Data Mining Approaches for Genomic Biomarker Development: Applications Using Drug Screening Data from the Cancer Genome Project and the Cancer Cell Line Encyclopedia

Developing reliable biomarkers of tumor cell drug sensitivity and resistance can guide hypothesis-driven basic science research and influence pre-therapy clinical decisions. A popular strategy for developing biomarkers uses characterizations of human tumor samples against a range of cancer drug responses that correlate with genomic change; developed largely from the efforts of the Cancer Cell Line Encyclopedia (CCLE) and Sanger Cancer Genome Project (CGP). The purpose of this study is to provide an independent analysis of this data that aims to vet existing and add novel perspectives to biomarker discoveries and applications. Existing and alternative data mining and statistical methods will be used to a) evaluate drug responses of compounds with similar mechanism of action (MOA), b) examine measures of gene expression (GE), copy number (CN) and mutation status (MUT) biomarkers, combined with gene set enrichment analysis (GSEA), for hypothesizing biological processes important for drug response, c) conduct global comparisons of GE, CN and MUT as biomarkers across all drugs screened in the CGP dataset, and d) assess the positive predictive power of CGP-derived GE biomarkers as predictors of drug response in CCLE tumor cells. The perspectives derived from individual and global examinations of GEs, MUTs and CNs confirm existing and reveal unique and shared roles for these biomarkers in tumor cell drug sensitivity and resistance. Applications of CGP-derived genomic biomarkers to predict the drug response of CCLE tumor cells finds a highly significant ROC, with a positive predictive power of 0.78. The results of this study expand the available data mining and analysis methods for genomic biomarker development and provide additional support for using biomarkers to guide hypothesis-driven basic science research and pre-therapy clinical decisions.