DNA copy number aberrations (CNAs) for determining cancer phenotype

Abstract

The present disclosure provides methods for generating a computed cancer signature of a cancer-associated phenotype based on copy number aberrations (CNAs) in a patient sample. The computed cancer signature may correspond to a somatic mutation, an mRNA expression signature, or a protein expression signature. The present disclosure also provides methods for treating a patient using the computed cancer phenotype. In addition, the present disclosure provides methods for generating a computed signature based on CNAs to reproduce a cancer phenotype.

Description

[Technical field]

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims the benefit of U.S. Application No. 62 / 912,727, filed October 9, 2019, Perou et al., Attorney Docket No. 150-32-PROV, which is incorporated herein by reference in its entirety.

[0002] Federally Sponsored Research or Development Claims This invention was made with Government support under Grant Nos. CA58223, CA148761, and CA195740 awarded by the National Institutes of Health. The Government has certain rights in this invention.

[0003] Reference to a "Sequence Listing," table, or computer program listing appendix submitted as an ASCII text file This application contains fifteen (15) tables and one (1) computer program listing as exhibits, which have been submitted electronically via EFS Web as ASCII text files. They have the following file attributes: (1) Annotation of gene expression signatures entitled Supplementary Table 1.Supp_Table1.txt, which has a size of 2,042,486 bytes and was created on Mar. 4, 2019; (2) Annotation of copy number segments entitled Supplementary Table 2.Supp_Table2.txt, which has a size of 728,822 bytes and was created on Oct. 9, 2019; (3) Elastic net model summary of genes entitled Supplementary Table 3.Supp_Table3.txt, which has a size of 1,236,182 bytes and was created on Oct. 9, 2019; (4) Elastic net model summary of breast cancer molecular subtypes and histology entitled Supplementary Table 4.Supp_Table4.txt, which has a size of 37,355 bytes and was created on Oct. 9, 2019; (5) Supplementary Table 5.Supp_Table5.txt, which has a size of 37,355 bytes and was created on Oct. 9, 2019; (6) a summary of the Elastic Net Model of Breast Cancer Protein Expression and Clinical Receptor Status entitled Supplementary Table 6.Supp_Table6.txt, which has a size of 120,280 bytes and was created on October 9, 2019; (7) a summary of the Elastic Net Model of Breast Cancer Somatic Mutations entitled Supplementary Table 7.Supp_Table7.txt, which has a size of 1,001,837 bytes and was created on October 9, 2019; (8) a summary of the Elastic Net Model of Lung Cancer Gene Expression Signatures entitled Supplementary Table 8.Supp_Table8.txt, which has a size of 1,199,707 bytes and was created on October 9, 2019; (9) a summary of the Elastic Net Model of Lung Cancer Gene Expression Signatures entitled Supp_Table9.(10) Gene Expression Signature Score of 1038 TCGA Breast Tumors entitled Supplementary Table 10.Supp_Table10.txt, which has a size of 6,875,948 bytes and was created on October 9, 2019; (11) Gene Expression Signature Score of 512 TCGA LUAD Tumors and 498 TCGA LUAD Tumors entitled Supplementary Table 11.Supp_Table11.txt, which has a size of 6,875,948 bytes and was created on October 9, 2019; (12) Gene Expression Signature Score of 1689 METABRIC Breast Tumors entitled Supplementary Table 12.Supp_Table12.txt, which has a size of 6,235,101 bytes and was created on October 9, 2019; (13) Gene Expression Signature Score of 972 Breast Tumors entitled Supplementary Table 13.Supp_Table13.txt, which has a size of 6,235,101 bytes and was created on October 9, 2019; (14) a summary of pan-cancer signature predictions entitled Supplementary Table 14.Supp_Table14.txt, which has a size of 577,688 bytes and was created on October 9, 2019; (15) a listing of amplicon signatures entitled Supplementary Table 15.Supp_Table15.txt, which has a size of 2,739 bytes and was created on October 9, 2019. (16) a computer program listing entitled helper.R, which has a size of 15,697 bytes and was created on September 25, 2019. All sixteen (16) files are incorporated herein by reference in their entirety.

[0004] 1. Field The present disclosure provides a method for generating a computed cancer signature of a cancer-related phenotype based on copy number aberrations (CNAs) of a patient sample. The computed cancer signature may correspond to a somatic mutation, an mRNA expression signature, or a protein expression signature. The present disclosure also provides a for method for treating a patient with the computed cancer phenotype. In addition, the present disclosure provides a method for generating a computed signature based on CNAs to recapitulate a cancer phenotype. [Background technology]

[0005] 2.Background 2.1. Introduction The "Background" description provided herein is for the purpose of generally presenting the context of the present disclosure. Works of the inventors named in this Background section, to the extent that they are described therein, and aspects of the present specification that may not otherwise qualify as prior art at the time of filing, are not admitted, expressly or implicitly, as prior art to the present disclosure. Tumor formation is often driven by multiple types of DNA abnormalities, leading to diseases of great complexity and heterogeneity. The ability to analyze this heterogeneity is crucial for understanding cancer mechanisms and for identifying patient subgroups for personalized treatment. One limitation to capturing this heterogeneity is in the characterization of disease phenotypes. Thanks to the efforts of many consortia, including The Cancer Genome Atlas (TCGA), large-scale multi-platform genomic data is now available, providing the opportunity to study cancer phenotypes at the molecular level and by using multiple technology types (Non-Patent Documents 1-4). 1-4 In particular, many gene expression signatures have been developed to define specific cancer phenotypes, ranging from proliferation rate to characteristics of the tumor microenvironment (Non-Patent Documents 5-7). 5-7These mRNA expression signatures, together with protein expression, somatic mutations, and clinical features, provide a comprehensive molecular portrait of the tumor. Integrating multi-platform genomic data together to unravel the relationships between genotype and phenotype is crucial to understanding the genetic causes behind tumor behavior (8). 8 Building predictive models of key tumor-driven phenotypes would be beneficial to stratify patients for personalized treatment. [Prior art documents] [Non-patent literature]

[0006] [Non-Patent Document 1] Perou, CM et al., Molecular portraits of human breast tumours, Nature, vol. 406, pp. 747-752, doi:10.1038 / 35021093 (2000). [Non-Patent Document 2] "Comprehensive molecular portraits of human breast tumours" Nature 490, 61-70, doi:10.1038 / nature11412 (2012) [Non-Patent Document 3] Curtis, C. et al., "The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups," Nature, 486, 346-352, doi:10.1038 / nature10983 (2012). [Non-Patent Document 4] Hoadley, KA et al., "Cell-of-Origin Patterns Dominate the Molecular Classification of 10,000 Tumors from 33 Types of Cancer," Cell 173, 291-304 e296, doi:10.1016 / j.cell.2018.03.022 (2018) [Non-Patent Document 5] Nevins, JR, & Potti, A., "Mining gene expression profiles: expression signatures as cancer phenotypes." Nature Reviews Genetics, 8, 601-609, doi:10.1038 / nrg2137 (2007). [Non-Patent Document 6] Bild, AH et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439: 353-357, doi:10.1038 / nature04296 (2006). [Non-Patent Document 7] Kwa, M., Makris, A., & Esteva, FJ. "Clinical utility of gene-expression signatures in early stage breast cancer." Nature Reviews Clinical Oncology, vol. 14, pp. 595-610, doi:10.1038 / nrclinonc.2017.74 (2017). [Non-Patent Document 8] Gatza, ML, Silva, GO, Parker, JS, Fan, C., & Perou, CM. An integrated genomics approach identifies drivers of proliferation in luminal-subtype human breast cancer. Nature genetics, 46, 1051-1059, doi:10.1038 / ng.3073 (2014). Summary of the Invention [Problem to be solved by the invention]

[0007] The present invention has been made to solve the problems in the conventional techniques described above. [Means for solving the problem]

[0008] 3. Disclosure Overview The present disclosure provides a method for generating a computed cancer signature of a patient sample, comprising: (a) obtaining or obtaining a sample from a patient; (b) measuring or measuring a plurality of copy number abnormalities (CNAs) at a plurality of locations on a plurality of chromosomes; and (c) analyzing the measured CNAs using a mathematical model based on mRNA expression data and molecular subtypes, the mathematical model being validated by at least two different statistical methods to generate a computed cancer signature of the sample. In one embodiment, more than 50 CNAs are measured, alternatively more than 100 CNAs are measured, alternatively between about 250 and about 400 CNAs are measured. In another embodiment, more than 400 CNAs are measured. The copy number abnormalities (CNAs) are obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0009] In the methods disclosed herein, the calculated cancer signature may correspond to a somatic mutation signature. The mathematical model for modeling the somatic mutation signature may be based on a beta coefficient value of 10 or more in Supplementary Table 6. Alternatively, the mathematical model may be based on a beta coefficient value of 20 or more, a beta coefficient value of 40 or more, a beta coefficient value of 60 or more, or a beta coefficient value of 100 or more. In another embodiment, the mathematical model may be based on the top 5%, top 10%, top 25%, or top 50% of the beta coefficient values.

[0010] In another embodiment of the methods disclosed herein, the calculated cancer signature may correspond to an mRNA expression signature, which may be a signature of a breast cancer subtype. The mathematical model for determining the breast cancer subtype signature may be based on a beta coefficient value of 10 or more in Supplementary Table 4. Alternatively, the mathematical model may be based on a beta coefficient value of 20 or more, a beta coefficient value of 40 or more, a beta coefficient value of 60 or more, or a beta coefficient value of 100 or more. In another embodiment, the mathematical model may be based on the top 5%, top 10%, top 25%, or top 50% of the beta coefficient values.

[0011] In yet another embodiment of the methods disclosed herein, the calculated cancer signature may correspond to a protein expression signature. The mathematical model for modeling the protein expression signature may be based on a beta coefficient value of 10 or greater in Supplementary Table 5. Alternatively, the mathematical model may be based on a beta coefficient value of 20 or greater, a beta coefficient value of 40 or greater, a beta coefficient value of 60 or greater, or a beta coefficient value of 100 or greater. In another embodiment, the mathematical model may be based on the top 5%, top 10%, top 25%, or top 50% of the beta coefficient values.

[0012] In the methods disclosed herein, the protein expression signature can be an immunohistochemistry (IHC) signature. The IHC signature can be an estrogen receptor (ER), epidermal growth factor receptor (EGFR), human epidermal growth factor receptor 2 (HER2), progesterone receptor (PR), or retinoblastoma (RB) signature.

[0013] The calculated cancer signature may correspond to a result from a commercial vendor of a cancer diagnostic. For example, the calculated cancer signature may correspond to a FoundationOne® CDX result, a MAMMAPRINT® 70-gene recurrence score, an OncotypeDX™ recurrence score, or a Prosigna® recurrence risk score. The calculated cancer signature may be a FoundationOne® result, and the mathematical model for fitting the FoundationOne® result may be based on a beta coefficient value of 10 or more in Supplementary Table 9. Alternatively, the mathematical model may be based on a beta coefficient value of 20 or more, a beta coefficient value of 40 or more, a beta coefficient value of 60 or more, or a beta coefficient value of 100 or more. In another embodiment, the mathematical model may be based on the top 5%, top 10%, top 25%, or top 50% of the beta coefficient values.

[0014] The calculated cancer signature may be associated with a mutation, substitution, or insertion or deletion (indel) in any of the following genes: (C17orf39), (MLL), (MLL2), ABL1, ACVR1B, AKT1, AKT2, AKT3, ALK, ALOX12B, AMER1, APC, AR, ARAF, ARFRP1, ARID1A, ASXL1, ATM, ATR, ATRX, AURKA, AURKB, AXIN1, AXL, BAP1, BARD1, BCL2, BCL2L1, BCL2L2, BCL6, BCOR, BCORL1, BRAF, BRCA1, BCL2L2, BCL6, BCOR, BCORL1, BRAF, BRCA1, BCL2L1, BCL2L2, BCL6, BCOR, BCORL1, BRAF, BRCA1, BCL2L1, BCL2L2 ... RCA2, BRD4, BRIP1, BTG1, BTG2, BTK, C11orf30, CALR, CARD11, CASP8, CBFB, CBL, CCND1, CCND2, CCND3, CCNE1, CD22, CD274, CD70, CD79A, CD79B, CDC73, C DH1, CDK12, CDK4, CDK6, CDK8, CDKN1A, CDKN1B, CDKN2A, CDKN2B, CDKN2C, CEBPA, CHEK1, CHEK2, CIC, CREBBP, CRKL, CSF1R, CSF3R, CTCF, CTNNA1, CTNNB1, CUL3, CUL4A, CXCR4, CYP17A1, DAXX, DDR1, DDR2, DIS3, DNMT3A, DOT1L, EED, EGFR, EP300, EPHA3, EPHB1, EPHB4, ERBB2, ERBB3, ERBB4, ERCC4, ERG, ERRFI1 , ESR1, EZH2, FAM46C, FANCA, FANCC, FANCG, FANCL, FAS, FBXW7, FGF10, FGF12, FGF14, FGF19, FGF23, FGF3, FGF4, FGF6, FGFR1, FGFR2, FGFR3, FGFR4, FH, F LCN, FLT1, FLT3, FOXL2, FUBP1, GABRA6, GATA3, GATA4, GATA6, GID4, GNA11, GNA13, GNAQ, GNAS, GRM3, GSK3B, H3F3A, HDAC1, HGF, HNF1A, HRAS, HSD3B1, ID 3, IDH1, IDH2, IGF1R, IKBKE, IKZF1, INPP4B, IRF2, IRF4, IRS2, JAK1, JAK2, JAK3, JUN, KDM5A, KDM5C, KDM6A, KDR, KEAP1, KEL, KIT, KLHL6, KMT2A, KMT2D,KRAS, LTK, LYN, MAF, MAP2K1, MAP2K2, MAP2K4, MAP3K1, MAP3K13, MAPK1, MCL1, MDM2, MDM4, MED12, MEF2B, MEN1, MERTK, MET, MITF, MKNK1, MLH1, MPL, MRE11A, MSH2, MSH3, MSH6, MST1R, MTAP, MTOR, MUTYH, MYC, MYCL, MYCN, MYD88, NBN, NF1, NF2, NFE2L2, NFKBIA, NKX2-1, NOTCH1, NOTCH2, NOTCH3, NPM1, NRAS, NT5C2, NTRK1, NTRK2, NTRK3, P2RY8, Page 4 of 36 RAL-0003-01, PALB2, PARK2, PARP1, PARP2, PARP3, PAX5, PBRM1, PDCD1, PDCD1LG2, PDGFRA, PDGFRB, PDK1, PIK3C2B, PIK3C2G, PIK3CA, PIK3CB, PIK3R1, PIM1, PMS2, POLD1, POLE, PPARG, PPP2R1A, PPP2R2A, PRDM1, PRKAR1A, PRKCI, PTCH1, PTEN, PTPN11, PTPRO, QKI, RAC1, RAD21, RAD51, RAD51B, RAD51C, RAD51D, RAD52, RAD54L, RAF1, RARA, RB1, RBM10, REL, RET, RICTOR, RNF43, ROS1, RPTOR, SDHA, SDHB, SDHC, SDHD, SETD2, SF3B1, SGK1, SMAD2, SMAD4, SMARCA4, SMARCB1, SMO, SNCAIP, SOCS1, SOX2, SOX9, SPEN, SPOP, SRC, STAG2, STAT3, STK11, SUFU, SYK, TBX3, TEK, TET2, TGFBR2, TIPARP, TNFAIP3, TNFRSF14, TP53, TSC1, TSC2, TYRO3, U2AF1, VEGFA, VHL, WHSC1, WHSC1L1, WT1, XPO1, XRCC2, ZNF217, or ZNF703.

[0015] The calculated cancer signatures were: ALK intron 18, 19; BCL2 3'UTR; BCR intron 8, 13, 14; BRAF intron 7-10; BRCA1 intron 2, 7, 8, 12, 16, 19, 20; BRCA2 intron 2; CD74 intron 6-8; EGFR intron 7, 15, 24-27; ETV4 intron 5, 6; ETV5 intron 6, 7; ETV6 intron 5, 6; EWSR1 intron 7-13; EZR intron 9-11; FGFR1 intron 1, 5, 17; FGFR2 intron 1, 17; FGFR3 intron 17; KIT It may be associated with rearrangements of intron 16; KMT2A (MLL) introns 6-11; MSH2 intron 5; MYB intron 14; MYC intron 1; NOTCH2 intron 26; NTRK1 intron 8-10; NTRK2 intron 12; NUTM1 intron 1; PDGFRA introns 7, 9, 11; RAF1 intron 4-8; RARA intron 2; RET intron 7-11; ROS1 intron 31-35; RSPO2 intron 1; SDC4 intron 2; SLC34A2 intron 4; TERC ncRNA; TERT promoter; or TMPRSS2 intron 1-3.

[0016] The computed cancer signatures include bladder urothelial carcinoma (BLCA), cervical squamous cell carcinoma and adenocarcinoma (CESC), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), clear cell renal carcinoma (KIRC), papillary renal cell renal carcinoma (KIRP), acute myeloid leukemia (LAML), brain low-grade glioma (LGG), hepatocellular carcinoma (LIHC), The signature may be a lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), pancreatic adenocarcinoma (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate adenocarcinoma (PRAD), rectal adenocarcinoma (READ), sarcoma (SARC), skin cutaneous melanoma (SKCM), gastric adenocarcinoma (STAD), testicular germ cell tumor (TGCT), thyroid carcinoma (THCA), thymoma (THYM), or uterine endometrial carcinoma (UCEC) signature. The mathematical model for fitting the calculated signature is based on a beta coefficient value of 10 or more in Supplementary Table 14. Alternatively, the mathematical model may be based on a beta coefficient value of 20 or more, a beta coefficient value of 40 or more, a beta coefficient value of 60 or more, or a beta coefficient value of 100 or more. In another embodiment, the mathematical model may be based on the top 5%, top 10%, top 25%, or top 50% of the beta coefficient values.

[0017] The present disclosure also provides a method for treating a cancer patient with chemotherapy, comprising the steps of determining whether the patient has a particular cancer subtype, by: (a) obtaining or obtaining a biological sample from the patient; (b) performing or having performed a gene-level copy number aberration (CNA) assay on the biological sample, in which copy number is measured at multiple locations on multiple chromosomes; (c) comparing the results of the CNA assay to a standard set to determine whether the patient has a particular cancer subtype; and (d) administering a suitable chemotherapy regimen to the cancer patient if the patient has a particular cancer subtype based on the determined cancer subtype. The chemotherapy regimen can be an ongoing therapeutic intervention. An ongoing therapeutic intervention includes discontinuing a particular treatment.

[0018] In another embodiment, the present disclosure provides a method for generating a computed cancer signature of a cancer phenotype, the method comprising: (a) receiving a plurality of gene expression signatures and subtype information of the cancer phenotype; (b) receiving a plurality of copy number aberration (CNA) datasets of the cancer phenotype; (c) analyzing the plurality of CNA datasets by an artificial intelligence algorithm to obtain a preliminary set of CNA segment level signatures of the cancer phenotype; (d) modifying the preliminary set of CNA segment level signatures using a gene expression training set to obtain a final set of CNA segment level signatures; and (e) using the final set of CNA segment level signatures to refine the computed cancer signature of the cancer phenotype. The cancer phenotype may be associated with somatic mutations, levels of mRNA expression, levels of protein expression, or immunohistochemistry (IHC) signatures.

[0019] Additionally, the disclosure provides a method for generating a computed cancer signature for a patient, the method comprising: (a) receiving copy number aberration (CNA) data for the patient; (b) receiving one or more CNA(s) signature(s) associated with a cancer phenotype, the CNA signature(s) being based on cancer expression analysis, cancer subtype information, and CNA gain / loss information; (c) processing the patient's CNA data with an algorithm that utilizes the one or more CNA(s) signature(s) associated with the cancer phenotype in a manner that characterizes a characteristic of the CNA data of the patient relative to the one or more CNA(s) signature(s); and (d) determining the computed cancer signature for the patient. The cancer phenotype may be associated with somatic mutations, levels of mRNA expression, levels of protein expression, or an immunohistochemistry (IHC) signature.

[0020] In the methods disclosed herein, the cancer phenotype may be associated with adrenal, bladder, bone, breast, cervical, colon, liver, lymphatic, ovarian, pancreatic, penile, prostate, rectal, salivary gland, skin, spleen, testicular, thymus, thyroid, trachea, or uterine cancer. In a preferred embodiment, the cancer phenotype is associated with breast cancer.

[0021] In another embodiment, the disclosure provides a method for treating a subject having cancer, comprising: (i) receiving copy number aberration (CNA) data of the patient; (ii) receiving one or more CNA(s) signature(s) associated with a cancer phenotype, the CNA signature(s) being based on cancer expression analysis, cancer subtype information, and CNA gain / loss information; (iii) processing the patient's CNA data by an algorithm that utilizes the one or more CNA(s) signature(s) associated with the cancer phenotype to characterize the characteristics of the CNA data of the patient characteristics relative to the one or more CNA(s) signature(s); (iv) adjusting the patient's calculated cancer signature based on the characterized characteristics; and (b) treating the patient based on a treatment plan based on the calculated cancer signature. The treatment can be an ongoing therapeutic intervention. The ongoing therapeutic intervention can include discontinuing a particular treatment.

[0022] The present disclosure also provides a device, comprising a processor configured to algorithmically process patient CNA data and one or more CNA(s) signature(s) associated with a cancer phenotype to generate a calculated cancer signature as described above. A system is also provided that includes the device of claim 36. Additionally, the present disclosure provides a device of claim 36, comprising software that includes an algorithm for comparing patient CNA data with one or more CNA(s) signature(s) associated with a cancer phenotype. [Brief description of the drawings]

[0023] 4. Brief description of the drawings [Figure 1]Identification of gene expression signature-specific CNA associations in breast cancer. Figure 1A. Schematic overview of the strategy used to identify CNAs associated with gene signatures. Gain / Loss indicates DNA copy number gain or loss; Pos / Neg indicates positive or negative association. Figures 1A-1D. Linear regression landscapes considering molecular subtypes were used to identify genes positively (dark grey above the axis) or negatively (dark grey below the axis) associated with gene signatures, and Fisher's exact test was used to compare the frequency of copy number gain (light grey above the axis) or loss (light grey below the axis) for retinoblastoma gene loss of heterozygosity (RB-LOH) (Figure 1B), basal signaling (Figure 1C), and estrogen signaling (Figure 1D) gene program signatures. Dotted lines indicate the significance threshold (q=0.01). Only q values ​​of genes significant in both analyses were plotted. Black arrowheads indicate known pathway drivers. In each figure, chromosome boundaries are indicated by black vertical lines. [Diagram 2] Patterns of DNA CNAs and gene expression signatures in breast cancer. Figure 2A. Heatmap showing DNA CNAs indicating gain and loss. Samples are co-ordered on the X-axis according to molecular subtype. Genes are co-ordered on the Y-axis according to chromosomal location. Figure 2B. Heatmap showing gene expression signatures. Samples are co-ordered on the X-axis according to molecular subtype. Gene signature scores are median centered and clustered by centroid linkage hierarchical clustering based on Pearson correlation. [Diagram 3] Patterns of association between DNA CNAs and amplicon signatures. Genes with positive correlation and increased frequency of copy number gain (q<0.01) in samples with high signature scores (upper quartile) are shown in light grey, and those with negative correlation and increased frequency of copy number loss (q<0.01) are shown in dark grey. Each amplicon signature is positively associated with its corresponding amplicon. [Figure 4]Elastic Net prediction model based on DNA CNAs of gene signatures in breast cancer. Figure 4A. Schematic overview of the strategy used to build an Elastic Net model for predicting gene expression signature levels. Figure 4B. Area under the curve (AUC) values ​​of 543 signatures displayed using box plots indicating median score (horizontal line), interquartile range (IQR, box boundaries), and 1.5 times the IQR (whiskers); horizontal line indicates AUC=0.75, which is considered "highly predictive". Three signatures are highlighted and their feature landscapes are also shown in Figure 4F-H. Figure 4C-4E. Receiver operating characteristic (ROC) curves and corresponding AUC values ​​of the TCGA test set and the Molecular Taxonomy of Breast Cancer International Consortium (METABRIC) validation set for predicting RB-LOH (Figure 4C), basal signaling (Figure 4D), and estrogen signaling (Figure 4E). Figure 4F-4H. Selected CNA segments and / or whole chromosome arms of elastic net and their coefficients for predictive models of RB-LOH (Figure 4F), basal signaling (Figure 4G), and estrogen signaling (Figure 4H). [Diagram 5] Identification of breast cancer subtype-adjusted gene signature-specific CNAs. Figure 5A. Schematic overview of the strategy used to identify CNAs associated with gene signatures considering molecular subtypes. Gain / Loss indicates DNA copy number gain or loss; Pos / Neg indicates positive or negative association. Figure 5B-D. Linear regression considering molecular subtypes was used to identify genes positively (above axis) or negatively (below axis) associated with gene signatures, and Fisher's exact test was used to compare the frequency of copy number gain (above axis) or loss (below axis) for RB-LOH (Figure 5B), basal signaling (Figure 5C), and estrogen signaling (Figure 5D) gene program signatures. The dotted line indicates the significance threshold (q=0.01). Only q values ​​of genes significant in both analyses were plotted. [Figure 6]Histograms of permutation test set AUC values. Test set AUC values ​​from 100 permutations per phenotype were plotted for each highly predictive gene expression signature, clinical receptor status, somatic mutations, and unique molecular subtypes. The vertical line indicates AUC=0.75, which is used as the threshold for defining "highly predictive." [Figure 7] Elastic Net prediction models based on multiple key expression signatures and CNAs of prognosis. Figure 7A. ROC curves and corresponding AUC values ​​for TCGA test set and METABRIC validation set of HER1-C2 signature. Figure 7B. Selected CNA segments and / or whole chromosome arms and their coefficients for the prediction model of HER1-C2 signature. Known drivers of EGFR pathway are highlighted by black arrows. Figure 7C-F. Kaplan-Meier curves of 10-year breast cancer-specific survival stratified by gene signature score (gene expression) and corresponding Elastic Net prediction models (DNA CNAs) for RB-LOH (Figure 7C), basal signaling (Figure 7D), estrogen signaling (Figure 7F), and HER1-C2 (Figure 7F) signatures. Event statistics were expressed as number of events / total patients in both high and low groups. [Figure 8]Elastic Net prediction models based on CNAs of three clinically used breast cancer assays. Figures 8A-I. ROC and Kaplan-Meier curves of 10-year breast cancer-specific survival for Oncotype DX® Recurrence Score (Figures 8A-C), Prosigna Risk of Recurrence Score (Figures 8D-F), and MAMMAPRINT® 70-gene Recurrence Score (Figures 8G-I). Kaplan-Meier curves were stratified by gene signature score (gene expression) and corresponding Elastic Net copy number prediction model (DNA CNA). Event statistics were expressed as number of events / total patients in both high and low groups. Figures 8J-L. Selected CNA segments and / or whole chromosome arms of Elastic Net and their coefficients for predictive models of Oncotype DX® Recurrence Score (Figure 8J), Prosigna® Recurrence Risk Score (k), and MAMMA PRINT® 70-gene Recurrence Score (Figure 8L). [Figure 9]Elastic net models predicting individual protein expression and mutation status in breast cancer. Figure 9A. Box plot showing median score (horizontal line), interquartile range (IQR, box boundaries), and 1.5 times IQR (whiskers) of AUC values ​​for predicting protein expression of 216 proteins and phosphoproteins from TCGA RPPA array. Five proteins of interest are highlighted by dots. Horizontal line indicates AUC=0.75. Figures 9B-9D. ROC curves and corresponding AUC values ​​of TCGA test set and METABRIC validation set for predicting clinical ER status (Figure 9B), clinical PR status (Figure 9C), and clinical HER2 status (Figure 9D). Figure 9E. Selected CNA segments and / or whole chromosome arms and their coefficients of the clinical ER status prediction model. Figure 9F. Dot plot showing AUC values ​​for predicting individual somatic mutations. Horizontal line indicates AUC=0.75. Figure 9G-9H. ROC curves and corresponding AUC values ​​of TCGA training set and TCGA test set for predicting TP53 mutations (Figure 9G) and mutation burden (Figure 9H). Figure 9I. Selected CNA segments and / or whole chromosome arms and their coefficients for the mutation burden prediction model. [Figure 10] Breast cancer subtype-specific CNA-based elastic net prediction models of gene signatures. Figure 10A. Box plots indicate AUC values ​​for predicting gene signatures in basal-like (n=185), luminal (n=853), and luminal A (n=556) samples; note that plots are stratified into immune signatures (n=78) and other signatures (n=465). Horizontal lines indicate AUC=0.75. Figure 10B. ROC curves and corresponding AUC values ​​for the TCGA test set and METABRIC validation set for predicting CD8 T cell expression signatures in basal-like samples. Figure 10C. Selected CNA segments and / or whole chromosome arms and other coefficients of the predictive model of CD8 T cell expression signatures in basal-like samples. [Figure 11]CNA-based elastic net prediction models of intrinsic and histological subtypes of breast cancer. Figures 11A-11E. ROC curves and corresponding AUC values ​​for predicting basal-like (Figure 11A), HER2 high (Figure 11B), luminal A (Figure 11C), and luminal B (Figure 11D) subtypes, and breast cancer histology IDC vs. ILC (Figure 11E). Figures 11F-11J. Selected CNA segments and / or whole chromosome arms and their coefficients for the prediction models of basal-like (Figure 11F), HER2 high (Figure 11G), luminal A (Figure 11H), and luminal B (Figure 11I) subtypes, and histology (Figure 11J). Positive weights favor ILC classification. [Figure 12] Selected CNA landscapes of DNA-based Elastic Net predictive models of clinical receptor status and corresponding protein expression measured by RPPA. Selected CNA segments and / or whole chromosome arms and their coefficients for predictive models of ER IHC status (Figure 12A), ER RPPA expression (Figure 12B), PR IHC status (Figure 12C), PR RPPA expression (Figure 12D), HER2 IHC status (Figure 12E), and HER2 RPPA expression (Figure 12F). Models predicting RPPA expression and IHC status of the same protein have similar landscapes. [Figure 13] Comparison of Elastic Net model performance using all genes and Foundation One® 313 gene set predictors. Box plots showing median score (horizontal line), interquartile range (IQR, box boundaries), and 1.5 times the IQR (whiskers) of AUC values ​​for predicting gene signatures and individual protein expression using all genes (left data points in each column) and the Foundation One® test 313 genes (right data points in each column) in breast cancer. AUC values ​​are highly correlated between the two categories. [Figure 14]Pan-cancer DNA CNA-based elastic net prediction models of gene signatures. Figure 14A. Line plots indicate the number of highly predictive signatures (i.e., AUC>0.75) (upper line) and highly predictive non-amplicon signatures (lower line) for each tumor type. Figure 14B. Box plots indicate the percentage of copy number aberrant genes for each tumor type. Figure 14C. Heatmaps show the predictability of each gene signature for each tumor type. Grey indicates predictive, black indicates non-predictive. Tumors and gene signatures are clustered by hierarchical clustering using Euclidean distance and complete linkage. Figure 14D and Figure 14E. Selected CNA segments and / or whole chromosome arms and other coefficients for the multi-tumor prediction models of claudin-low and immune signatures. [Figure 15] Elastic net prediction model based on DNA CNAs for lung cancer gene signature. Figure 15A. Box plots indicate AUC values ​​for predicting lung cancer gene signature using models built on lung cancer data (TCGA training and TCGA testing on the x-axis) and breast cancer data (breast cancer model prediction on the x-axis). Horizontal lines indicate AUC=0.75. Figure 15B. ROC curves and corresponding AUC values ​​for predicting TP53 status signature showing that both models built on lung cancer data and breast cancer are successful (AUC>0.75). Figure 15C. ROC curves and corresponding AUC values ​​for predicting lung histology, lung adenocarcinoma (LUAD) vs. lung squamous cell carcinoma (LUSC). Figures 15D-15E. Selected CNA segments and / or whole chromosome arms and other coefficients of predictive models built on lung cancer (Figure 15D) and breast cancer (Figure 15E) of the TP53 status signature show similar feature landscapes. Figure 15F. Selected CNA segments and / or whole chromosome arms and other coefficients of predictive models for classifying lung histology LUAD vs. LUSC. Positive weights favor LUSC classification. [Figure 16]DNA CNA-based elastic net prediction of continuous RB-LOH signature scores in breast cancer. Figure 16A. Scatter plot of predicted RB-LOH signature scores against observed signature scores for TCGA training set, TCGA test set, and METABRIC validation set. Grey lines are fitted regression lines. Pearson correlations are indicated. Figure 16B. Selected CNA segments and / or whole chromosome arms and their coefficients for the prediction model. [Figure 17] A DNA CNA-based elastic net model predicting key cancer phenotypes according to copy number values ​​determined by DNA sequencing of breast cancer. DNA sequencing data from exome sequencing was analyzed. Receiver operating curve (ROC) and area under ROC (AUC) values ​​for predicting key gene expression signatures (Figure 17A), clinical assays (Figure 17B), clinical receptor status (Figure 17C), and molecular subtypes (Figure 17D). DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0024] 5. Detailed Description of the Disclosure Those skilled in the art will recognize that there are a number of methods for obtaining copy number abnormality (CNA) data for use in the methods described herein. The source of CNA data can be conventional, including but not limited to, fluorescent in situ hybridization (FISH), comparative genomic hybridization (CGH), array comparative genomic hybridization (aCGH), or single nucleotide polymorphism (SNP) arrays. More recently, methods have been described for obtaining CNA data of samples using whole genome sequencing (WGS), whole exome sequencing (WES), or a combination of WGS and WES. For a review, see Hehir-Kwa, et al. (2018) "The clinical implementation of copy number detection in the age of next-generation sequencing," Expert Review of Molecular Diagnostics, vol. 18:10, 907-915. Several tools are available for obtaining CNA data from WGS or WES. Examples include ADTEx, ControlFREEC, VarScan2, and SynthEx.See, e.g., Silva et al. (2017) "SynthEx: a synthetic-normal-based DNA sequencing tool for copy number alteration detection and tumor heterogeneity profiling." Genome Biology 18:66; or Zare et al. (2017) "An evaluation of copy number variation detection tools for cancer using whole exome sequencing data." BMC Bioinformatics (2017) 18:286.Software programs (copy number variant (CNV) callers) for obtaining CNA data from next-generation sequencers and their release years are as follows: ADTex (2014), BIC-seq (2011), BreakDancer (2009), CANOES (2014), Canvas (2011), CLAMMS (2015), cn.MOPS (2012), CNVem (2013), CNVer (2010), CnvHiTSeq (2012). , CNVkit (2016), CNVnator (2011), CNVrd2 (2014), CNV-seq (2009), CODEX (2015), CONIFER (2012), CONTRA (2012), Cont rol-FREEC (2011), CoNVaDING (2015), Copy-Seq (2010), Cortex (2012), DECoN (2016), Delly (2012), Excavator (2013), ExomeCNV (2011), ExomeCopy (2011), ExomeDepth (2012), FermiKit (2015), GASV (2009), GATK (2010), GenomeSTRiPv2 (2 015), GROM-RD (2015), iCopyDAV (2018), JointSLM (2011), LUMPY (2014), Magnolya (2009), m-HMM (2013), mrCaNaVAR (2 2009), PEMer (2009), Pindel (2009), RDXplorer (2009), ReadDepth (2011), RSICNV (2017), Samblaster (2013), SegSeq (2009), SeqCNV (2017), SOAPsv (2011), Ulysses (2015), VariationHunter (2009), VarScan (2012), and XHMM (2012).See McCormick (August 8, 2019) "CNV Analysis Shifts Focus to NGS Sequences," https: / / www.biocompare.com / Editorial-Articles / 363086-CNV-Analysis-Shifts-Focus-to-NGS-Sequences / .

[0025] There are many examples of approved drugs that would benefit from the methods described herein. Drugs include small molecule kinase inhibitors, such as Imatinib (Gleevac®), an inhibitor of breakpoint cluster region-Abelson (BCR-ABL), originally approved for chronic myeloid leukemia (CML). An example of a monoclonal antibody kinase inhibitor is Trastuzumab (Herceptin®), an inhibitor of ERB-B2, approved for breast cancer, or Bevacizumab (Avastin®), an inhibitor of vascular endothelial growth factor (VEGF), approved for colon cancer. Other examples of drugs approved for companion diagnostics include drugs approved for BRCA1 / 2 mutations, KRAS mutations, and cKIT expression. Table 1 lists several approved drugs, including several kinase inhibitors. See Janne et al., 2009, Nature Reviews Drug Discovery 8:709-723; Levitzki and Klein, 2010, Molecular Aspects of Medicine 31:287-329; and Mellor et al., 2011, Toxicological Sciences 120(1):14-32; and the package insert for the particular drug.

[0026] [Table 1-1] [Table 1-2] [Table 1-3]

[0027] Abbreviations: For gene targets see Gene Cards (http: / / www.genecards.org / ). Indications: ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; BrCA, breast cancer; CML, chronic myeloid leukemia; CMML, chronic myelomonocytic leukemia; CRC, colorectal cancer; GIST, gastrointestinal stromal tumor; HCC, hepatocellular carcinoma; HNSCC, head and neck squamous cell carcinoma; MDS / MPD, myelodysplastic syndromes / myeloproliferative disorders; NSCLC, non-small cell lung cancer; OvCA, ovarian cancer; RCC, renal cell carcinoma; STS, soft tissue sarcoma; and TNBC, triple-negative breast cancer.

[0028] Many of these drugs are approved for use in companion diagnostics, for example, trastuzumab (Herceptin®) for breast cancers overexpressing ERB-B2, and cetuximab (Erbitux®) for patients with wild-type KRAS. Amado et al., 2008, J. Clinical Oncology 26(10): 1626-1634; Allegra et al., 2009, J. Clinical Oncology 27: 2091-2096. Another kinase inhibitor approved for diagnostic use is crizotinib (Xalkori®), which has been approved for fluorescent in situ hybridization (FISH) testing for ALK rearrangements (Vysis LSI ALK Dual Color, Break-Apart Rearrangement Probe; Abbott Molecular, Abbott Park, IL). Shah et al., 2011, The Lancet Oncology 12:1004-1012; Shaw et al., 2009, Journal of Clinical Oncology 27:4247-4253. Vemurafenib (Zelboraf®) is approved for use in patients with the BRAF V600E mutation (Cobas 4800 BRAF V600 Mutation Assay, Roche Molecular Diagnostics, Pleasanton, CA). Chapman et al., 2011, The New England Journal of Medicine 364, 2507-2516.Additional details can be found on the US FDA website for companion diagnostics (https: / / www.fda.gov / MedicalDevices / ProductsandMedicalProcedures / InVitroDiagnostics / ucm301431.htm). See the Biomarker column in Table 1 for additional companion diagnostics.

[0029] The methods disclosed herein can be used to aid in the diagnosis and treatment of a number of cancers. Once a particular cancer is diagnosed, there are a variety of targeted therapies that clinicians can use to treat the patient. Non-limiting examples for bladder cancer include erdafitinib (Barvasa™) or pembrolizumab (Keytruda™) in Table 1. Additional therapies include avelumab (Bavencio™), durvalumab (Imfinzi™), or nivolumab (Opdivo™). Non-limiting examples of BrCA include abemaciclib (Verzenio®), adotrastuzumab emtansine (Kadcyla®), alpelisib (Picray®), atezolizumab (Tecentriq®), everolimus (Afinitor®), lapatinib (Tykerb®), olaparib (LYNPARZA®), palbociclib (Ibrance®), pertuzumab (Perjeta®), ribociclib (Kisqali®) or trastuzumab (Herceptin®), or trastuzumab (Herceptin Hylecta™) in Table 1. Additional therapies include anastrozole (Arimidex®), exemestane (Aromasin®), fulvestrant (Faslodex®), letrozole (Femara®), neratinib (Nerlynx™), tamoxifen (Soltamox®), or toremifene (Fareston®). Non-limiting examples of CRC include bevacizumab (Avastin®), cetuximab (Erbitux®), panitumumab (Vectibix®), ramucirumab (Cyramza®), or regorafenib (Stivarga®) in Table 1. Additional therapies include ipilimumab (Yervoy®), nivolumab (Opdivo®), or dibuflibercept (Zaltrap®). Non-limiting examples of HCC include pembrolizumab (Keytruda®), ramucirumab (Cyramza®), regorafenib (Stivarga®), or sorafenib (Nexavar®) in Table 1.Additional treatments include cabozantinib (Cabometyx™), lenvatinib (Lenvima®), or nivolumab (Opdivo®). Non-limiting examples of renal cancer include axitinib (Inlyta®), bevacizumab (Avastin®), cabozantinib (Cabometyx®), everolimus (Afinitor®), pazopanib (Votrient®), pembrolizumab (Keytruda®), sorafenib (Nexavar®), sunitinib (Sutent®), temsirolimus (Torisel®) in Table 1. Additional treatments include avelumab (BAVENCIO®), ipilimumab (YERVOY®), lenvatinib mesylate (LENVIMA®), or nivolumab (OPDIVO®). Non-limiting examples of leukemia include dasatinib (Sprycel®), enasidenib (IDIHAIFA®), gilteritinib (XOSPATA®), imatinib (Gleevec®), ivosidenib (TIVSOVO®), midostaurin (RYDAPTO®), nilotinib (TASIGNA®), or venetoclax (VENCLEXTA®) in Table 1. Additional treatments include alemtuzumab (Campus®), blinatumomab (Blincyto®), bosutinib (Boslif®), duvelisib (Copictra®), gemtuzumab ozogamicin (Mylotarg®), glasdegib (Daurismo®), ibrutinib (Imbruvica®), idelalisib (Zydelig®), inotuzumab ozogamicin (Besponsa®), moxetumomab pasudotide (Pfizer®), and rituximab (Tregs®). These include rituximab-tdfk (Lumoxitit™), obinutuzumab (Gazyva™), ofatumumab (Arzera™), ponatinib (Iclusig™), rituximab (Rituxan™), rituximab and human hyaluronidase (Rituxan Hythera™), tagraxofuspumab (Erzonris™), tisagenlecleucel (Kymriah™), or tretinoin (Vesanoids™).Non-limiting examples of lung cancer, e.g., NSCLC, include those in Table 1, such as afatinib (GILORAF®), alectinib (Alecensa®), atezolizumab (Tecentriq®), bevacizumab (Avastin®), ceritinib (LDK378 / Zykadia®), crizotinib (Xalkori®), dabrafenib (Tafinlar®), and others. NAR®), dacomitinib (Vizinpro®), erlotinib (Tarceva®), gefitinib (Iressa®), osimertinib (Tagrisso®), pembrolizumab (Keytruda®), pemetrexed (Alimta®), ramucirumab (Cyramza®), trametinib (Mechanist®). Additional treatments include brigatinib (Alumbrig™), durvalumab (Imfinzi™), lorlatinib (Lobrena®), necitumumab (Portraza™), nivolumab (Opdivo®). Non-limiting examples of lymphoma include acalabrutinib (Calkens®), pembrolizumab (Keytruda®), and venetoclax (Veneclexta®) in Table 1.Additional treatments include axicabtagene ciloreucel (Yescarta™), belinstat (Bereodac™), bexarotene (Targretin™), bortezomib (Velcade™), brentuximab vedotin (Adcetris™), copanlisib (Alicopa™), denileukin diftitox (Ontak™), duvelisib (Copictra™), ibritumomab tiuxetan (Zevalin™), ibrutinib (Imbruvica™), idelalisib (Zydelig™), mogamulizumab (Mogamulizumab™), rituximab (Richardson ... These include lizumab-kpkc (Potelizio®), nivolumab (Opdivo®), obinutuzumab (Gazyva®), polatuzumab vedotin-piiq (Poraivy™), pralatrexate (Folotin®), rituximab (Rituxan®), rituximab and human hyaluronidase (Rituxan Hythera™), romidepsin (Istodax®), siltuximab (Sylvant®), tisagenlecleucel (Kymriah®), and vorinostat (Zolinza®). Non-limiting examples of melanoma include alitretinoin (Panretin®), binimetinib (Mektovi®), cobimetinib (Cotellic®), dabrafenib (TAFINAR®), encorafenib (Birafutovi™), pembrolizumab (Keytruda®), trametinib (Mechanist®), or vemurafenib (Zelboraf®) in Table 1. Additional treatments include avelumab (Bavencio®), cemiplimab-rwlc (Libtayo®), ipilimumab (Yervoy®), nivolumab (Opdivo®), sonidegib (Odomzo®), or vismodegib (Elivege®).Non-limiting examples of multiple myeloma (MM) include bortezomib (Velcade®), carfilzomib (Kyprolis®), daratumumab (Darazalex™), elotuzumab (Empliciti™), ixazomib (Ninlaro®), panobinostat (Farydak®), and selinexor (Expovio™). Non-limiting examples of prostate cancer include abiraterone acetate (Zytiga®) in Table 1. Additional treatments include apalutamide (Arleada™), cabazitaxel (Jevtana®), darolutamide (Nuvequo®), enzalutamide (Xtandi®), and radium-223 dichloride (Xofigo®). Additional drugs that may be used to treat cancer include denosumab (Zygeva®), dinutuximab (Unituxin™), iobenguane I131 (Azedra®), lanreotide acetate (Somatuline® depot), lutetium Lu177-dotatate (Lutatera®), niraparib (Zedula™), rucaparib camsylate (Rubraca™), ruxolitinib phosphate (Jakafi®), sirolimus (Rapamune®), or talazoparib (Tarzena®).

[0030] 5.1.Definition Although the following terms are believed to be well understood by those of skill in the art, the following definitions are submitted to aid in the explanation of the subject matter of the present disclosure.

[0031] As used herein, the "area under the curve" or "AUC" of a calculated signature is predictive if the AUC is greater than 0.60 or 0.65. The AUC of a calculated signature is "highly predictive" if it is 0.75 or more. The AUC of a calculated signature may be 0.80, 0.85, 0.90, 0.95, 0.97, or more.

[0032] As used herein, "clinical signs of cancer" means and includes any sign or indicator of the presence of cancer in a subject, which signs or indicators would be known to one of skill in the art (e.g., oncologist, clinical nurse). A clinical sign of cancer can be any symptom known to be associated with cancer. Some clinical signs of cancer include, for example, chronic pain, nausea, vomiting, abnormal taste, constipation, urinary symptoms (e.g., bladder spasms), respiratory symptoms, skin problems (e.g., itching, hair loss), or high fever, among others.

[0033] As used herein, "remission," as it is conventionally defined in the oncology field, means and includes a period during which symptoms of cancer are reduced or eliminated.

[0034] As used herein, "continuously measuring" the level of a biomarker in a sample refers to measuring the level of a biomarker in a sample more than once, for example, quarterly, bimonthly, monthly, biweekly, weekly, every third day, daily, or several times daily. Continuous monitoring of levels includes periodically measuring the level of a biomarker at regular intervals, as deemed necessary by one of skill in the art.

[0035] The term "standard level" as used herein refers to a baseline level of a biomarker determined in one or more normal subjects. For example, the baseline can be obtained from at least one subject, preferably from an average of subjects (e.g., n=2 to 100 or more), where the subject(s) have no prior history of cancer. In the present invention, the measurement of biomarker levels can be performed using multiple copy numbers as described.

[0036] As used herein, an "elevation" of a measured level of a biomarker relative to a standard level means that the amount or concentration of the biomarker in a sample is sufficiently great in a subject to be detected by the methods described herein relative to the standard. For example, an increase in a measured level relative to a standard level can be any statistically significant increase that is detectable. Such an increase can include, but is not limited to, about 1%, about 10%, about 20%, about 40%, about 80%, about 2-fold, about 4-fold, about 8-fold, about 20-fold, or about 100-fold increase, or more, relative to the standard. As used herein, the term "about" refers to a value plus or minus 10% of the value.

[0037] Non-limiting examples of signal transduction pathway modulators or chemotherapeutic agents known in the art include 5-fluorouracil; asparaginase; bevacizumab (Avastin®); bleomycin; campatecin; cetuximab (Erbitux®); crizotinib (Xalkori®); cyclophosphamide; cytarabine; dacarbazine; dactinomycin; dasatinib (Sprycel®); daunorubicin; DNA methyltransferase inhibitors (DNMTs), such as azacitidine (Vidaza®) and decitabine. ;doxorubicin;doxorubicin;epirubicin;erbstatin;erlotinib (Tarceva®);estramustine;etoposide;etoposide;gefitinib (Iressa®), gemcitabine, genistein, histone acetyltransferase inhibitors (HATs);histone deacetyltransferase inhibitors (HDACs), e.g. belinostat, entinostat (MS-275), panobinostat, PCI-24781, romidepsin (depsipeptide, FK-228), valproic acid, vorinostat (Zolinza (registered trademark), SAHA), or heat shock protein inhibitors, including HSP90 inhibitors, such as alvespimycin (IPI-493), AT13387, AUY922 (resorcinol isoxazolamide), CNF2024 (BIIB021), HSP990, MPC-3100, retaspimycin (IPI-504), SNX-2112, SNX-5422, STA-9090, tanespimycin (17-AAG; KOS-953), or XL888; herbimycin A; hexamethylmelamine; hedgehog pathway inhibitors, such as salidegial nephropathy inhibitors; (IPL-926), vismodegib (Elivege™); hydroxyurea, idarubicin, ifosfamide, imatinib (Gleevec®), irinotecan, lapatinib (Tykerb®), lavendustin A, leucovorin, levamisole, mercaptopurine, methotrexate, mitomycin, mitoxantrone, mTOR inhibitors such as everolimus (Afinitor®), sirolimus (Rapamune®), temsirolimus (Torisel®); nilotinib (Tasigna®);Nitrosoureas, such as carmustine and lomustine; paclitaxel; panitumumab (Vectibix®); pazopanib (Votrient®); pegaptanib (Macugen®); platinum compounds, such as carboplatin, cisplatin, oxaplatin; plicamycin, procarbizine; proteasome inhibitors, such as bortezomib (Velcade®); ranibizumab (Ruset®); Intis(R); sorafenib (Nexavar(R); sunitinib (Sutent(R); taxanes, e.g. docetaxel, paclitaxel, taxol; thioguanine; topotecan; trastuzumab (Herceptin(R)); tyrosine kinase inhibitors; tyrphostins; vandetanib (Caprelsa(R)); vemurafenib (Zelboraf(R)); vinblastine; vinca alkaloids; vincristine;or vinorelbine. In a preferred embodiment, the chemotherapy agent is bevacizumab (Avastin®), cetuximab (Erbitux®), crizotinib (Xalkori®), dasatinib (Sprycel®), erlotinib (Tarceva®), everolimus (Afinitor®), gefitinib (Iressa®), imatinib (Gleevec®), lapatinib (Tykerb®), nilotinib (Tanaka®), or rituximab (Tanaka®). Examples of chemotherapy drugs include: Signa®), panitumumab (Vectibix®), pazopanib (Votrient®), sirolimus (Rapamune®), sorafenib (Nexavar®), sunitinib (Sutent®), temsirolimus (Torisel®), trastuzumab (Herceptin®), vandetanib (Caprelsa®), or vemurafenib (Zelboraf®). Further examples of chemotherapy drugs can be found in Table 1 above, in standard publications and texts. See, for example, the National Comprehensive Cancer Network Guidelines (NCCN Guidelines™) or "Manual of Clinical Oncology," Dennis A. Casciato and Barry B. Lowitz, eds., 4th ed., July 15, 2000, Little, Brown and Company, USA;

[0038] Non-limiting examples of proteins whose expression signatures can be calculated using the methods disclosed herein are: 14-3-3_zeta, 4E-BP1, 4E-BP1_pS65, 4E-BP1_pT37_T46, 4E-BP1_pT70, 53BP1, ACC_pS79, ACC1, ADAR1, Akt_pS473, Akt_pT308, AMPK_alpha, Annexin_VII, AR, A-Raf_pS299, ASNS, Bap1-c-4, Bcl-2, Bim, B-Raf, B-Raf_pS445, BRD4, Caspase-8, CDK1_pY15, Chk2, Chk2_pT68, cIAP, COG3, Cyclin_B1, Cyclin_E1, DJ-1, DUSP4, Dvl3, eEF2K, EGFR, eIF4E, eIF4G, ER, ER-alpha, ER-alpha_pS118, ERK2, FASN, FoxM1, GAPDH, GATA3, HER2, HER2_pY1248, INPP4B, IRS1, JNK2, MSH2, MSH6, NF2, p16_INK4a, p53, p62-LCK ligand, p70S6K, p90RSK, P-cadherin, PCNA, PDK1, PDK1_pS241, PI3K-p110-alpha, PI3K-p85, PR, PREX1, Rab25, Rad50, Raptor, S6, Smac, Smad1, Src, VEGFR2, XRCC1, or YAP_pS127. For additional details and proteins see Figures 9A-9E and Supplementary Table 5.

[0039] As used herein, the terms "about" and / or "approximately" may be used for numerical values ​​and / or ranges. The term "about" is understood to mean a value close to the stated value. For example, "about 40 [units]" may mean within ±25% of 40 (e.g., 30 to 50), ±20%, ±15%, ±10%, ±9%, ±8%, ±7%, ±6%, ±5%, ±4%, ±3%, ±2%, ±1%, less than ±1%, or any other value or range of values ​​therein or therebelow. Alternatively, depending on the context, the term "about" may mean ±one-half a standard deviation, ±one standard deviation, or ±two standard deviations. Furthermore, the phrase "less than about [a value]" or "greater than about [a value]" should be understood in light of the definition of the term "about" provided herein. The terms "about" and "approximately" may be used interchangeably.

[0040] Numerical ranges are provided for certain quantities herein. It is to be understood that these ranges include all subranges therein. Thus, the range "50 to 80" includes all possible ranges therein (e.g., 51 to 79, 52 to 78, 53 to 77, 54 to 76, 55 to 75, 60 to 70, etc.). Furthermore, all values ​​within a given range may be endpoints of the range encompassed thereby (e.g., the range 50 to 80 includes ranges having endpoints such as 55 to 80, 50 to 75, etc.).

[0041] As used in this specification and claims, the verb "comprise" and its conjugations are used in their open-ended sense to mean that items following the word are included but not excluding items not specifically mentioned.

[0042] As used herein, the word "comprising" or variations such as "comprises" or "comprising" will be understood to imply the inclusion of the claimed element, integer, or step, or group of elements, integers, or steps, but not the exclusion of any other element, integer, or step, or group of elements, integers, or steps. The present disclosure may suitably "comprise", "consist of", or "consist essentially of" the steps, elements, and / or reagents recited in the claims.

[0043] 5.2. Computing Devices The computing device may be implemented in a programmable hardware device, such as a processor, digital signal processor, central processing unit, field programmable gate array, programmable logic array, programmable logic device, cloud processing system, or the like. The computing device may also be implemented in software for execution by various types of processors. The identified device may contain executable code, e.g., one or more physical or logical blocks of computer instructions, which may be organized as, e.g., objects, procedures, functions, or other constructs. Yet, the executable files of the identified device need not be physically located together, and may include different instructions stored in different locations, which, when logically linked together, constitute a computing device and achieve the purported purpose of the computing device. In another example, the computing device may be a server or other computer located in a hospital or outpatient environment and communicatively connected to other computing devices (e.g., POS machines or computers) for managing accounting, purchase processing, and other processes in the hospital or outpatient environment. In another example, the computing device may be a mobile computing device. For example, but not limited to, a smart phone, a mobile phone, a pager, a personal digital assistant (PDA), a mobile computer with a smart phone client, or the like. In another example, the computing device may be any type of wearable computer, such as a computer with a head mounted display (HMD), or a smart watch or some other wearable smart device. Some of the computer sensing may be part of the fabric of the clothing worn by the user. The computing device may also include any type of conventional computer, such as a laptop computer or a tablet computer.A typical mobile computing device is a wireless data access enabled device (e.g., an iPHONE® smartphone, a BLACKBERRY® smartphone, a NEXUS ONE™ smartphone, an iPAD® device, a smart watch, or the like) that can send and receive data in a wireless manner using protocols such as the Internet Protocol or IP and the Wireless Application Protocol or WAP. This allows a user to access information via wireless devices such as smart watches, smartphones, cell phones, pagers, two-way radios, communicators, and the like. Wireless data access is supported by many wireless networks, including but not limited to Bluetooth, short-range wireless communication, CDPD, CDMA, GSM, PDC, PHS, TDMA, FLEX, ReFLEX, iDEN, TETRA, DECT, DataTAC, Mobitex, EDGE, and other 2G, 3G, 4G, 5G, and LTE technologies, which run on many handheld device operating systems, such as PalmOS, EPOC, Windows CE, FLEXOS, OS / 9, JavaOS, iOS, and Android. Typically, such devices use a graphical display and can access the Internet (or other communication networks) by a so-called mini- or micro-browser, which is a web browser with a small file size that can accommodate the reduced memory constraints of wireless networks. In a representative embodiment, the mobile device is a mobile phone or a smartphone or a smartwatch, which operates by GPRS (General Packet Radio Service), the data technology of the GSM network, or by short-range wireless communication, such as Bluetooth.In addition to traditional voice communications, a given mobile device may communicate with another such device by way of many different types of message transfer technologies, including Bluetooth, near field communication, SMS (Short Message Service), Enhanced SMS (EMS), Multimedia Message (MMS), e-mail, WAP, paging, or other known or later developed wireless data formats. Although many of the examples provided herein are implemented in a smartphone, the examples may similarly be implemented in any suitable computing device, e.g., a computer.

[0044] The executable code of a computing device may be a single instruction or many instructions, and may even be distributed across several different code segments among different applications and across several memory devices. Similarly, operational data may be identified and illustrated herein within a computing device, and may be embodied in any suitable form and organized within any suitable type of data structure. Operational data may be collected as a single data set or may be distributed in different locations, including different storage devices, and may exist at least in part as electronic signals across a system or network.

[0045] The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to provide a thorough understanding of embodiments of the presently disclosed subject matter. However, one of ordinary skill in the art will recognize that the presently disclosed subject matter can be practiced without one or more of the specific details, or with other methods, components, materials, etc. In other instances, well-known structures, materials, or operations are not shown or described in detail to avoid obscuring aspects of the presently disclosed subject matter.

[0046] The term "memory" as used herein generally refers to a storage device of a computing device. Examples include, but are not limited to, read-only memory (ROM) and random access memory (RAM).

[0047] A device or system for performing one or more operations on a memory of a computing device may be software, hardware, firmware, or a combination thereof. A device or system is further intended to encompass or otherwise cover any software or computer program capable of performing various previously disclosed determinations, calculations, or the like for the purposes of the present disclosure. For example, the exemplary embodiments are intended to cover any software or computer program capable of enabling a processor to implement the processes of the present disclosure. The exemplary embodiments are also intended to cover any and all currently known related technologies or later developed non-transitory recording or storage media (e.g., CD-ROM, DVD-ROM, hard drive, RAM, ROM, floppy disk, magnetic tape cassette, etc.) that record or store such software or computer program. The exemplary embodiments are further intended to cover such software, computer program, system, and / or process provided by any other currently known related technologies or later developed media (e.g., transitory media, carrier waves, etc.) that can be used to implement the exemplary operations disclosed below.

[0048] According to an exemplary embodiment, the computer program of the present disclosure may be implemented in many exemplary ways, such as an application resident in the memory of the device or as a host application running on a server and communicating with the device application or browser via any number of standard protocols, such as TCP / IP, HTTP, XML, SOAP, REST, JSON, and other sufficient protocols. The computer program of the present disclosure may be written in an exemplary programming language, such as BASIC, COBOL, C, C++, Java, Pascal, or a scripting language, such as JavaScript, Python, Ruby, PHP, Perl, or other suitable programming language, that executes from the memory on the device or from a host server.

[0049] The terms "computing device" and "entity" as used herein should be interpreted broadly and should be understood to be interchangeable and may include any type of computing device, such as a server, a desktop computer, a laptop computer, a smart phone, a mobile phone, a pager, a personal digital assistant (e.g., a PDA with a GPRS NIC), a mobile computer with a smart phone client, or the like.

[0050] A user interface, as referred to herein, is generally a system by which a user interacts with a computing device. A user interface may include inputs that allow a user to operate a computing device and may include outputs that allow the system to present information and / or data, indicate the effect of the user's actions, etc. Examples of user interfaces for computing devices (e.g., mobile devices) include graphical user interfaces (GUIs) that allow a user to interact with a program in more ways than typing. GUIs typically offer display objects and visual indicators rather than text-based interfaces, typed command labels, or text navigation to represent information and actions available to a user. For example, an interface may be a display window or display objects that are selectable by a user of a mobile device for interaction. A user interface may include inputs that allow a user to operate a computing device and may include outputs that allow the computing device to present information and / or data, indicate the effect of the user's actions, etc. Examples of user interfaces for computing devices include graphical user interfaces (GUIs), which allow a user to interact with a program or application in more ways than typing. A GUI may typically offer display objects and visual indicators rather than a text-based interface, typed command labels, or text navigation to represent information and actions available to a user. For example, a user interface may be a display window or display objects that are selectable by a user of a computing device for interaction. Display objects may be displayed on a display screen of a computing device and may be selected and interacted with by a user using the user interface.In one example, the display of the computing device may be a touch screen that may display display icons. To select a display icon, a user may press down on an area of ​​the display screen where the display icon is displayed. In another example, a user may select a display icon or display object using any other suitable user interface of the computing device, such as a keypad. For example, a user may use a trackball or arrow keys to move a cursor to highlight and select a display object.

[0051] Display objects may be displayed on a display screen of the mobile device and may be selected and interacted with by a user using an interface. In one example, the display of the mobile device may be a touch screen, which may display display icons. To select a display icon, a user may press down on an area of ​​the display screen where the display icon is displayed. In another example, a user may select a display icon or display object using any other suitable interface of the mobile device, such as a keypad. For example, a user may use a trackball or times program instructions to cause a processor to implement aspects of the present disclosure.

[0052] A computer network, as referred to herein, can be any group of computing systems, devices, or equipment that are linked together. Examples include, but are not limited to, local area networks (LANs) and wide area networks (WANs). A network can be categorized based on its design model, topology, or architecture. In one example, a network can be characterized as having a hierarchical internetworking model, which divides the network into three layers: an access layer, a distribution layer, and a core layer. The access layer focuses on connecting client nodes, such as workstations, to the network. The distribution layer manages routing, filtering, and server quality of service (QoS) policies. The core layer can provide high-speed, highly redundant forwarding services to move packets between distribution layer devices in different regions of the network. The core layer typically includes multiple routers and switches.

[0053] The subject matter may be systems, methods, and / or computer program products. The computer program products may include computer-readable storage medium(s) having computer-readable program instructions thereon for causing a processor to implement aspects of the subject matter.

[0054] A computer readable storage medium may be a tangible device that can hold and store instructions for use by an instruction execution device. A computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer readable storage media includes: portable computer diskettes, hard disks, RAM, ROM, erasable programmable read only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disk read only memory (CD-ROM), digital versatile disk (DVD), memory stick, floppy disk, mechanically encoded devices such as punch cards or ridge structures in grooves having instructions recorded thereon, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as a transitory signal per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagated by wave guides or other transmission media (e.g., light pulses through fiber optic cable), or electrical signals transmitted over wires.

[0055] The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium into each computing / processing device or to an external computer or storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network or near field communication. The network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface of each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium within the respective computing / processing device.

[0056] The computer readable program instructions for carrying out the operations of the present subject matter may be assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state configuration data, or either source or object code, and may be written in any combination of one or more programming languages, including object oriented programming languages, such as Java, Smalltalk, C++, Javascript, or the like, and traditional procedural programming languages, such as the "C" programming language or similar programming languages. The computer readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer by any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (e.g., by the Internet using an Internet Service Provider). In some embodiments, to perform aspects of the present subject matter, electronic circuitry including, for example, a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may execute computer-readable program instructions by utilizing state information in the computer-readable program instructions to individualize the electronic circuitry.

[0057] Aspects of the present subject matter are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the subject matter. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0058] These computer-readable program instructions may be provided to a processor of a computer, a special purpose computer, or other programmable data processing apparatus to produce a machine. The instructions, when executed by the processor of the computer or other programmable data processing apparatus, create means for implementing the function / role specified in the block(s) of the flowcharts and / or block diagrams. These computer-readable program instructions may also be stored on a computer-readable storage medium that may direct a computer, programmable data processing apparatus, and / or other device to function in a particular manner. The computer-readable storage medium having instructions stored therein may then constitute an article of manufacture that includes instructions that implement aspects of the function / role specified in the block(s) of the flowcharts and / or block diagrams.

[0059] Also, the computer readable program instructions may be loaded into a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus, or other device to produce a computer-implemented process, such that the instructions executing on the computer, other programmable apparatus, or other device implement the function / role specified in the block(s) of the flowcharts and / or block diagrams.

[0060] The flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present subject matter. In this regard, each block of the flowcharts or block diagrams may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the figures. For example, depending on the functionality involved, two blocks shown in succession may in fact be executed substantially simultaneously, or the blocks may possibly be executed in the reverse order. It is also noted that each block of the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, may be implemented by a dedicated hardware-based system, which performs the specified functions or roles, or implements a combination of dedicated hardware and computer instructions.

[0061] It is further noted that the claims may be drafted to exclude any optional element, and therefore this statement is intended to serve as an antecedent basis for use of exclusive terminology such as "solely," "only," and the like with respect to the claim elements or for use of a "negative" limitation.

[0062] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure belongs. Preferred methods, devices, and materials are described, although any methods and materials similar or equivalent to those described herein may be used in the practice or testing of this disclosure. All references cited herein are incorporated by reference in their entirety.

[0063] The following examples further illustrate the present disclosure and are not intended to limit the scope. In particular, it is to be understood that the present disclosure is not limited to the specific embodiments described, as they can, of course, vary. It is also to be understood that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to be limiting, since the scope of the present disclosure will be limited only by the appended claims. EXAMPLES

[0064] 6. Example 6.1. Overview The ability to accurately characterize and predict tumor phenotypes is crucial for predicting patient prognosis. In addition, building predictive models of key phenotypes, such as signaling pathway activity, would be beneficial for guiding treatment selection. In this study, a comprehensive archive of tumor DNA information and gene expression signatures was utilized as a framework to characterize multiple aspects of tumor biology in detail. An integrated computational approach using genome-wide association analysis and elastic net prediction methods is presented to analyze the relationship between DNA copy number abnormalities and gene expression signatures. The approach was able to quantitatively predict, with high accuracy, the levels of many expression signatures within individual tumors of breast cancer based solely on DNA copy number features, including proliferation status and EGFR pathway activity. The elastic net model was also able to predict many other key phenotypes, including unique molecular subtypes, several protein expression features, including estrogen receptor status, and somatic mutation status, including TP53 and CDH1. This approach has been successfully applied to multiple other tumor types (pan-cancer), identifying several repeatably predictive signatures that encompass immune cell features of squamous / basaloid carcinomas. These elastic net DNA predictors could also be called from commonly used DNA-based gene panels to inform non-genetic tumor features that would therefore often guide treatment decisions. See Xia et al. (published December 11, 2019) Genetic Determinants of the Molecular Portraits of Epithelial Cancers Nature Communications 10:5666, the contents of which are incorporated in their entirety.

[0065] Characterization of multigene signature-specific DNA copy number abnormalities The possible association between DNA copy number aberrations (CNAs) and multiple gene expression signatures was first explored. The initial focus was on breast tumors, where multiple gene expression signatures are already in routine clinical use. 10-12 A panel of 543 published gene expression signatures measuring diverse phenotypes encompassing multiple signaling pathways, known prognostic / predictive models, tumor microenvironment features, and DNA amplicon and deletion signatures. 13 was applied to 1038 breast cancers using RNA-seq data from the TCGA Breast Cancer Project. 2 (Supplementary Table 1). DNA copy number data were used to identify possible associations linking DNA CNAs to each signature-based phenotype. Previously, Gatza et al. 8 developed an association analysis method for a much smaller cohort of patients to examine possible associations between CNAs and a limited panel of 52-gene signatures. This association analysis was modified to include another 491 signatures and to take into account intrinsic molecular subtype information (Figure 1A), because some gene signatures showed subtype-specific associations (Figures 2A-B). For each signature, two independent statistical methods were used to test for association. Linear regression with subtype as a covariate was used to identify positive or negative correlations between expression signature scores and gene-level DNA segment values. Following the work of Gatza et al., Fisher's exact test was used to compare the frequency of CNA gains or losses in samples with high signature scores (top quartile) and those with low signature scores (all others). Both tests were Benjamini-Hochberg corrected to control for false discovery rate. 14To further reduce possible false positive results, DNA CNA features associated with an expression signature were called only if the value was statistically significant in both analyses (q<0.01). In samples with high signature scores, likely DNA CNA drivers of the signature should have positive correlations and increased copy number gains, and likely repressors should have negative correlations and increased frequency of copy number losses. With this approach, the association landscape was analyzed for each signature, indicating that many expression signatures have no such associations.

[0066] The reproducibility of the association landscape was analyzed by comparing these results with those from Gatza et al. for the same signatures. 8 Here, all 52 signatures from Gatza et al. were included. In particular, the RB-LOH signature 15 We note that the current analysis used data from a much larger cohort of TCGA breast tumors (n = 1038 vs. n = 476); in addition, another systematic difference between the two studies is that Gatza et al. used gene expression microarrays, whereas mRNA-seq was used. More importantly, molecular subtypes were considered to identify universal associations regardless of subtype. Despite these methodological differences, there was a high concordance between the association landscapes of the RB-LOH signature and those published by Gatza et al. (Figure 1B); both landscapes highlighted the identification of known RB-E2F components, including DNA loss of RB1 and gain of E2F1 and E2F3, and amplification of multiple cell cycle drivers, including MYC and CCND2. 16 There were also associations present in previous results such as SOS1 that were absent in this analysis; this was due to intrinsic subtype correction, because the correlation between SOS1 and the RB-LOH signature was only present in basal-like tumors.

[0067] New and old possible associations were examined with all 543 gene signatures. Associations to previously determined DNA amplicon gene expression signatures were found, all encompassing regions of the corresponding amplicons (Figure 3), demonstrating that association analysis can identify known DNA-based drivers of expression signatures. Hoadley et al. 2014 17 Two universal expression signatures of key "gene programs" defined from PanCan (n=3500) tumor analysis of 12 tumor types, the so-called "basal signaling" signature and the "estrogen signaling" signature, both showed many informative associations. The basal signaling signature identified previously known associations including loss / deletion of genes involved in DNA repair such as RAD17, RAD50, PALB2, and BRCA1 (Figure 1C). The estrogen signaling signature identified many unique luminal tumor DNA copy number alterations, including 16p gains and 16q losses. 2 (FIG. 1D). Collectively, these results demonstrate that the strategies disclosed herein can objectively discover associations linking CNAs to specific gene signatures, many of which were previously known.

[0068] CNA-based gene signature prediction by elastic net model Given the strength of these associations, the feasibility of constructing a gene expression signature-level computational predictor based solely on DNA CNA features was assessed. Based on the fact that associations could be found between CNA and gene signatures, it was possible that at least some of the expression signatures would be predictable using DNA information alone. To successfully construct a predictive model, a statistical modeling approach called Elastic Net was used. This is a regularized regression model that can handle a large number of possible collinear variables, from which the most plausible features can be selected to construct the final model. 9Instead of using gene-level CNA scores as predictors, 536 segment-level CNA scores were calculated using defined chromosomal regions shown to be important in cancer. 18-22 (Supplementary Table 2). These DNA segments encompassed significant pan-cancer somatic CNA and breast cancer subtype-specific CNA regions. The 1038-sample TCGA breast cancer dataset was divided into a training set (70%) and a test set (30%). Models were built exclusively from the TCGA training set and validated with both the TCGA test set and a large independent breast tumor dataset (Molecular Taxonomy of Breast Cancer International Consortium (METABRIC), n=1,689). 3 A model was trained to classify samples into those with high signature scores (top third) versus low signature scores (bottom two thirds). Area under the ROC curve (AUC) values ​​were used to evaluate model performance (Figure 4A).

[0069] The AUC distribution of all gene signatures demonstrated high predictability for some, but not all, of the signatures (Figure 4B). Not surprisingly, the AUC values ​​were highest in the TCGA training set. However, the AUC values ​​were high and very similar between the TCGA test set and the METABRIC validation set for many signatures, indicating that successful models were developed for multiple expression features. Notably, 142 of the 543 signatures had AUC values ​​above 0.75 in both validation sets and were determined to be "highly predictive." Of these 142 signatures, only 33 were DNA-based amplicon signatures that essentially measure specific CNA events and were therefore expected to yield high AUC values. For example, signature 16q23 amplicon 23had the highest AUC value in the METABRIC validation set (AUC = 0.96). Notably, the three signatures highlighted in the association landscape, the clogged RB-LOH signature, the basal signaling signature, and the estrogen signaling signature, were all highly predictive (AUC > 0.85) as shown by the corresponding receiver operating characteristic (ROC) curves (Figure 4C-E). The most predictive signatures included multiple proliferation signatures and a few oncogenic pathways, while the least predictive signatures were mostly representative of immune infiltrates and other features of the tumor microenvironment. In particular, the HER1-C2 signature, previously developed by Hoadley et al. (2007), which indicates EGFR pathway activity, 24 had an AUC value of ∼0.90 in both validation sets (Figure 5A). Furthermore, three study-based implementations of commercially available signatures commonly used in breast cancer clinics, including the OncotypeDX® Recurrence Score, 25 , MAMMAPRINT® 70-Gene Recurrence Score 26 , and Prosigna® Recurrence Risk Score 11 were among the highly predictive signatures, with corresponding TCGA test set AUC values ​​of 0.89, 0.88, and 0.91 (Figure 5B-D). Permutation tests showed that a test set AUC of 0.75 indicated significant predictive power (Figure 6), and as might be expected, with the permuted data, the highest AUC achieved was 0.63, with the majority being close to 0.5. Of these 142 signatures, only 33 were DNA-based amplicon signatures that essentially measure specific CNA events, and thus were expected to yield high AUC values. For example, signature 16q23 amplicon 23 had the highest AUC value in the METABRIC validation set (AUC=0.96).

[0070] To better understand these predictive models, the CNA regions selected by the Elastic Net model (Supplementary Table 3) were examined. Model feature landscapes of the three signatures are shown for direct comparison against the association landscapes. Notably, for the RB-LOH signature and basal signaling signature, which had many associations with CNAs, there was a significant amount of overlap between the association landscape and the Elastic Net model feature landscape (Figure 4F-G). For example, the RB-E2F component and the cell cycle component were significantly associated with the RB-LOH signature and were selected by Elastic Net for prediction of the RB-LOH signature score. In contrast, the estrogen signaling signature had a simple association landscape, but the Elastic Net model selected many additional features beyond those of the regions based on the association analysis (Figure 4H). This suggests that Elastic Net provides additional information about the relationships between CNAs and gene signatures by combining whole genome data rather than association analysis evaluating genes one by one. In addition, in the highly predictive HER1-C2 signature, many EGFR pathway-related genes were selected by the Elastic Net model, including EGFR itself, KRAS, SOS1, and NRAS (Figure 7B). Together, these results demonstrate the ability to predict many gene expression signatures using only DNA CNAs with high accuracy and a biologically promising and informative feature set.

[0071] CNA-based prediction of unique molecular subtypes The elastic net DNA feature modeling strategy was then applied to predict other complex tumor phenotypes, including prediction of individual sample molecular subtypes of breast cancer. 11All intrinsic subtype prediction models demonstrated high AUC values ​​(i.e., >0.75), indicating that these RNA-based phenotypes could be well explained by DNA-based information (Figures S11A-S11D, Supplementary Table 4). Consistent with previous findings that basal-like breast cancer represents a unique disease entity, the basal-like subtype had the highest test set AUC value (>0.9). 17 Regions frequently abnormal in basal-like samples, such as 1p gain and 5q loss, were selected by the predictive elastic net model. 20,21 (Figure 11F). The HER2-high subtype also had a high AUC value (>0.82) and, not surprisingly, regions selected by the model included the ERBB2 region, which is a strong driver of this subtype (Figure 11G). Luminal A and Luminal B subtypes were more difficult to predict but were still highly predictable (AUC=0.82 for Luminal A and 0.76 for Luminal B in the METABRIC validation set). A notable difference between these two subtypes was the proliferation rate, where Luminal B tumors generally have a higher proliferation rate than Luminal A tumors; as might be expected, proliferation-related regions including 8q(MYC) amplification and RB1 deletion were present only in the Luminal B prediction model (Figures 11H-I). Since histological subtype also influences clinical treatment decisions, we evaluated how the DNA CNA-based elastic net model predicts the two major breast cancer histologies, infiltrating ductal carcinoma (IDC) and invasive lobular carcinoma (ILC). The ROC curves showed high AUC values ​​in both the TCGA training set (0.87) and test set (0.8) (Figure ​(Figure11E).1E). Loss of CDH1, located on chromosome 16q22.1, a hallmark of ILC, was reflected in the feature landscape of the model (Figure ​(Figure11J).11).

[0072] CNA-based prediction of individual protein expression In addition, the Elastic Net DNA-based modeling strategy was applied to build predictive models for individual proteins using reverse-phase protein array (RPPA) data measuring 216 proteins and phosphoproteins from TCGA breast cancer samples. 2 Many studies have addressed the relationship between protein levels and mRNA abundance, concluding that mRNA transcript levels predict protein levels approximately 50% of the time. 27 A few studies have also examined the effect of DNA copy number on protein expression and found several proteins with significant correlations, typically those that are targets of amplification or deletion. 28,29 However, these studies have assessed the correlation for individual genes. Here, an elastic net model is constructed to take into account the whole genome to predict protein expression. Using the aforementioned definition of a "high accuracy" AUC equal to or greater than 0.75, the model was able to accurately predict 16 of the 216 protein expression levels present in the RPPA array (test set AUC>0.75) (Figures 12A-F, Supplementary Table 5). Clinically, in breast cancer, the protein expression of estrogen receptor (ER), progesterone receptor (PR), and human epidermal growth factor receptor (HER2) directs breast cancer treatment and is routinely assessed by immunohistochemistry (IHC) staining. The expression of these three proteins assessed by RPPA was all highly predictable (AUC>0.75) by the model disclosed herein (Figures 9A-D, Supplementary Table 5). Among the other 13 highly predictable proteins, many of which are related to the cell cycle, were CCNB1, CCNE1, and FOXM1. Another interesting predictable protein was ASNS (test set AUC = 0.82); recently, ASNS has been shown to play an important role in breast cancer metastasis, where its high expression was linked to the increased metastatic potential of lung metastases, representing a possible therapeutic target. 30,31 .

[0073] In breast cancer, the most definitive therapeutic biomarkers are ER, PR, and HER2, scored by immunohistochemistry. For HER2 prediction, HER2 and 17q were selected with the highest coefficients by both the model of HER2 RPPA protein expression and the model guided by HER2 clinical IHC status (Figures 12E-F). In contrast, ESR1 copy number gain / loss is rare, so ER protein expression cannot be explained by ESR1 copy number changes. 32 Moreover, the elastic net model was able to accurately predict ER RPPA protein expression (AUC 0.82) and ER clinical IHC status (AUC 0.89 on the METABRIC validation set) when using only DNA copy number information. The feature landscape of these two models was complex, with many positive and negative predictors. Notably, luminal features 16p gain and 16q loss were included in the model, consistent with the fact that ER positivity is widespread in luminal tumors (Figure 9E). PR positivity was highly concordant with ER positivity, consistent with many regions predicting ER expression also predicting PR expression (Figures 12A-D).

[0074] Finally, an increasingly common clinical assay today for cancer patients is the gene panel assay, where typically hundreds of genes are DNA sequenced using massively parallel sequencing, thus providing somatic mutation status and DNA copy number values ​​for each gene. 33 One of the most widely used gene panels is FoundationOne®, which contained 313 genes at the time of writing this draft. Using only DNA copy number information of these 313 genes, the elastic net prediction model of all gene signatures and protein expression achieved essentially identical results and AUC values ​​(Figure 13); therefore, these complex expression and protein phenotypes can be accurately predicted when using only a small subset of the human genome.

[0075] CNA-based prediction of somatic mutations Next, the ability to predict individual somatic mutations was examined. Mutation data from TCGA breast tumors with highly confident mutation calls 34 was utilized, and the analysis was limited to the list of significantly mutated genes identified by previous work and to highly mutated genes (frequency >5%), excluding HLA and IGH genes. Only a few mutations, namely TP53, CDH1, and MAP3K1, passed the test set AUC threshold of 0.75 (Figure 9F-F); GATA3 and PIK3CA also had relatively high AUC values, slightly below 0.75. Consistent with their known tumor suppressor phenotype, both TP53 and CDH1 models selected CNA segments encompassing these two genes as negative predictors. Luminal subtype-specific mutation models, namely GATA3 and MAP3K1, selected luminal copy number changes encompassing 16p gain. Interestingly, tumor mutation burden, defined here as the total number of mutations per sample, has been shown to be associated with immunotherapy response. 35,36 was highly predictable from DNA CNAs (Figure 9F).

[0076] Subtype-specific prediction of gene signatures To examine whether molecular subtype influences the predictive potential of gene signatures, the same elastic net analysis described above was performed but applied only to basal-like subtype tumors, luminal A subtype tumors, or luminal tumors (HER2 high expression, luminal A and luminal B combined). Results showed that the predictive accuracy differed in the different subtypes (Supplementary Table 7), and in some cases, which signatures were predictive also varied. One striking phenomenon was that some immune signatures with low AUC values ​​using the model built with all breast cancer samples had higher AUC values ​​when using only basal-like tumors (Figure 10A). For example, the CD8 T cell signature was significantly lower when using all samples vs. basal-like samples. 37had AUC values ​​of 0.71 and 0.84 (Figure 10B). The segments selected to predict this signature included genes encoding the CD8 T cell chemokines CXCL9, CXCL10, and CXCL11, and SEC31A, a gene involved in chemokine secretion, both of which affect T cell trafficking. 38 Interestingly, EGFR was selected as a negative predictor, providing evidence that tumor-intrinsic mechanisms shape the tumor immune microenvironment. 39 (Figure 10C). This finding demonstrates the heterogeneity behind the different subtypes and provides insight into prioritizing basal-like tumors for immunotherapy.

[0077] Prediction of gene signatures in lung cancer To assess the generalizability of the elastic net modeling strategy, predictive models were evaluated using TCGA lung cancer data, including both lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC). 40,41Again, gene expression signatures were used. First, a predictive model based on DNA from breast cancer was applied to lung cancer. Results identified 37 signatures that passed the AUC threshold of 0.75 in the lung training set, lung test set, and breast cancer test set (Figure 15A, Supplementary Table 8). Not surprisingly, most of these signatures were amplicon signatures; however, two were not, and were related to TP53 mutation status and PTEN / PI3K pathway activity. The CNA segments and / or whole chromosome arms selected by models built in breast or lung cancer were similar for the TP53 status signatures (Figure 15D-E), indicating that the elastic net approach can consistently select the most plausible features. Finally, a larger number of signatures were found to be predictive when trained on lung cancer data, suggesting that some models may be tumor type-dependent and others may be tumor type-independent (i.e., TP53). The question of whether the elastic net model based on DNA CNAs could successfully classify the two lung cancer histologies was evaluated. The results showed that it was possible to classify LUAD vs. LUSC with very high accuracy (AUC=0.98 and 0.97 in the training and test sets), consistent with previous findings that LUAD and LUSC have unique somatic drivers. 42 (Figure 15C, Figure 15F).

[0078] Pan-cancer prediction of gene signatures Finally, the elastic net modeling approach was applied to 23 other tumor types from TCGA, which had multi-platform data and at least 100 samples. 4504 median expression-based gene signatures for each tumor type were examined for the predictive potential of each signature for each tumor type. Results showed that successful models with high accuracy (AUC>0.75) were constructed for multiple tumor types other than breast and lung cancer. Not surprisingly, there were more gene signatures that were highly predictive in tumor types with more CNA events (Figure 14A, Figure 14B). Hierarchical clustering of tumor types based on gene signature predictive potential revealed informative tumor subgroupings consistent with previous findings of similarities between tumor types, including the pan-squamous group (Figure 14C). 4 Basal-like breast cancer clustered with squamous cell carcinomas, including LUSC and head and neck squamous cell carcinoma (HNSC). Many immune-related signatures were uniquely predictive in these tumor types, including the previously described CD8 T cell signature and the PD1 and CTLA4 signaling pathways. Luminal breast cancer, on the other hand, clustered with LUAD and bladder urothelial carcinoma (BLCA), where multiple signatures measuring proliferation rate were highly predictive, including the RB-LOH signature. Finally, amplicon signatures were universally predictive in tumor types with a high percentage of copy number aberrant genes.

[0079] Claudin-low subtypes and a claudin-low signature representing an epithelial-mesenchymal transition (EMT)-like state 43Noting that the CNA regions that were universally important in predicting this signature were constructed as a model from the combined data from these tumor types, the CNA regions that were universally important in predicting this signature were highly predictive in the 11 tumor types. The resulting model had a training set AUC of 0.8 and a test set AUC of 0.74, indicating that the multi-tumor model was able to predict the signature in the 11 tumor types. In addition, the CNA regions selected by this model highlighted many RAS / MAPK pathway components, including the lesser known gene ERAS (Figure 14D), consistent with the finding that forced expression of the CNA regions induced EMT in human mammary cells. 44 Collectively, these results demonstrate that the elastic net approach can robustly construct predictive models of key gene signatures in multiple tumor types.

[0080] 6.2. Discussion The ability to predict key tumor phenotypes or complex expression phenotypes, such as mutation status or biomarker levels, is critical to understanding the biological complexity of solid epithelial cancers. In recent years, gene expression testing is commonplace in breast cancer, with protein expression analysis of ER, PR, and HER2 being required. In lung cancer, gene panel testing has been incorporated into the standard of care, and expression analysis (both mRNA and protein) has received increasing attention, primarily due to immunotherapy. Many solid epithelial cancers, particularly breast and lung, are thought to be driven, at least in part, by DNA copy number, since multiple copy number events occur, many with known genetic drivers. 45,46When investigating the proposed copy number driven tumor types, it was reasoned that many key tumor phenotypes may be predictable when using the diversity of DNA copy number alterations. To address this hypothesis, a large archive of manually curated gene expression signatures taken from multiple publications was used to study tumor phenotypes and estimate their predictability. The relationship between DNA copy number abnormalities and each gene expression signature was analyzed by two means; the first was a genome-wide association method, and the second was to build an elastic net predictive model to evaluate their accuracy. The association study allowed us to find DNA features that were positively or negatively correlated with genes to the expression signatures by evaluating them one by one. These two methods jointly generated a large picture of the links between CNAs and gene signatures. Known associations between CNAs and gene signatures were consistently found, including gene signatures of DNA amplification and loss, and for more complex phenotypes including signaling pathway activity (i.e., TP53 and EGFR), as well as gene signatures of cell proliferation status; in fact, the method was able to predict many of these signatures with very high accuracy (AUC>0.9) on true test sets even using different gene expression and DNA copy number technologies (i.e., METABRIC). Taken together, the association landscapes and Elastic Net feature landscapes of each gene signature provide CNA regions for further investigation of possible genetic drivers. In addition, further application of the Elastic Net modeling strategy to a variety of other molecular phenotypes including intrinsic molecular subtypes, protein expression levels, and somatic mutation status (including tumor mutation burden) revealed the ability to accurately predict many key phenotypes of breast cancer. With the increasing use of DNA exome and gene panels in the cancer clinic, these models may be clinically useful and, at a minimum, may provide an orthogonal approach to calling key features such as ER, PR, and HER2 status in breast cancer, and perhaps even be used to eventually call key phenotypes themselves.

[0081] In the analysis presented here, the expression signatures were divided into the highest tertile versus the two lowest tertiles; however, elastic net models in which the expression signatures were treated as continuous variables were also analyzed, and these too were successful for models that showed high AUC when tested as binary variables, although still with acceptable but lower accuracy (Figure 12A-F). Therefore, it may even be feasible to predict quantitative traits in addition to simple high versus low as has been done for many of the predictors.

[0082] With the goal of improving precision medicine, many commercial gene panel tests have been developed. Using DNA CNA information of only 313 genes that can be derived by the FoundationOne® genomic test, gene expression signature prediction and protein expression prediction accuracy remained the same when compared to using the whole exome of CNA values. The 313 genes were selected as being highly cancer relevant and have been reported to be important in tumorigenesis. This result suggests that a small portion of the genome accounts for a large portion of the predictive power of cancer phenotypes seen in several solid epithelial cancer types. It also sheds light on the application of the elastic net model in the clinic. For example, various proliferation signatures, including the RB-LOH signature evaluated here, could serve as possible biomarkers for CDK4 / 6 inhibitors targeting the RB / E2F pathway. 47 The elastic net model of the RB-LOH signature can be used to stratify patients into those with high proliferation rates, which typically identifies those with RB loss, for whom CDK4 / 6 inhibitors would not be recommended. Further validation is required to confirm this particular hypothetical application. However, if validated, a whole new set of prognostic and predictive biomarkers could be read out from existing DNA-based gene panels, thus providing more guidance for precision medicine at no additional cost.

[0083] Finally, the generalizability of this approach was demonstrated by a pan-cancer analysis of 23 different tumor types. Consistent with the working hypothesis, various gene signatures other than amplicon signatures were predictive in tumor types with many copy number alterations. Tumor types that were shown to share similar features had similar patterns of signature predictability. More importantly, shared key features such as immune features in squamous tumors and proliferation rates in adenocarcinomas were often highly predictive.

[0084] Collectively, these results demonstrate the ability to construct CNA-based predictors of multiple key cancer phenotypes for breast and non-small cell lung cancer patients. Although most studies have focused on finding the genetic drivers of tumorigenesis, the results hold important implications that critical complex tumor phenotypes may be predicted using DNA information, which could potentially be used in the clinic.

[0085] 6.3. Method Gene expression data. Illumina HiSeq 2000 RNA sequencing data for human breast and lung cancer (both lung adenocarcinoma and lung squamous cell carcinoma) were obtained from the Broad Institute TCGA GDAC Firehose. 4 Illumina HT-29 v3 expression data from the METABRIC project (n=1,992 samples) were obtained from the European Genome-Phenome Archive of the European Bioinformatics Institute. 3 For TCGA breast and lung cancer gene expression data, gene-level RNA-Seq reads were upper quartile normalized and log2 transformed, filtered to genes expressed in >70% of samples, median centered, and normalized for each sample within each dataset. For METABRIC microarray gene expression data, acquired data were filtered to genes expressed in >70% of samples, median centered for each gene, and normalized for each sample. For both TCGA and METABRIC breast cancer data, PAM50 subtyping was applied as previously described.2,3,11 Gene expression data for all other tumor types was downloaded from the GDC PanCanAtlas public site. For each tumor type, gene expression data were filtered to genes expressed in >70% of samples, median centered, and normalized per sample within each tumor type.

[0086] DNA copy number data. GISTIC2 gene-level copy number data for human breast and lung cancers were obtained from the Broad Institute TCGA GDAC Firehose without further processing. For the METABRIC project, copy number segmentation data using the circular binary segmentation (CBS) algorithm were obtained from the European Genome-Phenome Archive. 3 Using the Ensembl 54 (hg18) genome build, gene-level copy number scores were calculated using GISTIC2 48 The copy number data were derived by the extreme value method used in: genes that fell completely within a copy number segment identified by CBS were assigned the corresponding segment value. Genes that overlapped with multiple segments were assigned the most significant amplification or smallest deletion value among the overlapping segments. Genes with no overlapping segments were excluded from further analysis. GISTIC2 gene-level copy number data for all other tumor types were downloaded from the GDC PanCanAtlas public site without further processing.

[0087] Protein Expression Data. Normalized protein expression data for human breast cancer were obtained from the Broad Institute TCGA GDAC Firehose without further processing.

[0088] Mutation data. Mutation annotation format (MAF) data from the 2015 TCGA lobular breast cancer dataset was used. 34The MAF files were first filtered to include only the following variant classifications: Frame_Shift_Del, Frame_Shift_Ins, In_Frame_Del, In_Frame_Ins, Missense_Mutation, Nonsense_Mutation, Nonstop_Mutation, RNA,Splice_Site, and Translation_Start_Site. A binary gene by sample matrix with 1 indicating any mutation and 0 indicating no mutation was then constructed based on the filtered MAF (Supplementary Table 15). The mutation load of each sample was then determined based on the total number of mutated genes in that sample.

[0089] Gene expression signatures. A panel of 543 previously published gene expression signatures was used to further characterize the cancer phenotype. These 543 signatures were derived from multiple publications or GSEA. 49 , partially summarized by Tanioka et al. 13A complete list of genes and their references for each signature is shown in Supplementary Table 1. Signature scores were calculated in a manner consistent with their origin. For 504 signatures with homogeneous expression among genes, the median expression value was used as the signature score. The rest of the signatures were based on correlation to predefined gene centroids or on published algorithms. For correlation-based signatures, all predefined training sets are available for download from the GitHub repository (https: / / github.com / xyouli / DNA-based-predictors-of-non-genetic-cancer-phenotypes). For each such signature, DWD53 was used to first merge the gene expression matrix with the corresponding training set, and then Pearson / Spearman correlation / Euclidean distance was calculated for each sample of the merged data. For some algorithm-based signatures, corresponding R code was provided to calculate each signature. See the computer program listing appendix. All 543 signatures were applied to the TCGA breast and lung cancer data and the METABRIC data (see Supplementary Tables 12-14). 504 median expression-based signatures were applied to the pan-cancer data.

[0090] Identification of gene signature-specific CNAs. Two independent statistical tests were used to identify associations between CNAs and gene expression signatures. 8, the TCGA breast cancer cohort with matched gene expression and copy number data was used, excluding all normal-like samples (n=1038). For each signature, each gene was tested for significant association with molecular subtype, which was taken into account as a confounding variable: signature ~ CNA + (1|basal) + (1|HER2) + (1|LumA) + (1|LumB). Positive / negative correlation was determined by the coefficient and p-value of CNA in the model. Fisher's exact test was used to compare either the frequency of CNA gain or loss in samples with high signature scores (top quartile) and those with low signature scores (bottom three quartiles). In each analysis, Benjamini-Hochberg multiple testing correction was used to adjust the p-value of each signature for all genes. The significance threshold was set at 0.01 to identify genes that were significant in both analyses.

[0091] Constructing an Elastic Net Prediction Model. The Elastic Net modeling approach, a regularized regression method that linearly combines the L1 and L2 penalties of ridge regression and the least absolute shrinkage and selection operator (Lasso), was used to construct DNA CNA-based predictors of cancer phenotypes. 9Alternatively, other methods of machine learning can be used to build the model. Examples of such methods include: Lasso: Tibshirani, Robert, 1996, "Regression shrinkage and selection via the lasso," Journal of the Royal Statistical Society: Series B (Methodology), Vol. 58.1: pp. 267-288; Ridge: Hoerl, Arthur E. and Robert W. Kennard, 1970, "Ridge regression: Biased estimation for nonorthogonal problems," Technometrics, Vol. 12.1: pp. 55-67; ​​Support Vector Machines: Cortes, Corinna, and Vladimir Vapnik, 1999, "Ridge regression: Biased estimation for nonorthogonal problems," Technometrics, Vol. 12.1: pp. 55-67; Examples of such methods include, but are not limited to, Vapnik, 1995, "Support-vector networks," Machine learning, vol. 20.3: pp. 273-297; or deep learning: LeCun, Yann, Yoshua Bengio, and Geoffrey Hinton, 2015, "Deep learning," Nature, vol. 521.7553: pp. 436. Generally, gene-level CNA scores were first collapsed to segment-level CNA scores. A complete list of genes in each segment is shown in Supplementary Table 2. 18-22Each segment score was calculated as the average CNA score of the genes in the segment. For each cancer phenotype, the total samples were split into a 70% training set and a 30% test set (R package sampling) and stratified by clinical variables: overall survival, sex, race, ER status, PR status, and HER2 status, histological subtype, pathological stage, and molecular subtype when available. Models were built from the training set only; 200 rounds of Monte Carlo cross-validation (R package caret) were used to select tuning parameters. Lambda values ​​were chosen from an alpha range of 0.1 to 1 in increments of 0.1. The optimal parameter combination was determined to have the best classification accuracy. Then, the model with the optimal parameters was applied to the test set and other validation sets, when available. Receiver operating characteristic (ROC) curves were constructed and the area under the ROC curve (AUC) values ​​were used to evaluate model performance. Phenotypes with AUC values ​​above 0.75 are considered highly predictive.

[0092] Models were built to classify samples with high scores (top third) versus low scores (bottom two thirds) to predict gene expression signatures, protein expression, and mutation burden with continuous scores. Models were built to classify molecular subtype, clinical receptor status, cancer histology, and mutations with binary outcomes. For breast cancer gene expression signatures, normal-like samples were excluded (n=1038) as per the association tests described above. For somatic mutations, all IGH and HLA genes were removed and the 2015 TCGA lobular breast cancer analysis 34 Only genes with a mutation frequency of >5% and / or significantly mutated genes identified in were included (Supplementary Table 15).

[0093] For subtype-specific gene signature prediction, the same elastic net model approach was repeated within samples of a particular subtype split into 70% training and 30% testing, and the model was applied to METABRIC samples with the same subtype.

[0094] For gene signature and histology prediction using non-small cell lung cancer data, the full TCGA lung dataset combining both LUAD (n=498) and LUSC (n=512) was used. These were split into training and test sets and homogenized for clinical variables: overall survival, sex, pathological stage, and histology (LUAD or LUSC). Models were built from the training set and applied to the test set. Also, models built from the TCGA breast cancer training set were applied to the full lung dataset. Models were also built within LUAD and LUSC separately.

[0095] For pan-cancer gene expression signature prediction, the analysis was limited to tumor types with at least 100 samples with RNA, DNA, and clinical data. 504 median expression-based gene signatures were applied to each tumor type. For each signature prediction for each tumor type, the total samples were split into a 70% training set and a 30% test set, homogenized for gender, race, and overall survival. Models were then built from the training set and applied to the test set to obtain training and test AUC values.

[0096] To view the features selected by each prediction model, the coefficients of the CNA segments were remapped to the genes within each segment and plotted. A summary of all elastic net models, including the coefficients of the CNA segments and AUC values, is reported in the Supplementary Table.

[0097] Data Availability. The datasets generated and / or analyzed during the current study are available from this disclosure and its Supplementary Information tables. All raw and primary data come from the TCGA and METABRIC public data repositories.

[0098] Application to different copy number aberration (CNA) data sources. method Copy number alteration profiles determined by DNA exome sequencing were called by CNVkit for 1067 breast cancer samples from the TCGA Breast Cancer Project. 50 536 segment values ​​were calculated as described above. Elastic net predictions of key gene expression signatures, clinical assays, clinical receptor status, and molecular subtypes were made by applying segment values ​​from exome sequencing to existing models built from SNP array copy number data as described herein. Area under the receiver operating curve (AUC) values ​​were calculated to reflect model performance.

[0099] result To examine whether the elastic net model can be called from DNA exome sequencing data, copy number data determined by DNA exome sequencing was applied to models constructed from DNA SNP array data for representative highly predictive phenotypes. The results shown in Figures 17A-17D demonstrate that all 14 phenotypes, including four key gene expression signatures, three clinical assays, three receptor statuses, and four molecular subtypes, have high AUC values ​​(>=0.88), demonstrating the applicability of the models and methods disclosed herein to DNA copy number data from different platforms.

[0100] 7. References 1 Perou, CM et al., Molecular portraits of human breast tumours, Nature, vol. 406, pp. 747-752, doi:10.1038 / 35021093 (2000). 2 “Comprehensive molecular portraits of human breast tumours” Nature 490, 61-70, doi:10.1038 / nature11412 (2012). 3 Curtis, C. et al., "The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups," Nature, 486, 346-352, doi:10.1038 / nature10983 (2012). 4 Hoadley, KA et al., "Cell-of-Origin Patterns Dominate the Molecular Classification of 10,000 Tumors from 33 Types of Cancer," Cell, 173, 291-304 e296, doi:10.1016 / j.cell.2018.03.022 (2018). 5 Nevins, JR, & Potti, A., Mining gene expression profiles: expression signatures as cancer phenotypes. Nature Reviews Genetics, 8, 601-609, doi:10.1038 / nrg2137 (2007). 6 Bild, AH et al., Oncogenic pathway signatures in human cancers as a guide to targeted therapies. Nature 439:353-357, doi:10.1038 / nature04296 (2006). 7 Kwa, M., Makris, A. & Esteva, FJ. Clinical utility of gene-expression signatures in early stage breast cancer. Nature Reviews Clinical Oncology, vol. 14, p. 595-610, doi:10.1038 / nrclinonc.2017.74 (2017). 8 Gatza, ML, Silva, GO, Parker, JS, Fan, C., & Perou, CM. An integrated genomics approach identifies drivers of proliferation in luminal-subtype human breast cancer. Nature genetics 46, 1051-1059, doi:10.1038 / ng.3073 (2014). 9 Zou, H. & Hastie, T., "Regularization and variable selection via the elastic net," Journal of the Royal Statistical Society, Vol. B 67, p. 301, 2005, doi:DOI 10.1111 / j.1467-9868.2005.00527.x (2005). 10 Paik, S. et al., Gene expression and benefit of chemotherapy in women with node-negative, estrogen receptor-positive breast cancer, Journal of Clinical Oncology: official journal of the American Society of Clinical Oncology, vol. 24, pp. 3726-3734, doi:10.1200 / JCO.2005.04.7985 (2006). 11 Parker, JS et al., "Supervised risk predictor of breast cancer based on intrinsic subtypes," Journal of Clinical Oncology: official journal of the American Society of Clinical Oncology, vol. 27, pp. 1160-1167, doi:10.1200 / JCO.2008.18.1370 (2009). 12 van de Vijver, MJ et al., A gene-expression signature as a predictor of survival in breast cancer. The New England Journal of Medicine, 347, 1999-2009, doi:10.1056 / NEJMoa021967 (2002). 13 Tanioka, M. et al., "Integrated Analysis of RNA and DNA from the Phase III Trial CALGB 40601 Identifies Predictors of Response to Trastuzumab-Based Neoadjuvant Chemotherapy in HER2-Positive Breast Cancer." Clinical Cancer Research: an official journal of the American Association for Cancer Research, doi:10.1158 / 1078-0432.CCR-17-3431 (2018). 14 Benjamini, Y. & Hochberg, Y., "Controlling the False Discovery Rate - a Practical and Powerful Approach to Multiple Testing," Journal of the Royal Statistical Society, B Met, Vol. 57, pp. 289-300 (1995). 15 Herschkowitz, JI, He, X., Fan, C., & Perou, C. M. The functional loss of the retinoblastoma tumour suppressor is a common event in basal-like and luminal B breast carcinomas. Breast Cancer Research: BCR 10, R75, doi:10.1186 / bcr2142 (2008). 16 Knudsen, ES & Wang, JY. Targeting the RB-pathway in cancer therapy. Clinical Cancer Research: an official journal of the American Association for Cancer Research, vol. 16, pp. 1094-1099, doi:10.1158 / 1078-0432.ccr-09-0787 (2010). 17 Hoadley, KA et al., "Multiplatform analysis of 12 cancer types reveals molecular classification within and across tissues of origin," Cell 158, 929-944, doi:10.1016 / j.cell.2014.06.049 (2014). 18 Beroukhim, R. et al., "The landscape of somatic copy-number alteration across human cancers," Nature, 463, 899-905, doi:10.1038 / nature08822 (2010). 19 Zack, TI et al., "Pan-cancer patterns of somatic copy number alteration," Nature Genetics, vol. 45, p. 1134-1140, doi:10.1038 / ng.2760 (2013). 20 Silva, GO et al., "Cross-species DNA copy number analyses identify multiple 1q21-q23 subtype-specific driver genes for breast cancer." Breast Cancer Research and Treatment 152:347-356, doi:10.1007 / s10549-015-3476-2 (2015). 21 Weigman, VJ et al., Basal-like Breast cancer DNA copy number losses identify genes involved in genomic instability, response to therapy, and patient survival. Breast Cancer Research and Treatment, 133, 865-880, doi:10.1007 / s10549-011-1846-y (2012). 22 Chao, HH, He, X., Parker, JS, Zhao, W. & Perou, CM. Micro-scale genomic DNA copy number aberrations as another means of mutagenesis in breast cancer. PLoS One, 7, e51719, doi:10.1371 / journal.pone.0051719 (2012). 23 Fan, C. et al., Building prognostic models for breast cancer patients using clinical variables and hundreds of gene expression signatures, BMC Medical Genomics, vol. 4, p. 3, doi:1755-8794-4-310.1186 / 1755-8794-4-3 (2011). 24 Hoadley, KA et al., EGFR associated expression profiles vary with breast tumor subtype. BMC Genomics, 8:258, doi:10.1186 / 1471-2164-8-258 (2007). 25 Paik, S. et al., A multigene assay to predict recurrence of tamoxifen-treated, node-negative breast cancer. The New England Journal of Medicine, 351, 2817-2826, doi:10.1056 / NEJMoa041588 (2004). 26 van't Veer, LJ et al., Gene expression profiling predicts clinical outcome of breast cancer. Nature, 415, 530-536, doi:10.1038 / 415530a (2002). 27 Liu, Y., Beyer, A., & Aebersold, R., "On the Dependency of Cellular Protein Levels on mRNA Abundance," Cell, 165, 535-550, doi:10.1016 / j.cell.2016.03.014 (2016). 28 Myhre, S. et al., "Influence of DNA copy number and mRNA levels on the expression of breast cancer related proteins," Molecular Oncology, 7, 704-718, doi:10.1016 / j.molonc.2013.02.018 (2013). 29 Geiger, T., Cox, J., & Mann, M., Proteomic changes resulting from gene copy number variations in cancer cells. PLoS genetics, 6, e1001090, doi:10.1371 / journal.pgen.1001090 (2010). 30 Knott, SRV et al., Asparagine bioavailability governs metastasis in a model of breast cancer, Nature, 554, 378-381, doi:10.1038 / nature25465 (2018). 31 Balasubramanian, MN, Butterworth, EA, & Kilberg, MS. Asparagine synthetase: regulation by cell stress and involvement in tumor biology. American journal of physiology Endocrinology and metabolism, 304, E789-799, doi:10.1152 / ajpendo.00015.2013. 32 Horlings, HM et al., "ESR1 gene amplification in breast cancer: a common phenomenon?" Nature Genetics, vol. 40, pp. 807-808; author reply, pp. 810-802, doi:10.1038 / ng0708-807 (2008). 33 Frampton, GM et al., "Development and validation of a clinical cancer genomic profiling test based on massively parallel DNA sequencing," Nature Biotechnology, 31, 1023-1031, doi:10.1038 / nbt.2696 (2013). 34 Ciriello, G. et al., “Comprehensive Molecular Portraits of Invasive Lobular Breast Cancer,” Cell, Vol. 163, p. 506-519, doi:10.1016 / j.cell.2015.09.033 (2015). 35 Liontos, M., Anastasiou, I., Bamias, A. & Dimopoulos, MA. DNA damage, tumor mutational load and their impact on immune responses against cancer. Annals of Translational Medicine, 4:264, doi:10.21037 / atm.2016.07.11 (2016). 36 Van Allen, EM et al., Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science 350, 207-211, doi:10.1126 / science.aad0095 (2015). 37 Iglesia, MD et al., Genomic Analysis of Immune Cell Infiltrates Across 11 Tumor Types, Journal of the National Cancer Institute, vol. 108, doi:10.1093 / jnci / djw144 (2016). 38 Sokol, CL & Luster, AD. The chemokine system in innate immunity. Cold Spring Harbor Perspectives in Biology, vol. 7, doi:10.1101 / cshperspect.a016303 (2015). 39 Wellenstein, MD & de Visser, KE. Cancer-Cell-Intrinsic Mechanisms Shaping the Tumor Immune Landscape. Immunity 48, 399-416, doi:10.1016 / j.immuni.2018.03.004 (2018). 40 Cancer Genome Atlas Research, N., Comprehensive genomic characterization of squamous cell lung cancers, Nature, vol. 489, p. 519-525, doi:10.1038 / nature11404 (2012). 41 Cancer Genome Atlas Research, N., Comprehensive molecular profiling of lung adenocarcinoma, Nature, vol. 511, p. 543-550, doi:10.1038 / nature13385 (2014). 42 Campbell, JD et al., Distinct patterns of somatic genome alterations in lung adenocarcinomas and squamous cell carcinomas, Nature genetics, vol. 48, p. 607-616, doi:10.1038 / ng.3564 (2016). 43 Herschkowitz, JI et al., "Identification of conserved gene expression features between murine mammary carcinoma models and human breast tumors." Genome Biology 8, R76, doi:10.1186 / gb-2007-8-5-r76 (2007). 44 Suarez-Cabrera, C. et al., "The Ras-related gene ERAS is involved in human and murine breast cancer," Scientific Reports, vol. 8, p. 13038, doi:10.1038 / s41598-018-31326-4 (2018). 45 Bailey, MH et al., Comprehensive Characterization of Cancer Driver Genes and Mutations, Cell 173, 371-385 e318, doi:10.1016 / j.cell.2018.02.060 (2018). 46 Ding, L. et al., "Perspective on Oncogenic Processes at the End of the Beginning of Cancer Genomics," Cell, vol. 173, p. 305-320 e310, doi:10.1016 / j.cell.2018.03.033 (2018). 47 Asghar, U., Witkiewicz, AK, Turner, NC & Knudsen, ES. The history and future of targeting cyclin-dependent kinases in cancer therapy. Nature Reviews Drug Discovery, 14, 130-146, doi:10.1038 / nrd4504 (2015). 48 Mermel, CH et al., GISTIC2.0 facilitates sensitive and confident localization of the targets of focal somatic copy-number alteration in human cancers. Genome Biology 12, R41, doi:10.1186 / gb-2011-12-4-r41 (2011). 49 Subramanian, A. et al., "Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles," Proceedings of the National Academy of Sciences of the United States of America, vol. 102, pp. 15545-15550, doi:10.1073 / pnas.0506580102 (2005). 50 Talevich E, Shain AH, Botton T, Bastian BC (2016). CNVkit: Genome-Wide Copy Number Detection and Visualization from Targeted DNA Sequencing. PLoS Computational Biology 12(4): e1004873.

[0101] 8. Generalized Claims of This Disclosure The following numbered claims provide a general description of the disclosure, but are not intended to limit the appended claims.

[0102] Claim 1: A method for generating a computed cancer signature of a patient sample, comprising: (a) obtaining or having obtained a sample from the patient; (b) measuring or having measured multiple copy number aberrations (CNAs) at multiple locations on multiple chromosomes; and (c) analyzing the measured CNAs using a mathematical model based on mRNA expression data and molecular subtypes, wherein the mathematical model is validated by at least two different statistical methods to generate a computed cancer signature of the sample.

[0103] Claim 2: The method of claim 1, wherein a CNA of more than 50 is measured.

[0104] Claim 3: The method of claim 1, wherein more than 100 CNA is measured.

[0105] Claim 4: The method of claim 1, wherein a CNA of between about 250 and about 400 is measured.

[0106] Claim 5: The method of any of claims 1 to 4, wherein the calculated cancer signature corresponds to a somatic mutation signature.

[0107] Claim 6: The method of claim 5, wherein the mathematical model for determining the somatic mutation signature is based on a beta coefficient value of 10 or greater in Supplementary Table 6.

[0108] Claim 7: The method of any of claims 1 to 4, wherein the calculated cancer signature corresponds to an mRNA expression signature.

[0109] Claim 8: The method of claim 7, wherein the calculated cancer signature is a signature of a breast cancer subtype.

[0110] Claim 9: The method of claim 7, wherein the mathematical model for determining the breast cancer subtype signature is based on a beta coefficient value of 10 or greater in Supplementary Table 4.

[0111] Claim 10: The method of any of claims 1 to 4, wherein the calculated cancer signature corresponds to a protein expression signature.

[0112] Claim 11: The method of claim 10, wherein the mathematical model for analyzing the protein expression signature is based on a beta coefficient value of 10 or greater in Supplementary Table 5.

[0113] Claim 12: The method of claim 10, wherein the protein expression signature is an immunohistochemistry (IHC) signature.

[0114] Claim 13: The method of claim 12, wherein the IHC signature is an estrogen receptor (ER), epidermal growth factor receptor (EGFR), human epidermal growth factor receptor 2 (HER2), progesterone receptor (PR), or retinoblastoma (RB) signature.

[0115] Claim 14: Any of the methods of claims 1 to 4, wherein the calculated cancer signature corresponds to a FoundationOne® CDX result, a MAMMAPRINT® 70-gene recurrence score, an OncotypeDX™ recurrence score, or a Prosigna® recurrence risk score.

[0116] Claim 15: The method of claim 14, wherein the calculated cancer signature is a FoundationOne® result and the mathematical model for fitting the FoundationOne® result is based on a beta coefficient value of 10 or greater in Supplementary Table 9.

[0117] Claim 16: Any of claims 1 to 4, wherein the calculated cancer signature is selected from the group consisting of bladder urothelial carcinoma (BLCA), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), clear cell renal carcinoma (KIRC), papillary renal cell renal carcinoma (KIRP), acute myeloid leukemia (LAML), and brain low-grade glioma (LGG). , hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), lung squamous cell carcinoma (LUSC), pancreatic adenocarcinoma (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate adenocarcinoma (PRAD), rectal adenocarcinoma (READ), sarcoma (SARC), skin cutaneous melanoma (SKCM), gastric adenocarcinoma (STAD), testicular germ cell tumor (TGCT), thyroid carcinoma (THCA), thymoma (THYM), or uterine endometrial carcinoma (UCEC) signature.

[0118] Claim 17: The method of claim 16, wherein the mathematical model for fitting the calculated signature is based on the beta coefficient values ​​in Supplementary Table 14.

[0119] Claim 18: The method of any of claims 1 to 17, wherein the multiple copy number abnormalities (CNAs) are obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0120] Claim 19: A method for treating a cancer patient with chemotherapy, comprising the steps of determining whether the patient has a particular cancer subtype: (a) obtaining or having obtained a biological sample from the patient; (b) performing or having performed a gene-level copy number aberration (CNA) assay on the biological sample, in which copy number is measured at multiple locations on multiple chromosomes; (c) comparing results of the CNA assay to a standard set to determine whether the patient has a particular cancer subtype; and (d) if the patient has a particular cancer subtype, administering an appropriate chemotherapy regimen to the cancer patient based on the determined cancer subtype.

[0121] Claim 20: The method of claim 19, wherein the chemotherapy regimen is an ongoing therapeutic intervention.

[0122] Claim 21: The method of claim 20, where the ongoing intervention includes discontinuing a particular procedure.

[0123] Claim 22: Any of the methods of claims 19 to 21, wherein the copy number aberration (CNA) assay is obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0124] Claim 23: A method for generating a computed cancer signature of a cancer phenotype, the method including: (a) receiving a plurality of gene expression signatures and subtype information of the cancer phenotype; (b) receiving a plurality of copy number aberration (CNA) datasets of the cancer phenotype; (c) analyzing the plurality of CNA datasets using an artificial intelligence algorithm to obtain a preliminary set of CNA segment level signatures of the cancer phenotype; (d) refining the preliminary set of CNA segment level signatures using a gene expression training set to obtain a final set of CNA segment level signatures; and (e) using the final set of CNA segment level signatures to refine the computed cancer signature of the cancer phenotype.

[0125] Claim 24: The method of claim 23, wherein the cancer phenotype is associated with a somatic mutation.

[0126] Claim 25: The method of claim 23, wherein the cancer phenotype corresponds to a level of mRNA expression.

[0127] Claim 26: The method of claim 23, wherein the cancer phenotype corresponds to a level of protein expression.

[0128] Claim 27: The method of claim 26, wherein the level of protein expression corresponds to an immunohistochemistry (IHC) signature.

[0129] Claim 28: Any of claims 23 to 26, wherein the multiple copy number abnormality (CNA) datasets are obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0130] Claim 29: A method for generating a computed cancer signature for a patient, the method comprising: (a) receiving copy number aberration (CNA) data for the patient; (b) receiving one or more CNA(s) signature(s) associated with a cancer phenotype, the CNA signature(s) being based on cancer expression analysis, cancer subtype information, and CNA gain / loss information; (c) processing the patient's CNA data using an algorithm that utilizes the one or more CNA(s) signature(s) associated with the cancer phenotype in a manner that characterizes a characteristic of the CNA data of the patient relative to the one or more CNA(s) signature(s); and (d) interrogating the patient's computed cancer signature.

[0131] Claim 30: The method of claim 29, wherein the cancer phenotype is associated with a somatic mutation.

[0132] Claim 31: The method of claim 29, wherein the cancer phenotype corresponds to a level of mRNA expression.

[0133] Claim 32: The method of claim 29, wherein the cancer phenotype corresponds to a level of protein expression.

[0134] Claim 33: The method of claim 29, wherein the level of protein expression corresponds to an immunohistochemistry (IHC) signature.

[0135] Claim 34: The method of claim 29, wherein the cancer phenotype is associated with adrenal, bladder, bone, breast, cervical, colon, liver, lung, lymph, ovarian, pancreatic, penile, prostate, rectal, salivary gland, skin, spleen, testicular, thymus, thyroid, trachea, or uterine cancer.

[0136] Claim 35: The method of claim 41, wherein the cancer phenotype is associated with breast cancer.

[0137] Claim 36: The method of any of claims 29 to 35, wherein the copy number abnormality (CNA) data is obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0138] Claim 37: A method for treating a subject having cancer, comprising: (a) generating a calculated cancer signature for the patient, comprising: (i) receiving copy number aberration (CNA) data for the patient; (ii) receiving one or more CNA(s) signature(s) associated with a cancer phenotype, the CNA signature(s) being based on cancer expression analysis, cancer subtype information, and CNA gain / loss information; (iii) processing the patient's CNA data using an algorithm that utilizes the one or more CNA(s) signature(s) associated with the cancer phenotype to characterize characteristics of the CNA data of patient characteristics relative to the one or more CNA(s) signature(s); (iv) adjusting the patient's calculated cancer signature based on the characterized characteristics; and (b) treating the patient based on a treatment plan based on the calculated cancer signature.

[0139] Claim 38: The method of claim 37, wherein the treatment is an ongoing therapeutic intervention.

[0140] Claim 39: The method of claim 37, where the ongoing intervention includes discontinuing a particular procedure.

[0141] Claim 40: Any of the methods of claims 37-39, wherein the copy number abnormality (CNA) data is obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0142] Claim 41: A device including a processor configured to process patient CNA data and one or more CNA(s) signature(s) associated with a cancer phenotype according to an algorithm to generate a calculated cancer signature for the patient of claim 29.

[0143] Allegation 42: A system including the device of Allegation 41.

[0144] Claim 43: The device of claim 41, including software including an algorithm for comparing patient CNA data to one or more CNA(s) signature(s) associated with a cancer phenotype.

[0145] Claim 36: The device of any of claims 41 to 43, wherein the copy number abnormality (CNA) data is obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

[0146] It should be understood that the above description is only representative of illustrative embodiments and examples. For the convenience of the reader, the above description has focused on a limited number of representative examples of all possible embodiments, examples that teach the principles of the present disclosure. The description does not attempt to exhaustively list all possible variations, or even combinations of variations described. The fact that alternative embodiments may be possible that are not presented in a particular part of the disclosure, or that further alternative embodiments not described may be available in some parts, should not be considered as a waiver of such alternative embodiments. Those skilled in the art will appreciate that many of these undescribed embodiments involve differences in technology and materials rather than differences in the application of the principles of the present disclosure. Thus, the present disclosure is not intended to be limited to less than the scope and equivalents set forth in the following claims.

[0147] Inclusion by reference All references, articles, publications, patents, patent publications, and patent applications cited herein are incorporated by reference in their entirety for all purposes. However, mention of any references, articles, publications, patents, patent publications, and patent applications cited herein is not, and should not be taken as, an acknowledgment or any form of indication that they constitute formal prior art or form part of the common general knowledge in any country in the world. Although the present disclosure has been described in its detailed description, it should be understood that the above description is intended to illustrate, but not limit, the scope. Other aspects, advantages, and modifications are within the scope of the claims filed below. All publications, patents, and patent applications cited herein are incorporated by reference herein, to the same extent as if each individual publication or patent application was specifically and individually indicated to be incorporated by reference.

[0148] 9. Computer Programs DNA-based-predictors-of-non-genetic-cancer-phenotypes Analysis scripts and relative data used in the paper "Genetic determinants of the molecular portraits of epithelial cancers" by Xia et al. The data contains many .rda files ready for analysis and can be found at https: / / github.com / xyouli / DNA-based-predictors-of-non-genetic-cancer-phenotypes.

[0149] R script helper.R: This script has many R functions that can be used in the analysis. signature_score _and_segment_score_calculation.R: This script is used to calculate the signature score from RNA expression data and the segment score from gene-level copy number data. association_test.R: This script is used to perform genome-wide association tests between gene signatures and DNA copy number abnormalities. Elastic_Net_modeling.R: This script is used to perform the elastic net modeling analysis. All data used to build the elastic net model is contained in the data folder. Rscript / Elastic_Net_modeling.R 57 lines (43 sloc) 1.81 library(caret) library(glmnet) library(ROCR) library(doMC) registerDoMC(cores = 8) set.seed(1992) source("~ / script / helper.R") # load data, BRCA_signatue for example # predictor: CN_score # response: signature_score / rppa / mutation / clinical variables # for balanced stratification: balancing_variables load('~ / data / BRCA_signature_elastic_net_data.rda') # choose a phenotype to model # for example, RB-LOH signature pheno <- "UNC_RB_LOH_Median_Breast.Cancer.Res.2008_PMID.18782450" score <- unlist(signature_score[pheno,]) score_bi <- ifelse(score >= quantile(score,0.67),1,0) # split sample into 70% training set and 30% test set strata <- score_bi pik <- rep(7 / 10,times=length(strata)) balanced_split <- balancedstratification(balancing_variables,strata,pik,comment = F) train <- which(balanced_split==1) test <- which(balanced_split==0) trainX <- CN_score[train,] testX <- CN_score[test,] trainY <- score_bi[train] testY <- score_bi[test] glmnet_obj <- caret_wrap(trainX,trainY,testX,testY,bi = T) # look at model performance: receiving operating curve and precision recall curve pred_train <- predict(glmnet_obj,newdata = trainX,type = 'prob') pred_test <- predict(glmnet_obj,newdata = testX,type = 'prob') pred_train <- prediction(pred_train$pos,labels = trainY) pred_test <- prediction(pred_test$pos,labels = testY) perf_train <- performance(pred_train,measure = 'tpr',x.measure = 'fpr') perf_test <- performance(pred_test,measure = 'tpr',x.measure = 'fpr') auc_train <- signif(performance(pred_train,measure = 'auc')@y.values[[1]][1],2) auc_test <- signif(performance(pred_test,measure = 'auc')@y.values[[1]][1],2) plot_ROC(perf_train,perf_test,auc_train,auc_test,pheno) # look at feature landscape beta <- as.matrix(coef(glmnet_obj$finalModel,glmnet_obj$bestTune$lambda)) beta <- beta[-1,1] plot_seg_ss(beta,pheno) -------------------------------------------------------------------------------- DNA-based-predictors-of-non-genetic-cancer-phenotypes / Rscript / association_test.R 27 lines (22 sloc) 1.14 KB # Test for associations between gene signatures and DNA CNA # Given a signature_score matrix and # gene-level CNA matrix CN_score: continuous CNA score for each gene, # CN_gain: binary matrix where 1 is copy number gain for a gene and 0 is no gain # CN_loss: binary matrix where 1 is copy number loss for a gene and 0 is no loss # Fisher's exact test and spearman correlation test / linear model control for subtypes # choose a gene signature to test, for example RB-LOH signature load('BRCA_association_test_data.rda') pheno <- "UNC_RB_LOH_Median_Breast.Cancer.Res.2008_PMID.18782450" score <- unlist(signature_score[pheno,]) p_value <- sigCNTest(score, CN_score, CN_gain, CN_loss) # Benjamini-Hochberg correct p values p_value_adj <- apply(p_value,2,function(x){return(p.adjust(x,method = "BH"))}) # log10 transform adjusted p values log_p <- -log(p_value_adj,10) # plot association landscape v <- vertical_lines # for unadjusted associations p <- log_p[,1:4] # for subtype-adjusted associations p <- log_p[,c(5:6,3:4)] P_Plot(p,main = 'UNC_RB_LOH_Median_Breast.Cancer.Res.2008_PMID.18782450',y1 = 35,y2=50,label_up = seq(0,35,by=10),label_down = seq(10,50,by=10)) -------------------------------------------------------------------------------- DNA-based-predictors-of-non-genetic-cancer-phenotypes / Rscript / signature_score_and_segment_score_calculation.R 48 lines (34 sloc) 2.2 KB # This script is for calculating gene signature scores based on RNA and segment scores based on gene-level DNA CNA # Given a gene expression data matrix (gene X sample): edata # rows are genes in Entrez ID and columns are samples # run calc_signatures signature_score <- calc_signatures(edata,"~ / data / gene_signatures_20170111.gmt",method = "median") # find NA signatures NAsig <- rownames(signature_score[is.na(signature_score[,1]),]) # remove NA signatures i <- match(NAsig,rownames(signature_score)) signature_score <- signature_score[-i,] # CD103_Ratio CD103_pos <- signature_score[rownames(signature_score)=="CD103_Positive_Median_Cancer.Cell.2014_PMID.25446897"] CD103_neg <- signature_score[rownames(signature_score)=="CD103_Negative_Median_Cancer.Cell.2014_PMID.25446897"] CD103_ratio <- CD103_pos - CD103_neg # log2 scale division signature_score <- rbind(signature_score,CD103_ratio) rownames(signature_score)[nrow(signature_score)] <- "CD103_Ratio_Cancer.Cell.2014_PMID.25446897" # differentiation score diff_centroid <- read.table("~ / data / special_gene_signature_training_sets / UNC_Differentiation.Score_Model_BCR.2010_PMID.20813035",sep = '\t',header = T,row.names = 1,check.names = F) diff_score <- assignDiffScore.dwd(diff_centroid,edata) signature_score <- rbind(signature_score,diff_score) rownames(signature_score)[nrow(signature_score)] <- "UNC_Differentiation.Score_Model_BCR.2010_PMID.20813035" # oncotype DX score oncotype <- GHI_RS(edata) signature_score <- rbind(signature_score,diff_score) rownames(signature_score)[nrow(signature_score)] <- "GHI_RS_Model_NJEM.2004_PMID.15591335" save(signature_score,file = 'signature_score.rda') # For special signatures calculated as correlation to predetermined gene centroids # All traing sets files are included in the ~ / data / special_gene_signature_training_sets folder # For calculation of such special signature, DWD was used to merge current edata with # training set, then DWD prediction tool was used to compute pearson / spearman correlation / euclidean distance # for each sample # Given a gene-level CNA score matrix (gene X sample): CNdata # rows are genes in Entrez ID and columns are samples segment_score <- calc_segments(CNdata,'CNA_segments.gmt',method = 'mean')

Claims

1. A method for generating a calculated cancer signature associated with a cancer phenotype for a patient sample using only copy number alteration (CNA) data, comprising: (a) measuring or having measured multiple CNAs at multiple locations on multiple chromosomes; (b) analyzing the measured CNAs using a mathematical model based on mRNA expression data, protein expression, somatic mutations, clinical features, and / or molecular subtypes; Including, The mathematical model is statistically validated to generate a calculated cancer signature for the sample. method.

2. The method of claim 1 , wherein more than 50 CNAs are measured.

3. The method of claim 1 , wherein more than 100 CNAs are measured.

4. The method of claim 1, wherein a CNA of between 250 and 400 is measured.

5. The method of claim 1 , wherein the calculated cancer signature corresponds to a somatic mutation signature.

6. The method of claim 1, wherein the calculated cancer signature corresponds to an mRNA expression signature.

7. The method of claim 1, wherein the calculated cancer signature is a signature of a breast cancer subtype.

8. The method of claim 1, wherein the calculated cancer signature corresponds to a protein expression signature.

9. The method described in claim 8, wherein the protein expression signature is an immunohistochemistry (IHC) signature.

10. The method of claim 9, wherein the IHC signature is an estrogen receptor (ER), epidermal growth factor receptor (EGFR), human epidermal growth factor receptor 2 (HER2), progesterone receptor (PR), or retinoblastoma (RB) signature.

11. The calculated cancer signature is selected from the group consisting of bladder urothelial carcinoma (BLCA), cervical squamous cell carcinoma and adenocarcinoma (CESC), colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), glioblastoma multiforme (GBM), head and neck squamous cell carcinoma (HNSC), clear cell renal carcinoma (KIRC), papillary renal cell renal carcinoma (KIRP), acute myeloid leukemia (LAML), brain low-grade glioma (LGG), hepatocellular carcinoma (LIHC), lung adenocarcinoma (LAML), and pulmonary adenocarcinoma (PAC).

2. The method of claim 1, wherein the cancer is a lung cancer (LUSC), lung squamous cell carcinoma (UAD), pancreatic adenocarcinoma (PAAD), pheochromocytoma and paraganglioma (PCPG), prostate adenocarcinoma (PRAD), rectal adenocarcinoma (READ), sarcoma (SARC), skin cutaneous melanoma (SKCM), gastric adenocarcinoma (STAD), testicular germ cell tumor (TGCT), thyroid cancer (THCA), thymoma (THYM), or uterine endometrial carcinoma (UCEC) signature.

12. The method described in claim 1, wherein multiple copy number abnormalities (CNAs) are obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

13. A method for generating a computed cancer signature of a cancer phenotype using only copy number alteration (CNA) data, the method comprising: (a) receiving a plurality of gene expression signatures, protein expression, somatic mutations, clinical features, and / or subtype information of a cancer phenotype; (b) receiving a plurality of copy number aberration (CNA) datasets of a cancer phenotype; (c) analyzing the plurality of CNA datasets with an artificial intelligence algorithm to obtain a preliminary set of CNA segment level signatures of cancer phenotypes; (d) refining the preliminary set of CNA segment level signatures using the gene expression training set to obtain a final set of CNA segment level signatures; (e) using the final set of CNA segment level signatures to determine a calculated cancer signature of the cancer phenotype; Including, method.

14. The method of claim 13, wherein the cancer phenotype is associated with a somatic mutation.

15. The method described in claim 13, wherein the cancer phenotype corresponds to the level of mRNA expression.

16. The method of claim 13, wherein the cancer phenotype corresponds to the level of protein expression.

17. The method described in claim 13, wherein the level of protein expression corresponds to an immunohistochemistry (IHC) signature.

18. The method of claim 13, wherein the multiple copy number abnormality (CNA) datasets are obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

19. A method for generating a computed cancer signature of a cancer phenotype for a patient using only copy number alteration (CNA) data, the method comprising: (a) receiving copy number aberration (CNA) data for a patient; (b) receiving one or more CNA(s) signature(s) associated with a cancer phenotype, the CNA signature(s) being based on cancer expression analysis, protein expression, somatic mutations, clinical features, cancer subtype information, and / or CNA gain / loss information; (c) processing the patient's CNA data with an algorithm that utilizes the one or more CNA(s) signature(s) associated with a cancer phenotype in a manner that characterizes the patient's characteristic CNA data relative to the one or more CNA(s) signature(s); (d) determining a computed cancer signature of the cancer phenotype for the patient; and Including, method.

20. The method of claim 19, wherein the cancer phenotype is associated with a somatic mutation.

21. The method of claim 19, wherein the cancer phenotype corresponds to the level of mRNA expression.

22. The method of claim 19, wherein the cancer phenotype corresponds to the level of protein expression.

23. The method described in claim 19, wherein the level of protein expression corresponds to an immunohistochemistry (IHC) signature.

24. The method of claim 19, wherein the cancer phenotype is associated with adrenal, bladder, bone, breast, cervical, colon, liver, lung, lymph, ovarian, pancreatic, penile, prostate, rectum, salivary gland, skin, spleen, testis, thymus, thyroid, trachea, or uterine cancer.

25. The method of claim 24, wherein the cancer phenotype is associated with breast cancer.

26. The method of claim 19, wherein copy number abnormality (CNA) data is obtained from whole genome sequencing (WGS), whole exome sequencing (WES), or a combination thereof.

Images