodiments, at least 30 mg of tissue sample is taken. In one embodiment, at least 20-30 mg of tissue sample is collected and RNA and / or DNA are extracted therefrom. In some embodiments, the sample from which RNA and / or DNA is extracted (e.g., a tumor The sample, or blood sample) should contain at least 0.2 μg (e.g., at least 200 ng, at least 300 ng , at least 400ng, at least 500ng, at least 600ng, at least 700ng, at least 800ng, at least 900ng, at least 1μg, at least 1.1μg, at least 1.2μg, at least at least 1.3 μg, at least 1.4 μg, at least 1.5 μg, at least 1.6 μg, at least 1. 7 μg, at least 1.8 μg, at least 1.9 μg, or at least 2 μg) of RNA is extracted therefrom. In some embodiments, RNA and / or DNA can be extracted. The sample (e.g., tumor sample or blood sample) to be analyzed should contain at least 0.1 μg (e.g., at least 1 00ng, at least 200ng, at least 300ng, at least 400ng, at least 500ng, at least At least 600ng, at least 700ng, at least 800ng, at least 900ng, at least 1μg, at least 1.1 μg, at least 1.2 μg, at least 1.3 μg, at least 1.4 μg, at least 1.5 μg, at least 1.6 μg, at least 1.7 μg, at least 1.8 μg, at least 1.9 μg, or at least 2 μg) of RNA can be extracted.

[0329] subject Aspects of the present disclosure relate to a biological sample obtained from a subject. In some embodiments, the subject , mammals (e.g., humans, mice, cats, dogs, horses, hamsters, cows, pigs, or other In some embodiments, the subject is a human. In some embodiments, the subject is a In some embodiments, the subject is an adult (e.g., 18 years of age or older). In some embodiments, the subject is a child (e.g., 18 years of age or older). In some embodiments, the human subject has at least one form of cancer. or has been diagnosed with at least one form of cancer. The cancer from which the subject is afflicted is a carcinoma, a sarcoma, a myeloma, a leukemia, a lymphoma, or a cancer, a sarcoma, a myeloma, a leukemia, a lymphoma, or a Carcinoma is a mixed type of cancer that includes more than one of the following: myeloma, leukemia, and lymphoma. Sarcomas are malignant neoplasms of epithelial origin or cancers of the inner or outer membranes of the body. Myeloma refers to cancer that originates in the supporting and connective tissues, such as muscle and fat. Leukemia ("liquid cancer" or "blood cancer") is a cancer that originates in the bone marrow (blood cell-producing Lymphoma is a cancer of the lymph nodes (the area of ​​the bloodstream). The glands of the lymphatic system, a network of blood vessels, nodes, and organs (especially the spleen, tonsils, and thymus) that produce lymphatic fluid Occurring in nodules. Non-limiting examples of mixed cancers include adenosquamous carcinoma, mixed mesodermal tumor, In some embodiments, the subject has a tumor. The tumor is In some embodiments, the cancer is skin cancer, lung cancer, breast cancer, pre-cancerous carcinoma, or ovarian cancer. Any one of the following cancers: prostate cancer, colon cancer, rectal cancer, cervical cancer, and uterine cancer In some embodiments, the subject is diagnosed with a genetic disorder, e.g., the subject has one or more genetic risk factors. or exposure to one or more carcinogens (e.g., tobacco smoke or chewing tobacco). Are at risk of developing cancer because they have been exposed to or have been exposed to do.

[0330] Expression data Expression data (e.g., indicative of expression levels) of a plurality of genes can be analyzed by the methods described herein. The number of genes that can be examined can be determined by the number of genes of interest. In some embodiments, the number of genes of interest may be up to an inclusive number. As a non-limiting example, expression levels can be examined using four or more genes, five genes, or or more, 6 or more, 7 or more, 8 or more , 9 or more, 10 or more, 11 or more, 12 or or more, 13 or more, 14 or more, 15 or more, 16 or more, 17 or more, 18 or more, 19 or more Above, 20 or more, 21 or more, 22 or more, 23 or less 24 or more, 25 or more, 26 or more, 27 1 or more, 28 or more, 29 or more, 30 or more or more, 35 or more, 40 or more, 50 or more, 60 or less or more, 70 or more, 80 or more, 90 or more, 100 or more, 125 or more, 150 or more, 175 or less or more, 200 or more, 225 or more, 250 or more , 275 or more, or 300 or more genes, as described herein. As another set of non-limiting examples, expression data can be used to assess any The data includes, for each cell type listed in Table 2, the At least 5, at least 10, at least 1 selected from a group of genes for cell type 5, at least 20, at least 25, at least 35, at least 50, at least 7 5, and may include expression data for at least 100 genes.

[0331] To obtain expression data (e.g., indicative of expression levels) for a plurality of genes, Any method can be used on samples taken from. A non-limiting set of examples are: The expression data may be RNA expression data, DNA expression data, or protein expression data. obtain.

[0332] In some embodiments, the DNA expression data refers to the level of DNA in a sample from a subject. The level of DNA in a sample from a subject having cancer, e.g., in a sample from a cancer patient, If the level is elevated compared to the level in DNA from a subject that does not have the gene duplication The level of DNA in a sample from a subject with cancer may be a measure of the level of cancer, e.g. reduced levels of DNA in the sample compared to levels in a sample from a subject not having the gene depletion In some cases, this may be the case.

[0333] In some embodiments, the DNA expression data is about expressed DNA (or genes) in a sample. data, e.g., sequencing data for expressed genes in patient samples. Such data may, in some embodiments, indicate that a patient has one or more cancer-related can be useful in determining whether a gene has multiple mutations.

[0334] RNA expression data can be obtained from any of a variety of sources known in the art, including, but not limited to, the following: It can be obtained using any method: whole transcriptome sequencing, Total RNA sequencing, mRNA sequencing, targeted RNA sequencing, small RNA sequencing Sequencing, Ribosome Profiling, RNA Exome Capture Sequencing, and / or deep RNA sequencing. DNA expression data may be obtained from any publicly available source of DNA sequencing. The nucleic acid sequence can be obtained using any method known in the art, including known methods. For example, DNA sequencing can be used to identify one or more mutations in a subject's DNA. Any technique used in the art for sequencing DNA is referred to herein. As a non-limiting set of examples, the methods and compositions described herein can be used in conjunction with A is single molecule real-time sequencing, ion torrent sequencing, pyrosequencing sequencing, sequencing by synthesis, sequencing by ligation (SOLiD sequencing), nanopore sequencing, or Sanger sequencing (chain transcription Proteins can be sequenced by cleavage, cleavage, and cleavage. Expression data may be obtained from any of the methods known in the art, including, but not limited to, the following: The method can be obtained using N-terminal amino acid analysis, C-terminal amino acid analysis, Doman degradation (including by mechanical means such as protein sequencing equipment), or Quantitative analysis method.

[0335] In some embodiments, the expression data is obtained by bulk RNA sequencing. Bulk RNA sequencing is a method to sequence RNA from multiple input cell populations in a single or multiple replicates. The method can include obtaining expression levels of a plurality of genes, the population including a plurality of different cells. In some embodiments, the expression data may include single-cell sequencing (e.g., Single-cell sequencing involves sequencing individual cells. This may include

[0336] In some embodiments, the expression data comprises whole exome sequencing (WES) data. In some embodiments, the expression data comprises whole genome sequencing (WGS) data. In some embodiments, the expression data comprises next generation sequencing (NGS) data. In an embodiment, the expression data comprises microarray data.

[0337] Obtaining RNA expression data In some embodiments, RNA expression data (e.g., data obtained from RNA sequencing ( The method of processing RNA-seq data (also referred to herein as RNA-seq data) includes administering to a subject (e.g., a person having cancer) The method includes obtaining RNA expression data for a subject (a subject diagnosed with cancer or a subject with cancer). In one embodiment, obtaining RNA expression data comprises obtaining a biological sample and performing an RNA sequencing analysis as described herein. 2. Processing the RNA for sequencing using one of the following sequencing methods: In some embodiments, the RNA expression data includes performing an experiment to obtain the RNA expression data. Some of the samples are obtained from the laboratory or facility that performed the analysis (e.g., the laboratory or facility that performed RNA-seq). In an embodiment, the laboratory or facility is a medical laboratory or medical facility.

[0338] In some embodiments, the RNA expression data is stored in a computer storage medium (e.g., In some embodiments, the RNA expression data is obtained by obtaining the RNA expression data. are available via a secure server (e.g., an SFTP server or Illumina BaseSpace) In some embodiments, the data is stored in a text-based file (e.g., a FASTQ file). In some embodiments, the sequencing data is stored in a file. The quality score of the sequencing data is also included in the profile. The file in which the sequencing data is stored also contains sequencing identifier information.

[0339] Alignment and annotation In some embodiments, RNA expression data (e.g., data obtained from RNA sequencing ( The method for processing the RNA expression data (also referred to herein as RNA-seq data) involves identifying genes in the RNA expression data. The annotation is performed by aligning the sequence with known sequences of the human genome. The method includes obtaining coded RNA expression data.

[0340] In some embodiments, alignment of the RNA expression data aligns the data to a specific species of interest. Known assembled genome (e.g., human genome) or transcriptome data A variety of sequence alignment software programs are available. The data are available in assembled genome or transcriptome databases. Alignment software can be used to align a sequence to a Non-limiting examples of aligners include short (unspliced) aligners (e.g., BLAT; BFAS T, Bowtie, Burrows-Wheeler Aligner, Short Oligonucleotide Analysis package, Mosaik), spliced ​​aligners, aligners based on known splice junctions (e.g. For example, Errange, IsoformEx, Splice Seq), or de novo splice aligners (e.g., A In some embodiments, the data alignment is performed using the Any suitable tool can be used to perform the indexing and annotation. For example, Kallisto (github.com / pachterlab / kallisto) aligned and annotated the data. In some embodiments, the known genome is referred to as a reference genome. A reference genome (also called a reference assembly) is a representative set of genes for a species. In some embodiments, the present invention provides a digital database of nucleic acid sequences. The human and mouse reference genomes used in any one of the methods described herein are provided by Genome It is maintained and improved by the Global Reference Consortium (GRC). Non-limiting examples include GRCh38, GRCh37, NCBI Build 36.1, NCBI Build 35, and NCBI B uild 34. Non-limiting examples of transcriptome databases include the Examples include transcriptome shotgun assembly (TSA).

[0341] In some embodiments, the annotation of RNA expression data is performed according to the data to be processed. The location of genes and / or coding regions in the assembled genome or Identifying by comparing with a transcriptome database. Non-limiting examples of data sources for the application include GENCODE (www.gencodegenes.org), R efSeq (see, e.g., www.ncbi.nlm.nih.gov / refseq / ), and Ensembl. In one embodiment, the annotation of genes in the RNA expression data is based on the GENCODE database. The annotations are based on the GENCODE V23 annotations (www.gencodegenes.org).

[0342] Consea et al. (A survey of best practices for RNA-seq data analysis; Genome Biology2 01617:13) provides best practices for analyzing RNA-seq data, which The present invention is applicable to any one of the methods described in the specification, and is hereby incorporated by reference in its entirety. Pereira and Rueda (bioinformatics-core-shared-training.github. io / cruk-bioinf-sschool / Day2 / rnaSeq_align.pdf) may also be used in accordance with any of the methods described herein. This paper describes a method for analyzing RNA sequencing data that can be applied to one of is incorporated herein by reference in its entirety.

[0343] Removal of non-coding transcripts In some embodiments, RNA expression data (e.g., data obtained from RNA sequencing ( The method for processing annotated RNA expression data (also referred to herein as RNA-seq data) is Removing non-coding transcripts from the data. The annotation allows for the identification of coding and non-coding reads. In one embodiment, the analytical effort is focused on identifying proteins (e.g., proteins that may be involved in cancer pathology). Remove non-coding reads from transcripts to focus on the expression of some In embodiments, removing non-coding transcript reads from the data, e.g. or the variance of data in replicate tests of similar samples (e.g., nucleic acids of the same cell or cell type). In some embodiments, non-limiting examples of expression data that is removed include: one or more non-coding transcripts belonging to one or more gene groups selected from the list consisting of (For example, 10-50 pieces, 50-100 pieces, 100-1,000 pieces, 1,000-2,500 pieces, 2,500-5,000 pieces, or more) (or more non-coding transcripts): pseudogenes, polymorphic pseudogenes, processing pseudogenes , transcribed processing pseudogenes, unitary pseudogenes, non-processing pseudogenes, transcribed Transcribed unitary pseudogenes, constant chain immunoglobulin (IGC) pseudogenes, binding chain immunoglobulin IG J pseudogene, variable immunoglobulin (IG V) pseudogene, transcribed non-processed Pseudogene, translated non-processing pseudogene, T-cell receptor binding chain (TR J) pseudogene, possible T cell receptor (TR V) pseudogene, small nuclear RNA (snRNA), small nucleolar RNA (snRNA), microRNA (miRNA), ribozyme, ribosomal RNA (rRNA), mitochondrial tRNA (Mt tRNA), Mitochondrial rRNA (Mt rRNA), Cajal body-specific small RNA (scaRNA), residual introns , sense intron RNA, sense overlap RNA, nonsense mediated decay RNA, nonstop RNA Sense RNA, antisense RNA, long intervening non-coding RNA (lincRNA), macro-long non-coding RNA (macroRNA) lncRNA), processed transcripts, 3' overlapping non-coding RNAs (3' overlapping ncrna), small RNAs (sRNAs), and Other RNAs (misc RNAs), vault RNAs, and TEC RNAs.

[0344] In some embodiments, one or more of these types of transcripts may be used. Information (e.g., sequence information) of a number of transcripts is available from nucleic acid databases (e.g., the Gencode database). Gencode V23, the Genbank database, the EMBL database, or other databases. In some embodiments, the non-coding sequences described herein can be obtained at transcripts, histone-encoding genes, mitochondrial genes, and interleukin-encoding genes. genes encoding collagen, and / or genes encoding T cell receptors Part of the child (for example, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, 95%, 98%, 99%, or 99.5% or more of the RNA expression data were aligned and annotated. is removed from the data.

[0345] Conversion to TPM and gene assembly In some embodiments, RNA expression data (e.g., data obtained from RNA sequencing ( The method for processing RNA-seq data (also referred to herein as RNA-seq data) involves the use of read lengths of transcripts. Convert RNA expression data into (e.g., transcripts per kilobase per million (TPM) format) In some embodiments, the normalized RNA expression per transcript length is First, align and annotate the current data. Convert the data to TPM. This allows for the presentation of expression in the form of concentrations rather than counts, which allows for the This allows for the comparison of samples with different counts and read lengths.

[0346] In some embodiments, the RNA expression data is normalized per transcript read length. By analyzing the data, gene expression data (expression data for each gene) is obtained. This is also called gene aggregation. A gene assembly is a set of reads for all isoform transcripts of a gene. The method includes combining expression data from the various gene sequences to obtain expression data for the gene. In some embodiments, gene aggregation to obtain gene expression data is after TPM normalization. In some embodiments, the method is performed prior to identifying the genes that introduce bias. Gene assembly occurs prior to conversion of the TA to TPM.

[0347] Wagner et al. (Theory Biosci. (2012) 131:281-285) explain how TPM is calculated. provides a description thereof, which is incorporated herein by reference in its entirety. In some embodiments, the following formula is used to calculate the TPM:

[0348]

number

[0349] Computer implementation and sample processing environment Any of the embodiments of the techniques described herein (e.g., the methods of FIG. 2, FIG. 4, and FIG. 6) An example implementation of a computer system 1000 that may be used with the computer system 1000, such as the computer system 1000 shown in FIG. The computer system 1000 includes one or more processors 1010 and a non-transitory computer readable medium. The system includes a removable storage medium (e.g., memory 1020 and one or more non-volatile storage media 1030). The processor 1010 includes a memory 1020 and a non-volatile storage device. The writing and reading of data to device 1030 may be controlled in any suitable manner. However, this is because aspects of the technology described herein are not limited in this respect. To perform any of the functions described herein, the processor 1010 A non-transitory computer that stores processor-executable instructions for execution by the processor 1010. One or more non-transitory computers that function as readable storage media (e.g., memory 1020). One or more processor-executable instructions stored on a computer-readable storage medium. can be executed.

[0350] The computing device 1000 may also be configured to communicate with other computing devices. A network input / output (I / O) interface that can communicate with a user interface 1040, where the computing device provides output to a user, One or more user I / O interfaces capable of receiving input from the user The user I / O interface may also include a keyboard, mouse, display device (e.g., monitor or touch screen), speakers, camera, and and / or various other types of I / O devices.

[0351] In some embodiments, the techniques described herein are implemented in an exemplary environment as shown in FIG. As shown in FIG. 11 , within the exemplary environment 1100, One or more biological samples from 1180 can be provided to the laboratory 1170. The laboratory 1170 The body sample is processed to obtain expression data (e.g., DNA, RNA, and / or protein expression data) and / or Alternatively, sequence information is obtained and information about a subject (e.g., a patient) 1180 is transmitted via a network 1110. The at least one database 1160 may store the

[0352] The network 1110 may be a wide area network (e.g., the Internet), a local area network (LON), or a networks (e.g., corporate intranets), and / or any other suitable type of network. Any of the devices shown in FIG. 11 may be a workpiece. Using a wire link, one or more wireless links, and / or any suitable combination thereof 1110.

[0353] In the illustrated embodiment of FIG. 11, at least one database 1120 stores information about the subject (e.g., expression data or sequence information about a subject (e.g., a patient); medical history data about a subject (e.g., a patient); data, laboratory test result data for a subject (e.g., patient), and / or other assignments related to the subject 1180. Any suitable information may be stored. Examples of data include biopsy results, imaging results (e.g., MRI results), and blood test results. The information stored in the at least one database 1120 may be stored in any suitable The data may be stored in any suitable format and / or structure, including, but not limited to, any of the formats and / or structures described herein. The present invention is not limited in this respect because the present invention is not limited in this respect. 120 may be implemented in any suitable manner (e.g., one or more databases, one or more files, At least one database 1120 may store data in a single database. It may be a single database or multiple databases.

[0354] As shown in FIG. 11, the exemplary environment 1100 may include information about a patient other than the patient 1180. One or more external databases 1120 in which the The database 1160 stores expression data and / or sequence information (e.g., , imaging test results, biopsy results, blood test results), medical history data of one or more patients, or multiple patient test result data, one or multiple patient demographic and / or Personal or biographical information, and / or any other suitable type of information may be stored. In some embodiments, the external database 1160 includes TCGA (The Cancer Genome Atlas), clinical trials, and one or more databases of experimental information and / or commercial sequencing vendors. one or more publicly accessible databases, such as one or more databases maintained by The available information can be stored in a number of databases. 60 may store such information in any suitable manner using any hardware. However, aspects of the technology described herein are not limited in this respect. This is why.

[0355] In some embodiments, at least one of the databases 1120 and the external database 1160 , may be the same database or part of the same database system; or may be physically co-located, although aspects of the technology described herein may This is because there is no limitation in this respect.

[0356] For example, in some embodiments, the server 1140 may store in the databases 1120 and / or 1160 access information stored in the database and use this information to identify one of the biological samples and / or sequence information; or to determine a number of characteristics (e.g., to determine the cellular composition thereof), The process can be carried out as described above.

[0357] In some embodiments, the server 1140 may include one or more computing devices. If the server 1140 includes multiple computing devices, the devices may be installed in the same physical They may be located in one place (e.g., in one room) or distributed across multiple physical locations. In some embodiments, the server 1140 may be a cloud computing infrastructure. In some embodiments, one or more servers 1140 may be part of In the same place, the doctor 1150 may be in a facility operated by an entity (e.g., a hospital or a research institute) to which the doctor 1150 belongs. In such an embodiment, the server 1140 may be located at a It is likely that access to all personal medical data will be easier.

[0358] As shown in FIG. 11, in some embodiments, the distribution performed by the server 640 The results of the analysis are displayed on a computing device 1130 (which may be a laptop or a smartphone). portable computing devices such as MacBooks, or fixed computing devices such as desktop computers The results may be provided to the physician 1150 via a computer device (which may be a The communication may be by written, email, graphical user interface, and / or any other suitable means. In the embodiment of FIG. 11, the results are provided to a physician 1150. In another embodiment, the results of the analysis are provided to the patient 1180 or a medical professional, such as a caregiver, nurse, etc., of the patient 1180. It should be understood that the information may be provided to a medical provider, or to participants in a clinical trial.

[0359] In some embodiments, the results are presented to a physician 1150 via a computing device 1130. In some embodiments, the display may be part of a graphical user interface (GUI) that is provided to the user. The GUI is displayed by a web browser running on the computing device 1130. In some embodiments, the GUI may be presented to the user as part of a web page. An application program (such as a web browser) running on the computing device 1130 For example, in some embodiments, the user may be presented with a The computing device 1130 may be a mobile device (e.g., a smartphone). Rather, a GUI is a user interface that is used by application programs (e.g., The user may be presented with the request via the "Request" ("Request"). EXAMPLES

[0360] Example 1 Establishing RNA transcript normalization and analyzing sequencing technical noise To establish an exemplary process for RNA transcript normalization as described herein, Experiments were performed to analyze the sequencing technical noise.

[0361] FIG. 12A shows purified B cells (as an example of a cell type) sequenced in different laboratories. The transcripts per million covering different biological types were calculated in different samples from The percentage of transcripts per million (TPM) is shown. The GEO and ArrayExpress IDs of the sets are shown as labels on the x-axis. are shown in the legend (according to GENCODE annotation, version 23). In addition, the total expression variability belonging to short RNA transcripts was reduced by the variability due to the length normalization of short transcripts. This strongly distorts the TPM value distribution of the gene of interest by increasing the number of copies of the non-coding transcripts. As mentioned above, including with regard to the section on “How to identify non-coding transcripts from the data” This may reduce the variance of the data.

[0362] FIG. 12B shows the sequence of the reference human transcriptome (GENCODE, v23), as indicated in the legend. The transcript distribution by transcript biotype and length is shown. The percentage of transcripts of various lengths in the biotypes is shown (all retained transcripts in Figure 12C). (with further categories of removed and all removed transcripts). Additionally, a significant amount of noise appears to correspond to V, D, or J regions in the transcriptome. The T- and B-cell lines were derived from short transcripts of the annotated TCR and BCR coding genes. Cells make long transcripts after VDJ recombination, and these short transcripts are never synthesized. Therefore, without specific rearrangement, the various TCR and BCR variants (TCR and BCR repertoires) Finally, we used a filter to measure short non-coding RNA sequences. In addition to being removed by TPM normalization, these TCR and BCR protein-encoding transcripts were The exclusion of non-coding transcripts and TCR- and BCR-transcripts was shown in Figure 12. As shown in B, this has the potential to reduce the variance of the data.

[0363] Figure 12C is a schematic diagram of an exemplary process of expression quantification and TPM renormalization. Expression was calculated using Kallisto (Bray et al. 2016). Next, non-coding transcripts, short V, D or The present study aims to identify the transcripts encoding the TCR / BCR associated with the J segment and their biological properties and quality / Filter out other transcripts that follow the evidence. Finally, aggregate the transcripts by genes. and normalized to 1 million TPM.

[0364] [Table 10A]

[0365] [Table 10B]

[0366] Figures 12D-E show the results before (red) and after (blue) transcript filtration and TPM renormalization. The expression of 3515 housekeeping genes (Eisenberg and Levanon 2013) in various cell types The data are from total RNA-seq (Figure 12D) or polynucleotides. A and B were grouped based on the type of library preparation using either RNA-seq (Figure 12E). The P values ​​shown are calculated by two-tailed Wilcoxon test. Median and rank biserial correlation coefficients are shown.

[0367] Figure 12F shows the total R before (left) and after (right) the proposed transcript filtration and renormalization. Selected results from experiments using either NA-seq (green) or polyA RNA-seq (red). 1 is a PCA projection of RNA expression of B cells. After the normalization procedure, there is an unwanted batch effect between the expression profiles. The techniques for TPM normalization are described herein, including with respect to the "Transcription and Gene Assembly" section. It will be published.

[0368] FIG. 12G shows the dependence of the relative standard deviation of technical replicates on gene expression level (TPM). RNA-se with total coverage of 0 million (pink), 5 million (yellow), and 10 million (green) read counts. q experiments are presented.

[0369] Figure 12H (left) shows the mean standard deviation of gene expression over total coverage of read counts in RNA-seq. The graphs shown show the dependence of noise levels on the threshold: Technical Poisson noise only (blue), all technical noise (yellow) and technical and biological Figure 12H (right) shows the results of the calculations in samples with different types of noise. FIG. 6 is a violin plot showing the distribution of the same standard deviation of gene expression. As mentioned above, the technical noise component may be characterized by a Poisson distribution. Another component of technical noise may be identified by non-Poisson noise. ,Biological noise may be specified by a normal distribution.

[0370] Figure 12I shows technical replicates of an RNA-seq experiment with different total read count coverage. 1 is a plot showing the measured Poisson noise coefficient. The coverage of the data is inversely proportional to the square root of the total read count coverage.

[0371] Figure 12J (left) shows the mean standard deviation of gene expression over total coverage of read counts in RNA-seq. The graphs shown show the dependence of gene expression on the level of the imputed Poisson noise (green). Figure 12J (right) shows the data for the same sample with (yellow) and with all technical noise. Dependence of the mean standard deviation of gene expression on the total coverage of read counts in RNA-seq The graph shown is based on the results presented in the left graph after subtracting the imputed Poisson noise. The same data as shown is shown, showing the non-Poissonian addition to the technical noise. Son technical noise does not show any dependence on sequencing coverage.

[0372] Figure 12K (left) shows the mean standard deviation of gene expression over total coverage of read counts in RNA-seq. The graphs shown show the dependence of the β-glucose concentration on the β-glucose concentration in a single cell across multiple laboratories and experiments. The gene expression of the strains is shown, accounting for both biological and technical noise. The calculated imputed Poisson technical noise is shown in green. Figure 1 shows the dependence of the mean standard deviation of the current read counts on the total coverage in RNA-seq. The graph shown is the gene expression pattern shown on the left after subtracting the imputed Poisson noise. This indicates the presence of pure biological noise in the sample, which is due to the sequencing coverage. It did not depend on the

[0373] Example 2 Deconvoluting the microenvironment from RNA-seq of multiple normal and cancer tissues Using RNA-seq data from multiple normal and cancer tissues, We carried out experiments to perform cellular deconvolution according to the technique described in the present invention. The cellular deconvolution technique developed by the authors is called "Kassandra" Specifically, cell type and / or subtype specific and and / or semi-specific genes are selected, artificial mixtures are generated, and multiple nonlinear regression models are trained. We then developed a method to determine the proportions of multiple cellular components of multiple cell types using a trained nonlinear regression model. The techniques for determining the cellular composition, as well as other pretreatment and and post-processing techniques.

[0374] FIG. 13A shows a schematic of the validation experiment for deconvolution based on TCGA data. Figure 1. Hematoxylin and eosin (H&E) slides and whole exome sequencing ( Data on cell numbers obtained from other methods (WES) are used.

[0375] FIG. 13B shows the B cell, CD4+, CD8+, and macrophage expression levels in 10,489 tumor biopsies from TCGA. Deconvolution of phage, fibroblasts and endothelial cells as described herein Cellular composition estimated using a technique (e.g., using a trained nonlinear regression model) 1 is a violin plot showing the distribution of tumor rates in the example shown. are divided according to the type of cancer.

[0376] Figure 13C shows the TCGA and GTEX assays calculated based on the deconvoluted cell ratios. t-SNE plot showing the

[0377] FIG. 13D shows the TCGA RNA-seq data predicted by the techniques described herein. , and lymphoid tissue predicted by mechanical analysis of histological TCGA data by (Saltz et al. 2018). 13 is a graph showing the Pearson correlation between the proportions of spheres.

[0378] FIG. 13E shows predicted genomic DNA sequences of malignant cells from RNA-seq by the techniques described herein. Plot showing correlation between ratio and WES-estimated tumor purity for 11 TCGA cancer types. It is.

[0379] Figure 13F shows the peer correlation between tumor purity and the predicted proportion of malignant cells based on RNA-seq data. Graph showing tumor correlation. Tumor data was derived from TCGA. The graph is Pearson correlation of predictions from the technique described in the present study as well as predictions from various alternative algorithms The Pearson correlation is shown. The nonlinear correlation coefficients developed by the present inventors are compared with other algorithms. Shape deconvolution technique more accurately predicts proportion of malignant cells and outperforms conventional techniques Improvement was demonstrated.

[0380] FIG. 13G shows the LUSC T cell RNA ratios predicted by the techniques described herein. Pearson correlation of MiXCR with T cell receptor (CDR3 region of TCR) reads in TCGA data 1 is a graph showing the relationship between the

[0381] FIG. 13H shows plasma B cell RNA ratios predicted by the techniques described herein. Peer matching of B cell receptor (IgH CDR3 region) leads by MiXCR in LUSC TCGA data 1 is a graph showing the correlation.

[0382] Figure 13I shows predicted T cell RNA proportions in various cancer types from TCGA data. 1 is a graph showing the Pearson correlation value with the cell receptor (CDR3 region of TCR) reads. Predictions according to the techniques described herein and predictions according to various alternative algorithms are shown. Each data point corresponds to a different cancer type (COAD, KIRC, LUAD, LUSC, READ, SKCM, TNBC). do.

[0383] Figure 13J shows predicted plasma B cell RNA ratios in various cancer types from TCGA. 1 is a graph showing Pearson correlation values ​​with B cell receptor (IgH CDR3 region) leads. Predictions from the technique described in the paper and from various alternative algorithms are shown. Each data point corresponds to a different cancer type (COAD, KIRC, LUAD, LUSC, READ, SKCM, TNBC). Respond.

[0384] In this study, we characterized the cellular composition of TCGA samples from different tumor types and healthy tissues. (Figure 13B). B cells, CD4+ T cells, CD8+ T cells, macrophages, fibroblasts and endothelial cells. Five major cell populations, including IL-17 cells, were quantified (Figure 13C). These values ​​are consistent with those reported For example, the DLBC RNA-seq data showed a strong enrichment for B cells. The tumor purity values ​​predicted by the techniques described herein and other deconvolution techniques are The correlation between the extraction algorithms was compared using an established purity algorithm (Figure 13E 13F). This analysis was performed using the techniques described herein to refine the bulk RNAseq data. This supported the ability to accurately predict cell populations.

[0385] In this example, we analyzed the expression of T cell receptor (TCR) and IgH / L (B cell The proportion of T cells or plasma B cells actively producing immunoglobulins is The sequences were realigned using MIXCR to identify various T and plasma B cell clones. The abundance and diversity of CDR3 transcripts associated with the genomic DNA were measured. The technique described herein is the only one in which the predicted T cell proportions in a sample can be predicted. Provided a strong correlation with the IgH / L transcript fraction with the TCR numbers found and with the plasma B cell proportions (Figures 13G to 13J).

[0386] Example 3 Deconvolution of single-cell and bulk RNA-seq data from blood Using single-cell RNA-seq data and bulk RNA-seq data from blood, Experiments were carried out to perform cellular deconvolution according to the technique described in

[2010] . The cellular deconvolution technique developed by the present inventors is called "Cassandra". Specifically, artificial mixtures are generated and cell types and / or subtypes are identified. Select genes specific and / or semi-specific to the group and train multiple nonlinear regression models. Determine multiple cellular constituent ratios of multiple cell types using a trained nonlinear regression model and other pretreatment and methods described herein to determine the cellular composition. Post-processing techniques.

[0387] Figure 14A shows the validation results for deconvolution using scRNA-seq samples from PBMCs. Schematic diagram of the recognition experiment. scRNA-seq data were artificially mixed to create a bulk RNA-seq data set. We created a mutant strain of this mutant.

[0388] Figure 14B shows the results across nine single-cell PBMC datasets provided by 10x Genomics. The concatenated plots are the t-SNE plots of cell phenotyping using the SCTransform normalization, The Seurat pipeline (Butler et al. 2018; Stuart et al. 2019) included batch correction and up-front PCA. As shown, various cell types and / or subtypes have distinct The cells express important cellular markers (e.g., specific and / or semi-specific genes) that are essential for the development of a novel therapeutic target.

[0389] Figure 14C shows true cell ratios from scRNA-seq of PBMCs and the corresponding values ​​for the bulk RNA-seq mixture. 1 is a graph showing correlation between predictions made using the techniques described herein.

[0390] Figure 14D shows the true ratios from scRNA-seq of PBMCs and the ratios calculated using this specification for the eight cell subtypes. Using techniques described in the literature (e.g., nonlinear regression to determine cellular composition). 4 is a plot showing the correlation of predictions made using the model.

[0391] Figure 14E shows the validity of deconvolution using bulk RNA-seq of PBMCs or whole blood. FIG. 1 is a schematic diagram of a sex confirmation experiment and FACS measurement of the same sample.

[0392] Figures 14F-1 and 14F-2 show the expression of various cell types (CD4+ T cells, CD8+ T cells, NK cells, B cells, monocytes, and and neutrophils) using the techniques described herein from bulk RNA-seq Correlation between the estimated cell ratio and the actual cell ratio obtained by flow cytometry was performed. The graph shows the results of the comparison of the GSE107572 dataset (Finotello et al. 201 9), GSE115823 (Altman et al. 2019), GSE60424 (Linsley et al. 2014), and SDY67 (Zimmermann et al. 20 16), GSE127813 (Newman et al. 2019), GSE53655 (Shin et al. 2014), and GSE64655 (Hoek et al. 2015). Pearson correlations are shown for all cell types combined.

[0393] In this experiment, we applied the techniques described herein to peripheral blood mononuclear cells (PBMs). C) was applied to artificial bulk RNA-seq constructed from scRNA-seq datasets derived from the 10-kDa RNA-seq dataset (Figure 1 4A-14B). High correlation was observed when aligning the true scRNA-seq ratios with the predicted RNA-seq ratios. Correlation values ​​were obtained (FIG. 14C). In this example, when the correlations for each cell type were graphed separately, ,Cell types present in high numbers have the most significant correlation between true and predicted values ​​( Figure 14D).

[0394] The blood samples that were available for FACS analysis were then analyzed using the techniques described herein. Bulk RNA-seq of the broth was analyzed (Figure 14E). Eight different PBMC samples were analyzed, and F The ACS analysis was compared to the cellular organization predicted by the techniques described herein. As shown, all analyses presented correlation coefficients ranging from 0.900 to 0.984 (Figure 14F-1 and and Figure 14F-2).

[0395] Example 4 Deconvolution of the microenvironment from various cancer tissues Using scRNA-seq data from several tumor tissues, including melanoma, head and neck cancer, and lung cancer, To perform cellular deconvolution according to the techniques described herein using The figure shows the results of the cellular deconvolution technique developed by the present inventors. The technique is sometimes called "Cassandra." Specifically, it creates an artificial mixture and Select genes specific and / or semi-specific for the type and / or subtype, and A shape regression model is trained to determine multiple cellular composition ratios of multiple cell types, and the trained nonlinear The technique for determining cellular composition ratios using a shape regression model, as well as the techniques described herein Other pre- and post-processing techniques that may be used.

[0396] Figure 15A shows, from left to right, melanoma (GSE72056) (Tirosh et al. 2016), lung cancer (E-MTAB-6149 and EM TAB-6653) (Lambrechts et al. 2018) and head and neck cancer (HNC) (GSE103322) (Puram et al. 2017) single t-SNE plot of cell phenotyping in the cell dataset. t-SNE plot of lung cancer. The Seurat pipeline (Butler et al. 2 ) includes SCTransform normalization, batch correction and up-front PCA. 2018; Stuart et al. 2019). Melanoma and head and neck cancer t-SNE plots show cell type specificity. The log TPM expression values ​​of differential genes were obtained by t-SNE transformation.

[0397] Figure 15B is a schematic of a validation experiment using scRNA-seq data derived from cancer tissues. The scRNA-seq data were artificially mixed to create a bulk RNA-seq dataset.

[0398] Figures 15C, 15D, 15E, and 15F show the true cell proportion values ​​derived from scRNA-seq data (Figure 15A). and decombination from artificial bulk RNA-seq data using the techniques described herein. 15C) (n=19) and lung cancer (Fig. 15D). (n=12), HNC (Fig. 15E) (n=22) and B-cell lymphoma (Fig. 15F) (n=12) in various cell subpopulations. A correlation has been shown.

[0399] Figures 15G and 15H show predictions from artificial bulk RNA-seq data for melanoma, lung cancer, and HNC. The average Pearson correlation values ​​between the estimated values ​​and the true values ​​derived from the scRNA-seq data (Figure 15G) and FIG. 15H is a heat map showing the average MAE (mean absolute error) scores. The results obtained from the techniques described herein are compared to those obtained from alternative algorithms. In particular, when compared to conventional techniques of deconvolution, The nonlinear regression techniques developed in this study can, on average, more accurately predict the cellular composition of various cell types. The results are shown to have a lower mean absolute error.

[0400] FIG. 15I shows the cell ratios predicted by the techniques described herein and the data set. FAC of lymphocytes, fibroblasts, and lung adenocarcinoma cell lines from the mouse GSE121127 (Wang et al. 2018) (top). S and bone marrow CYTOF from dataset GSE120444 (Oetjen et al. 2018) (bottom). The Pearson correlation value (r) is the mean correlation between the actual cell proportions measured and all the The correlation values ​​of cell types are shown.

[0401] In this experiment, cells obtained from scRNA-seq were manually annotated (Figure 15A) to identify the individual cell types. Specific ratios of the 100% ribosomal RNA were mixed to resemble the bulk RNA-seq sample (e.g., at least as shown in Figure 6A). These cell ratios are then compared to the The results were compared to values ​​predicted by the techniques described in the literature. The ability of the technique to reconstitute the cellular composition of each cell type was measured (Figures 15C-15F). The median correlation of the reconstruction reached approximately 0.97, the highest among the other methods.

[0402] The techniques described herein demonstrate that scRNA-seq data derived mixed samples When compared to alternative techniques in its ability to estimate relative cell number, The described technique produced the most accurate results with the highest correlation score (Figure 15G) and the lowest mean average error (MAE) (Figure 15H). Many cell types were achieved. Only the techniques described herein were able to identify CD4+ T cells and control The study was accurate in reconstituting active T cells, providing average Pearson correlation values ​​of up to 0.87 and 0.95. (Figure 15G). Thus, although these cell types have a high number of overlapping genes, The techniques developed by the inventors have successfully produced more accurate results than alternative algorithms. Drop.

[0403] [Table 11A]

[0404]

Table 11B

[0405]

Table 11C

[0406]

Table 11D

[0407]

Table 11E

[0408]

Table 11F

[0409]

Table 11G

[0410]

Table 11H

[0411]

Table 11I

[0412]

Table 11J

[0413]

Table 11K

[0414]

Table 11L

[0415] [Table 11M]

[0416] Thus, several aspects and embodiments of the techniques described in this disclosure have been described. It should be understood that various changes, modifications, and improvements will readily occur to those skilled in the art. Such changes, modifications, and improvements are within the spirit and scope of the technology described herein. For example, it is contemplated that one of ordinary skill in the art would be able to perform the functions described herein and and / or various other means for obtaining the result and / or one or more advantages. Any such variations and / or modifications will be readily envisioned within the scope of the present invention. It is considered to be within the scope of the embodiments described in the specification. It is possible to recognize or use no more than experimentation, numerous equivalents to the specific embodiments described herein. It will be appreciated that the above embodiments are provided by way of example only. Within the scope of the appended claims and their equivalents, embodiments of the present invention are specifically described. It should be understood that the present invention may be practiced other than as described herein. Any two or more of the features, systems, articles, materials, kits, and / or methods described herein may be Any combination of such features, systems, articles, materials, kits and / or methods may be combined in any manner that is inconsistent with one another. Where not inconsistent, they are within the scope of this disclosure.

[0417] The above-described embodiments can be implemented in any of numerous ways. One or more aspects and embodiments of the present disclosure may include a device (e.g., a computer, a processor, The program instructions are executable by a processor, processors, or other devices to execute a process or In this respect, various inventive concepts can be implemented or controlled. When implemented on one or more computers or other processors, the various implementations described above One or more programs implementing a method for implementing one or more of the aspects of the present invention. A computer-readable storage medium (or a plurality of computer-readable storage media) (e.g., computer memory, one or more floppy disks, compact Disks, optical disks, magnetic tapes, flash memories, field programmable gates Circuit configurations in chip arrays or other semiconductor devices, or other tangible computer The computer-readable medium or media may be embodied as a storage medium. The program or programs to be implemented may be executed on one or more different computers or other processes. The computer-implemented software may be transportable so that it can be loaded into a computer and used to implement various of the above aspects. In some embodiments, the computer-readable medium may be a non-transitory medium.

[0418] The terms "program" and "software" are used in their general sense in this specification. and may be used to configure a computer or other processor to implement various aspects as described above. Any kind of computer code that can be used to program a computer It should be further understood that, according to one aspect, One or more computer programs that, when executed, perform the methods of the present disclosure may be implemented as a single It does not have to reside on a single computer or processor to implement various aspects of the disclosure. For this reason, it can be distributed in a modular manner among several different computers or processors.

[0419] Computer-executable instructions may, for example, be implemented on one or more computers or other devices. The information may take many forms, such as a program module executed by a processor. Program modules include programs that perform particular tasks or implement particular abstract data types. This includes routines, programs, objects, components, data structures, etc. Typically, the functionality of the program modules may be combined or adapted as desired in various embodiments. It can be dispersed.

[0420] Also, data structures may be stored in computer-readable media in any suitable form. For simplicity of illustration, the data structure is organized as follows: Such a relationship may be represented as having a field in the field's storage. By assigning a location in a computer readable medium that conveys the relationship between the However, pointers, tags, or other mechanisms that establish relationships between data elements may be used. The information in the fields of the data structure may be retrieved using any suitable mechanism, including by use of Establish relationships between information sources.

[0421] When implemented in software, the software code may be provided to a single computer. Any suitable processor, whether centralized or distributed across multiple computers. Or it can be run on a collection of processors.

[0422] Furthermore, it will be appreciated that a computer may be implemented in any of several forms, e.g. Non-exclusive examples include rack-mounted computers, desktop computers, rack The present invention can be embodied in a laptop computer or a tablet computer. on devices that are not generally considered computers but have suitable processing power. The device may be implemented in a personal digital assistant (PDA), smartphone, tablet, or other Suitable portable or stationary electronic devices are included.

[0423] A computer may also have one or more input and output devices. The device can be used, among other things, to present a user interface. Examples of output devices that can be used to provide an interface include visual presentation of the output. a printer or display screen for and a speaker for audible presentation of the output or other sound generating devices. Examples of input devices that can be used include a keyboard and a pointing device, such as a mouse. Examples include touch pads and discretized tablets. Another example is a computer that can be operated by voice. The input information may be received in a recognizable or other audible form.

[0424] Such computers may be connected to a local area network, such as a business network, or Area Networks, Intelligent Networks (IN) or Internet The networks may be interconnected by one or more networks of any suitable form, including Such a network may be based on any suitable technology and may use any suitable protocol. It can operate according to the standard and can be used over wireless, wired, or fiber optic networks. It may contain a line.

[0425] Also, as described, some aspects can be embodied as one or more methods. The acts carried out as part of the Act may be ordered in any suitable way. Therefore, embodiments can be constructed in which the acts are performed in an order different from that illustrated, which may include Some acts may be performed simultaneously even though they are shown in the illustrative embodiment as sequential acts. and

[0426] All definitions as defined and used herein are taken from dictionary definitions, by reference. Definitions in incorporated documents and / or ordinary meanings of defined terms are intended to control the It should be understood.

[0427] As used herein and in the claims, indefinite coronary artery The words "a" and "an" are understood to mean "at least one" unless expressly indicated to the contrary. It should be understood.

[0428] As used herein and in the claims, "and The term "or" means "either or both" of the elements so conjoined, i.e., It is understood to mean elements which are conjointly present in some cases and disjunctively present in other cases. Multiple elements listed with "and / or" are treated in the same manner, i.e. should be construed as "one or more" of the elements so conjoined. Elements other than those specifically identified by the clauses are the same as those specifically identified elements. It may be present in any form, whether related or unrelated. For example, a reference to "A and / or B" may be used in conjunction with open-ended language such as "including." When used, in one embodiment, only A (optionally including elements other than B) is used in another embodiment. In one embodiment, only B (optionally including elements other than A) is used, and in yet another embodiment, both A and B are used. It may refer to a method (optionally including other elements) etc.

[0429] As used herein and in the claims, one or more With respect to a list of multiple elements, the phrase "at least one" refers to any should be understood to mean at least one element selected from one or more elements of but not necessarily at least one of every element specifically listed in the list of elements. It is not intended to be inclusive or exclusive of any combination of elements in the list of elements. This definition also includes all elements, whether related or not to those elements specifically identified. not include any element specifically identified in the list of elements referred to by the phrase "at least one" It is possible that other elements may be optionally present. Thus, as a non-limiting example, "A and at least one of B" (or, equivalently, "A or at least one of B" or equivalently In one embodiment, "at least one of A and / or B" refers to at least one, optionally , more than one A and no B (optionally including elements other than B), In this embodiment, there is at least one, and optionally more than one, B, and no A (and optionally In yet another embodiment, at least one, optionally more than one, contains more than one A and at least one, optionally more than one B (and optionally , and other elements).

[0430] In the claims and in the above specification, "comprising" and "including," "carrying," "having," "containing" "," "involving," "holding," "composed of" All transitional phrases, such as ", " ...) "consisting of" and "essentially consisting of" should be understood to mean Only the transitional phrase "consisting essentially of" is used in the closed or semi-closed This is a transitional phrase.

[0431] The terms "approximately," "substantially," and "about" refer in some embodiments to within ± the target value. 20%, in some embodiments within ±10% of the target value, and in some embodiments within ±5% of the target value "Approximately" may be used to mean within ±2% of a target value in some embodiments. The terms "substantially" and "about" can be inclusive of the target value. [Explanation of symbols]

[0432] 100 Systems 102 Biological specimens 104 Sequencing Platform 106 Sequence Information 108 Computer Devices 110 Cell composition ratio 122 Cell type A 124 Sequence information 126 Model A 128 Cell composition ratio 132 Cell type B 134 Sequence information 136 Model B 138 Cell composition ratio 140 t-SNE plot 142 Subtype A 144 Sequence information 146 Model C 148 Cell composition ratio 150 t-SNE plots 152 Tumor cells 156 Model D 158 Cell composition ratio 160 cell types 162 Subtype B 164 Sequence Information 170 Gene Expression 180 Cell population 182 Tumor cell lines 190 Genes 192 genes 200 Methods, Processes 202 Operation 204 Operation 206 Operation 212 operation 214 operation 216 operation 216a operation 216b operation 218 operation 220 Operation, implementation example, method 232 operation 234 operation 236 operation 302 Primary tumor samples 304 Nonlinear Regression Models 306 RNA ratio 308 Cell type A 310 Cell type B 312 Cell type C 314 Expression Data 316 Expression Data 318 Expression Data 320 Nonlinear Regression Models 322 Nonlinear Regression Models 324 Nonlinear Regression Models 326 1st submodel 328 1st submodel 330 1st submodel 332 1st value 334 First Value 336 1st value 338 Second submodel 340 Second submodel 342 Second submodel 344 Second Value 346 Second Value 348 Second Value 350 formula 360 RNA ratio 370 Cell composition ratio 380 method 382 First Step 384 Second Process 386 Third Step 400 ways 402 operation 404 operation 406 operation 408 operation 500 ways 502 operation 510 operation 520 operation 530 Diagram 540 Diagram 550 parameters 600 ways 602 operation 604 Rebalancing 608 operation 610 Branch 612 operation 614 operation 620 Branch 630 Branch 640 Branches, Servers 650 Branch 702 RNA expression data 704 RNA expression data 800 ways 802 operation 804 operation 806 operation 1000 Computer Systems 1010 Processor 1020 Memory 1030 Non-volatile storage media 1040 Network Input / Output (I / O) Interface 1050 User I / O Interface 1100 Environment 1110 Network 1120 Database 1130 Computer Devices 1140 Server 1150 Doctor 1160 Database 1170 Laboratory 1180 Target

Claims

1. Using at least one computer hardware processor, Obtaining expression data for a biological sample, said biological sample having been previously obtained from a subject. the expression data being a first set of genes associated with a first cell type; comprising expression data of one of the steps; The expression data and one or more nonlinear regression models, including a first nonlinear regression model. determining a first cellular composition ratio for said first cell type using said first indicates an estimated proportion of cells of the first cell type in the biological sample, The step of determining the first cell composition ratio for a first cell type includes: The first expression data is processed by the first nonlinear regression model to obtain the first determining the first cellular composition ratio for a cell type; and outputting the first cell constituent ratio A process including A process for carrying out The method includes:

2. The subject has, is suspected of having, or is at risk of having cancer.

2. The method of claim 1 .

3. The method of claim 1 or 2, wherein the expression data is RNA expression data.

4. Processing the first expression data with the first nonlinear regression model comprises: providing the first expression data as an input to the first nonlinear regression model; obtaining a corresponding output representing an estimated proportion of RNA from said cell types; determining a proportion of RNA from the first cell type based on the estimated proportion of RNA from the first cell type; determining a first cell composition ratio; 4. The method of claim 3 or any one of claims 1 or 2, comprising:

5. The expression data includes a first set of genes associated with the first cell type. Contains expression data for 2, The first nonlinear regression model is said prediction of RNA expression from said first cell type using said first expression data as input; a first sub-model configured to generate a first value for the ratio; second expression data and said first value for said estimated proportion of RNA from said first cell type and a second value for the estimated proportion of RNA from the first cell type using as input a second sub-model configured to generate 5. The method of claim 4 or any one of claims 1 to 3, comprising:

6. The expression data is for a second gene associated with a second cell type distinct from the first cell type. second expression data associated with the set; the one or more nonlinear regression models include a second nonlinear regression model; The method includes, at least in part, determining a second cellular composition for the second cell type. processing said second expression data with said second nonlinear regression model to obtain determining the second cell composition ratio for the second cell type; 6. The method according to claim 1 or any one of claims 2 to 5.

7. The first cell type may be a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibroblast, a lymphocyte, a cell selected from the group consisting of leukocytes, macrophages, monocytes, NK cells, neutrophils, and T cells. The method according to claim 1 or any one of claims 2 to 6.

8. The first expression data is a set of genes for the first cell type in Table 2. The method according to claim 1 or claim 2, further comprising: The method according to any one of items 2 to 7.

9. The expression data includes expression data associated with a plurality of gene sets associated with each of a plurality of cell types. the plurality of gene sets includes data that correlates the first gene set with the first cell type; and said plurality of cell types comprising the one or more nonlinear regression models include a plurality of nonlinear regression models; The method further comprises: determining a plurality of cellular composition ratios for a cell type; includes the first cellular constituent ratio, and the step of determining the plurality of cellular constituent ratios includes For each cell type of the plurality of cell types, at least in part, A nonlinear regression of each of the plurality of nonlinear regression models to determine the cellular composition ratio. The model is used to process expression data associated with a set of genes associated with said cell type. determining the respective cell composition ratios for the cell types; 9. The method according to claim 1 or any one of claims 2 to 8.

10. The genes in the plurality of gene sets are selected from the group of genes in Table 2. The step of determining the plurality of cellular constituent ratios comprises the step of: 9 or claim 1, comprising processing expression data for at least 25 genes.

9. The method according to any one of claims 1 to 8.

11. The genes in the plurality of gene sets are selected from the group of genes in Table 2. The step of determining the plurality of cellular constituent ratios comprises the step of: 9 or claim 1, comprising processing expression data for at least 35 genes.

11. The method according to any one of claims 8 to 10.

12. The genes in the plurality of gene sets are selected from the group of genes in Table 2. The step of determining the plurality of cellular constituent ratios comprises the step of: 9 or claim 1, comprising processing expression data for at least 50 genes. 8 or 10 or 11. A method according to any one of claims 1 to 8, 10 or 11.

13. The genes in the plurality of gene sets are selected from the group of genes in Table 2. The step of determining the plurality of cellular constituent ratios comprises: 9 or claim 1, comprising processing expression data for at least 75 genes.

13. The method according to any one of claims 1 to 8 or 10 to 12.

14. The genes in the plurality of gene sets are selected from the group of genes in Table 2. The step of determining the plurality of cellular constituent ratios comprises the steps of: The method of claim 9 or claim 10, further comprising the step of processing expression data for at least 100 genes. The method according to any one of claims 1 to 8 or 10 to 13.

15. The one or more nonlinear regression models may be one or more random forest regression models.

15. The method of claim 1 or any one of claims 2 to 14, comprising:

16. The one or more nonlinear regression models may be one or more neural network regression models. The method of claim 1 or any one of claims 2 to 15, comprising:

17. The one or more nonlinear regression models may be one or more support vector machine regression models. The method of claim 1 or any one of claims 2 to 16, comprising:

18. The first nonlinear regression model is determined, at least in part, by: obtaining simulated expression data; The simulated expression data is used to train the first nonlinear regression model. The process and 18. The method of claim 1 or any one of claims 2 to 17, wherein the method is trained by

19. obtaining said simulated expression data; The simulated expression data is used to train the first nonlinear regression model. The process and 19. The method of claim 18 or any one of claims 1 to 17, further comprising:

20. The step of obtaining the simulated expression data comprises: generating the simulated expression data, Obtaining a set of RNA expression data from one or more biological samples, said RNA expression data comprising: the set of data includes microenvironment cell expression data and malignant cell expression data; Simulated microenvironment cell expression data using the microenvironment cell expression data. and Using the malignant cell expression data, a simulated malignant cell expression data is generated. and The simulated microenvironment cell expression data and the simulated malignant cell expression data. and combining the current data to generate at least a portion of the simulated expression data. The process of 20. The method of claim 18 or any one of claims 1 to 17 or claim 19, comprising:

21. The expression profile for the first cell type and the first determining a malignant tumor expression profile using the cellular composition ratios of 21. The method of claim 1 or any one of claims 2 to 20, further comprising:

22. The first nonlinear regression model is Obtaining training data comprising simulated RNA expression data, The rated RNA expression data is for the first set of genes associated with the first cell type. comprising a first RNA expression data set for Training the first nonlinear regression model to estimate the proportion of RNA from the first cell type. The training step includes: Using the first nonlinear regression model and the first RNA expression data, generating an estimated ratio of RNA from each type; and said first nonlinear regression model using said estimated proportion of RNA from said first cell type; The process of updating the parameters A process including 22. The method of claim 1 or any one of claims 2 to 21, wherein the method is trained by

23. at least one hardware processor; When executed by the at least one hardware processor, One hardware processor, Obtaining expression data for a biological sample, said biological sample having been previously obtained from a subject. wherein the expression data is associated with a first set of genes associated with a first cell type. comprising expression data of one of the steps; The expression data and one or more nonlinear regression models, including a first nonlinear regression model. determining a first cellular composition ratio for said first cell type using said a cellular composition ratio of 1 indicates an estimated ratio of cells of the first cell type in the biological sample; The step of determining the first cell composition ratio for the first cell type includes: The first expression data is processed by the first nonlinear regression model to obtain the first determining the first cellular composition ratio for a cell type; and outputting the first cell constituent ratio A process including at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium A system including:

24. When executed by at least one hardware processor, the at least one The hardware processor Obtaining expression data for a biological sample, said biological sample having been previously obtained from a subject. wherein the expression data is a first set of genes associated with a first cell type. and using the expression data and one or more nonlinear regression models, including a first nonlinear regression model; determining a first cellular composition ratio for the first cell type using the first indicates an estimated ratio of cells of the first cell type in the biological sample; The step of determining the first cell composition ratio for the first cell type includes: The first expression data is processed by the first nonlinear regression model to obtain the first cell determining the first cell composition ratio for a cell type; and outputting the first cell constituent ratio A process including at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium.

25. Using at least one computer hardware processor, Obtaining RNA expression data for a biological sample, the biological sample comprising a cancer have been obtained previously from subjects suspected of having or at risk of having cancer 、 The RNA expression data includes a first set of genes associated with a first cell type. Contains RNA expression data, The first RNA expression data is collected from the gene expression data for the first cell type in Table 2. expression data for at least 10 genes selected from the group of The first cell type may be a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibroblast, a phospholipid, or a phospholipid. selected from the group consisting of lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells; The process and The first RNA expression data is used to determine a first cellular fraction for the first cell type. The first cell composition ratio is a step of determining the first cell type in the biological sample. determining said first cellular composition ratio for said first cell type; The procedure is providing the first RNA expression data as an input to a first nonlinear regression model; obtaining a corresponding output representing an estimated proportion of RNA from the cell types; and a pre-selection step for the first cell type based on the estimated proportion of RNA from the first cell type; The step of determining the first cell composition ratio. A process including The method includes the step of:

26. The RNA expression data is associated with the first set of genes associated with the first cell type. a second RNA expression data set, The first nonlinear regression model is Using the first RNA expression data as input, the prediction of RNA from the first cell type is performed. a first sub-model configured to generate a first value for the constant ratio; said second expression data and said first comparison of said estimated proportion of RNA from said first cell type; and a second estimation of the estimated proportion of RNA from the first cell type using the values ​​of a second sub-model configured to generate values ​​of 26. The method of claim 25, comprising:

27. The RNA expression data includes a second RNA expression data set associated with a second set of genes associated with a second cell type. A expression data included, The second RNA expression data is for the genes in Table 2 for the second cell type. The method comprises the steps of: The second cell type may be a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibroblast, a lymphocyte, a cytoplasmic reticulocyte, a macrophage, a monocyte, a NK cell, a neutrophil, and a T cell; and To The step of determining a second cellular composition ratio for the second cell type includes: The second RNA expression data is processed by a second nonlinear regression model to obtain determining the second cell composition ratio for the type; 27. The method of claim 25 or any one of claims 1 to 22 or claim 26, comprising:

28. The RNA expression data is R associated with a plurality of gene sets associated with each of a plurality of cell types. NA expression data, the plurality of gene sets being a first gene set and a first cell type. and the plurality of cell types comprising The method further comprises: using the RNA expression data associated with the plurality of gene sets to The method further comprises determining a plurality of cellular constituent ratios for a plurality of cell types, The composition ratio includes the first cellular composition ratio, and the step of determining the plurality of cellular composition ratios includes: For each cell type of the plurality of cell types, at least in part, Using each nonlinear regression model to determine the cell composition ratio associated with the cell type. The RNA expression data associated with a set of genes corresponding to the cell type is processed to identify the determining the ratio of each cell component The method according to any one of claims 25 or claims 1 to 22 or claims 26 or 27, Law.

29. 26. The method of claim 25, wherein the first nonlinear regression model comprises a random forest regression model.

29. A method according to any one of claims 1 to 22 or claims 26 to 28.

30. 26. The method of claim 25, wherein the first nonlinear regression model comprises a neural network regression model. Or a method according to any one of claims 1 to 22 or claims 26 to 29.

31. 26. The method of claim 25, wherein the first nonlinear regression model comprises a support vector machine regression model. Or a method according to any one of claims 1 to 22 or claims 26 to 30.

32. The first nonlinear regression model is based, at least in part, on simulated RNA expression data. generating training data including the training data, The process is as follows: Obtaining a set of RNA expression data from one or more biological samples, said RNA expression data comprising: The set of data includes microenvironment cell RNA expression data and malignant cell RNA expression data. and, Using the microenvironment cellular RNA expression data, a simulated microenvironment cellular RNA expression generating data; The malignant cell RNA expression data is used to generate simulated malignant cell RNA expression data. A generating step; The simulated microenvironment cellular RNA expression data and the simulated malignant cell and combining said simulated RNA expression data with said cellular RNA expression data. The process of creating a part 32. The method of claim 25 or any one of claims 1 to 22 or claims 26 to 31, comprising:

33. an RNA expression profile for said first cell type and determining a malignant tumor expression profile using the first cellular constituent proportion; The method according to any one of claims 25 or claims 1 to 22 or claims 26 to 32, further comprising Law.

34. The first RNA expression data comprises at least one gene selected from the group of genes in Table 2. Claim 25 or claims 1 to 22 or claim 26, comprising expression data for at least 25 genes.

34. The method of claim 33,

35. The first RNA expression data comprises at least one gene selected from the group of genes in Table 2. Claim 25 or claims 1 to 22 or claim 26, comprising expression data for at least 50 genes.

35. The method of claim 34,

36. The first RNA expression data comprises at least one gene selected from the group of genes in Table 2. Claim 25 or claims 1 to 22 or claim 26, comprising expression data for at least 100 genes.

35. The method of claim 1, wherein

37. The first nonlinear regression model is Obtaining training data comprising simulated RNA expression data, The rated RNA expression data is for the first set of genes associated with the first cell type. and Training the first nonlinear regression model to estimate the proportion of RNA from the first cell type. The training step includes: Using the first nonlinear regression model and the second RNA expression data, generating an estimated ratio of RNA from each type; and said first nonlinear regression model using said estimated proportion of RNA from said first cell type; The process of updating the parameters and Any one of claims 25, 1 to 22, or 26 to 36, The method according to claim 5.

38. at least one hardware processor; When executed by the at least one hardware processor, One hardware processor, Obtaining RNA expression data for a biological sample, the biological sample being from a subject having cancer. previously obtained from subjects suspected of having or at risk of having cancer. the law of nature, The RNA expression data includes a first set of genes associated with a first cell type. Contains RNA expression data of The first RNA expression data is generated for the first cell type in Table 2. expression data for at least 10 genes selected from the group of genes; The first cell type may be a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibroblast, a lymphocyte, or a lysosome. lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells. , process, The first RNA expression data is used to determine a first cellular fraction for the first cell type. The step of determining the first cell composition ratio is determining said first cellular composition ratio for said first cell type; The process is as follows: providing the first RNA expression data as an input to a first nonlinear regression model; obtaining a corresponding output representing an estimated proportion of RNA from the first cell type; and for the first cell type based on the estimated proportion of RNA from the first cell type. determining the first cell composition ratio A process including at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium A system including:

39. When executed by at least one hardware processor, the at least one The hardware processor Obtaining RNA expression data for a biological sample, the biological sample comprising a cancer have been obtained previously from subjects suspected of having or at risk of having cancer 、 The RNA expression data includes a first set of genes associated with a first cell type. Contains RNA expression data, The first RNA expression data is collected from the gene expression data for the first cell type in Table 2. expression data for at least 10 genes selected from the group of The first cell type may be a B cell, a CD4+ T cell, a CD8+ T cell, an endothelial cell, a fibroblast, a phospholipid, or a phospholipid. selected from the group consisting of lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells; The process and The first RNA expression data is used to determine a first cellular fraction for the first cell type. The first cell composition ratio is a step of determining the first cell type in the biological sample. determining said first cellular composition ratio for said first cell type; The procedure is providing the first RNA expression data as an input to a first nonlinear regression model; obtaining a corresponding output representing an estimated proportion of RNA from the cell types; and a pre-selection step for the first cell type based on the estimated proportion of RNA from the first cell type; The step of determining the first cell composition ratio. A process including at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium.

40. Using at least one computer hardware processor, Obtaining training data comprising simulated RNA expression data, The computed RNA expression data includes a first RNA sequence for a first gene associated with a first cell type. A. expression data for a second gene associated with a second cell type distinct from the first cell type. and Multiple nonlinear regression models were used to estimate the proportion of RNA from each of one or more cell types. training a plurality of nonlinear regression models to identify RNA from the first cell type; a first nonlinear regression model for estimating the proportion of RNA from said second cell type; a second nonlinear regression model for estimating the first nonlinear regression model, and training the plurality of nonlinear regression models. The step of computing the first nonlinear regression model at least in part by: Using the first nonlinear regression model and the first RNA expression data, generating an estimated proportion of RNA from each cell type; and The estimated proportion of RNA from the first cell type is used to generate the first nonlinear regression model. The process of updating the parameters of the training the sample by The trained composite including the first nonlinear regression model and the second nonlinear regression model. outputting a nonlinear regression model of the number of The method includes the step of:

41. The step of obtaining training data comprises: Obtaining a set of RNA expression data from one or more biological samples, said RNA expression data comprising: The current data set includes RNA expression data from microenvironment cells and RNA expression data from malignant cells. As Based on the microenvironment cell RNA expression data, a simulated microenvironment cell RNA expression obtaining current data; Simulated malignant cell RNA expression data based on the malignant cell RNA expression data. and The simulated microenvironment cellular RNA expression data and the simulated malignant cellular RNA expression data, Also, the process of creating a part generating said at least a portion of said simulated RNA expression data by 41. The method of claim 40 or any one of claims 1 to 22 or claims 25 to 37, comprising:

42. Prior to training the plurality of nonlinear regression models, the simulated RNA expression data The process of adding noise to Any one of claims 41 or claims 1 to 22 or claims 25 to 37 or claim 40 further comprising The method according to claim 5.

43. 42. The method of claim 42, wherein the noise comprises at least one of Poisson noise or Gaussian noise. or any one of claims 1 to 22, claims 25 to 37, or claims 40 or 41 method.

44. The step of generating the simulated microenvironment cellular RNA expression data includes: For a first microenvironment cell type, a first portion of the microenvironment cell RNA expression data is used to generating a first RNA expression profile. Claim 41 or claims 1 to 22 or claims 25 to 37 or claim 40 or claim 42 or 44. The method of claim 43.

45. The first portion of the microenvironment cell RNA expression data comprises a plurality of the first microenvironment cell types. 44 or claims 1 to 22 or claims 25 to 30, comprising RNA expression data from subtypes of 37 or a method according to any one of claims 40 to 43.

46. The step of generating the first RNA expression profile comprises: resampling the first portion of the microenvironment cellular RNA expression data using a number of subtypes. The method according to claim 45 or claim 1 to 22 or claim 25 to 37 or claim 40, comprising the step of ringing.

45. The method according to any one of claims 44 to 44.

47. The first portion of the microenvironment cellular RNA expression data comprises RNA expression data from a plurality of samples. Claim 44 or claims 1 to 22 or claims 25 to 37 or claims 40 to 43 or claim 4 5 or 46. The method of any one of claims 5 or 46.

48. The step of generating the first RNA expression profile comprises: resampling a first portion of the microenvironment cellular RNA expression data by taking as input The method according to claim 47 or claim 1 to 22 or claim 25 to 37 or claim 40, comprising the step of ringing.

46. ​​The method of claim 45, wherein

49. The step of generating the simulated microenvironment cellular RNA expression data includes: For a second microenvironment cell type, using a second portion of the microenvironment cell RNA expression data. generating a second RNA expression profile; The first RNA expression profile and the second RNA expression profile are combined to obtain a generating at least some of the simulated microenvironment cellular RNA expression data; and Claim 44 or claims 1 to 22 or claims 25 to 37 or claims 40 to 43 or claims 49. The method of any one of claims 45 to 48.

50. The first RNA expression profile and the second RNA expression profile are combined to obtain a generating at least some of the simulated microenvironment cellular RNA expression data; The process is as follows: determining a weighted sum of the first RNA expression profile and the second RNA expression profile. Process Any of claims 49, 1 to 22, 25 to 37 or 40 to 48, 13. The method according to claim 1.

51. The malignant cell RNA expression data comprises RNA expression data from a plurality of malignant cell samples. Any of claims 41, 1 to 22, 25 to 37, 40, or 42 to 50 13. The method according to claim 1.

52. The step of generating the simulated malignant cell RNA expression data includes: 51 or claim 1 to 22, comprising combining the RNA expression data from cell samples. Or a method according to any one of claims 25 to 37 or claims 40 to 50.

53. The step of generating the simulated malignant cell RNA expression data comprises: 41 or claim 1 to 22, comprising adding noise to the malignant cell RNA expression data obtained. Or a method according to any one of claims 25 to 37 or claim 40 or claims 42 to 52.

54. The coefficients of the weighted sum are determined using the output of a pre-trained nonlinear regression model. , claim 50, or claims 1 to 22, or claims 25 to 37, or claims 40 to 49, or claims 51 to 5 4. The method according to any one of claims 3 to 4.

55. The first RNA expression data is collected from the gene(s) for the first cell type in Table 2.

40. The method according to claim 40, further comprising:

55. A method according to any one of claims 1 to 22 or claims 25 to 37 or claims 41 to 54.

56. The second RNA expression data is for the genes in Table 2 for the second cell type.

40. The method according to claim 40, further comprising:

10. A method according to any one of claims 1 to 22 or claims 25 to 37 or claims 41 to 55.

57. The first cell type and the second cell type are B cells, CD4+ T cells, CD8+ T cells, endothelial cells, , consisting of fibroblasts, lymphocytes, macrophages, monocytes, NK cells, neutrophils, and T cells Selected from the group consisting of claim 40 or claims 1 to 22 or claims 25 to 37 or claims 41 to 56 2. The method according to claim 1 .

58. The simulated RNA expression data includes the first gene associated with the first cell type. The method of claim 40 or claim 1 to 22 or claim 25 to 3, further comprising the step of: 7 or a method according to any one of claims 41 to 57.

59. The first nonlinear regression model of the one or more nonlinear regression models: said estimation of RNA expression from said first cell type using said first RNA expression data as input; a first sub-model configured to generate a first value for the ratio; and said second RNA expression data and said estimated proportion of RNA from said first cell type; 1 and a value of 0.1 as input to generate a second estimate of the estimated proportion of RNA from the first cell type. A second submodel configured to generate values ​​of any of claims 58, 1 to 22, 25 to 37 or 40 to 57, 13. The method according to claim 1.

60. The second sub-model is a submodel for RNA from each of a plurality of cell types other than the first cell type. Using the estimated proportion as input, a priori analysis is performed on the estimated proportion of RNA from the first cell type. 59 or claims 1 to 22 or claims 59, further configured to generate the second value.

59. The method of any one of claims 25 to 37 or 40 to 58.

61. at least one computer hardware processor; When executed by the at least one computer hardware processor, The at least one computer hardware processor includes: Obtaining training data comprising simulated RNA expression data, The computed RNA expression data includes a first RNA sequence for a first gene associated with a first cell type. A. expression data for a second gene associated with a second cell type distinct from the first cell type. and Multiple nonlinear regression models were used to estimate the proportion of RNA from each of one or more cell types. training a plurality of nonlinear regression models to identify RNA from the first cell type; a first nonlinear regression model for estimating the proportion of RNA from said second cell type; a second nonlinear regression model for estimating the first nonlinear regression model, and training the plurality of nonlinear regression models. The process comprises, at least in part: Using the first nonlinear regression model and the first RNA expression data, generating an estimated proportion of RNA from each cytoplasmic type; and The estimated proportion of RNA from the first cell type is used to generate the first nonlinear regression model. The process of updating the parameters of the training the first nonlinear regression model by The trained composite including the first nonlinear regression model and the second nonlinear regression model. outputting a nonlinear regression model of the number of at least one non-transitory computer-implemented memory device storing processor-executable instructions for causing the A data-readable storage medium A system including:

62. When executed by at least one computer hardware processor, At least one computer hardware processor Obtaining training data comprising simulated RNA expression data, The rated RNA expression data includes a first RNA for a first gene associated with a first cell type. Expression data and the first cell type for a second gene associated with a different second cell type and Multiple nonlinear regression models to estimate the proportion of RNA from each of one or more cell types. training said plurality of nonlinear regression models to determine a degree of fidelity of RNA from said first cell type; A first nonlinear regression model for estimating the proportion of RNA from the second cell type. a second nonlinear regression model for determining whether the plurality of nonlinear regression models is a plurality of nonlinear regression models; The process is, at least in part, Using the first nonlinear regression model and the first RNA expression data, generating an estimated ratio of RNA from each type; and said first nonlinear regression model using said estimated proportion of RNA from said first cell type; The process of updating the parameters training the first nonlinear regression model by The trained plurality of models including the first nonlinear regression model and the second nonlinear regression model. A step of outputting a nonlinear regression model of At least one non-transitory computer program that stores processor-executable instructions to cause a program to execute a program. Data-readable storage medium.

63. Using at least one computer hardware processor, Obtaining expression data for a biological sample, the biological sample having cancer. have been obtained prospectively from subjects suspected of having or at risk of having cancer; The process and Obtaining a plurality of expression profiles for a corresponding plurality of cell types, Each of the current profiles includes one or more of the following: comprising respective expression data from a plurality of genes; At least in part, a piecewise correlation between the expression data and the plurality of expression profiles. A plurality of cell structures for the plurality of cell types are obtained by optimizing a continuous error function. A process to determine the composition ratio The method includes the step of carrying out the above.

64. The expression data is RNA expression data, and the plurality of expression profiles are RNA expression profiles.

64. The method of claim 63, wherein the file is a

65. The step of determining the plurality of cellular constituent ratios for the plurality of cell types comprises: wherein the error value is determined using a piecewise continuous error function. The method according to claim 63.

66. The step of determining the plurality of cellular constituent ratios for the plurality of cell types includes:

66. The method of claim 65, further comprising minimizing the weighting.

67. The one or more genes may be less than 5000 genes in Table 2.

64. The method of claim 63, wherein the gene is composed of at least two genes.

68. The plurality of expression profiles for the corresponding plurality of cell types and the plurality of cells determining a malignant tumor expression profile using the constituent ratios; Claim 63 or claims 1 to 22 or claims 25 to 37 or claims 40 to 60 or claims 68. The method of any one of claims 64 to 67.

69. at least one computer hardware processor; When executed by the at least one computer hardware processor, The at least one computer hardware processor includes: Obtaining expression data for a biological sample, the biological sample having cancer. have been obtained prospectively from subjects suspected of having or at risk of having cancer; The process and Obtaining a plurality of expression profiles for a corresponding plurality of cell types, Each of the current profiles includes one or more of the following: comprising respective expression data from a plurality of genes; At least in part, a piecewise correlation between the expression data and the plurality of expression profiles. By optimizing a continuous error function, multiple cell configurations are obtained for the multiple cell types. determining the ratio; at least one computer readable memory storing processor executable instructions for causing the Storage media and A system including:

70. When executed by at least one computer hardware processor, At least one computer hardware processor Obtaining expression data for a biological sample, the biological sample having cancer, Previously obtained data from subjects suspected of having or at risk of having cancer. With Obtaining a plurality of expression profiles for a corresponding plurality of cell types, Each of the profiles comprises one or more sequences associated with each cell type from the plurality of cell types. including respective expression data from a number of genes; At least in part, a piecewise correlation between the expression data and the plurality of expression profiles. A plurality of cell composition ratios for the plurality of cell types are obtained by optimizing a continuous error function. determining a ratio; at least one computer readable memory storing processor executable instructions for causing the A capable storage medium.

Images