Abstract 5035: IDWAS: Imputing drug response in large cohorts of cancer patients to discover novel predictive biomarkers

Abstract

Abstract There are currently only 10 cancer genes with FDA approved drug treatment options (source: OncoKB). Here, we present a conceptually novel analytical method that can rapidly expand this small list of clinically effective cancer drug treatment biomarkers. Large sequencing studies, such as The Cancer Genome Atlas (TCGA), have greatly accelerated our understanding of the molecular basis of cancer. However, because of the difficulty in collecting drug response data in large cohorts of cancer patients, these studies have not been effectively used for finding new biomarkers of drug response (i.e. pharmacogenomics discovery). Thus, much cancer pharmacogenomics research is conducted in pre-clinical disease models such as cell lines (e.g. the Genomics of Drug Sensitivity in Cancer (GDSC) project); but these studies are (among other limitations) always restricted by comparatively smaller sample sizes. Here, we present an analytical method that integrates data from large clinical studies (e.g. TCGA) with data from pre-clinical disease models (e.g. GDSC) and overcomes these critical obstacles, allowing studies such as TCGA to now be effectively used for pharmacogenomics discovery. We refer to this approach as an “Imputed drug-wide association study” (IDWAS). The method works by fitting a statistical model relating gene expression and drug response in pre-clinical data (here we use the GDSC cancer cell lines), then using this model to impute drug response from tumor gene expression data in a clinical cohort (here we use TCGA). Next, we compare these imputed drug response data to measured variants (e.g. somatic mutations, copy number changes) in TCGA, thus finding new biomarkers of drug response. We show that we can recapitulate known clinically effective biomarkers and we have validated new clinically relevant biomarkers, which remarkably could not have been identified using conventional approaches. Our method will set the stage for many future studies. Crucially, this approach could easily be applied to any of the vast number of clinical cancer sequencing studies now undertaken, meaning that it will be possible to use all of these datasets for pharmacogenomics research. We have included a set of computational tools to allow easy application of our method and replication of our results. Given that this is a conceptually novel methodology, it is also likely that many other studies will attempt to improve upon our proposed implementation. Furthermore, members of our group are currently leading the development of the new Genomic Data Commons (GDC; https://gdc.cancer.gov/), which is the NCI’s new access portal for TCGA data. We are currently working towards making the imputed drug response data directly accessible on GDC, along with all other TCGA data. This means that our imputed drug response data will be easily accessible to the thousands of researchers already accessing TCGA via the GDC. Citation Format: Paul Geeleher, Zhenyu Zhang, Fan Wang, Aritro Nath, Steven Bhutra, Robert Grossman, R. Stephanie Huang. IDWAS: Imputing drug response in large cohorts of cancer patients to discover novel predictive biomarkers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2017; 2017 Apr 1-5; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2017;77(13 Suppl):Abstract nr 5035. doi:10.1158/1538-7445.AM2017-5035