Abstract 5549: Novel network predictor for drug sensitivity in cell line response data

Abstract

Abstract Sequencing patient tumors has enabled the design of treatment regimens that exploit sensitizing genomic alterations. Associating drugs with mutations that enhance their effect is a key component of the American Society of Clinical Oncology vision for the next two decades of cancer treatment. Methods that distinguish drug sensitizing molecular events from the millions of extraneous alterations common within a tumor will help researchers implement precision medicine strategies. Here we use molecular data from the cancer cell line encyclopedia; including mutations, copy number alterations and gene expression changes along with drug response data, to identify subnetworks of interacting proteins that contain drug sensitizing alterations. Building upon the ability of fuzzy logic models to capture gene activity from different molecular data types, we create ‘Network-FLM’, a method to identify drug sensitizing molecular markers using a subnetwork model that distinguishes a drug sensitive sample from a drug insensitive one. Because the subnetwork model incorporates the sign and direction of network edges and the magnitude of gene activity changes, biologically meaningful features are captured. We integrate protein interaction information from Metacore database with somatic mutation, copy number and gene expression measurements, profiled by CCLE. Using cross validation, we evaluate the mean AUC of the predictor for 24 anti-cancer compounds (Barretina et al., 2012). We build subnetwork classifiers for each compound using mutation, copy number and expression data separately and in all possible combinations. We find that the Network-FLM approach performs well for targeted agents such as Sorafenib, and also for drugs with pleitropic mechanisms of action. Gene expression data alone creates effective predictive subnetworks for 8 compounds (Nilotinib, Sorafenib, Irinotecan, PLX4720, Paclitaxel, Topotecan, TAE684, and Erlotinib). Adding copy number changes to gene expression data created better predictive networks for 8 compounds (PF2341066, AZD6244, L685458, RAF265, PD0325901, ZD6474, PHA665752, and 17-AAG). Likewise, three compounds benefit from adding mutation data to expression data (Panobinostat, PD0332991, and Lapatinib). Combining all datatypes created the best predictors for AEW541 and AZD0530. For three compounds, mutation data alone and together with copy number alterations is the best input for building predictive models. We identified hyper-active subnetworks in cancer cell lines and used them to predict drug sensitivity. We plan to further explore the potential of these networks to improve patient response to anti-cancer drugs. Citation Format: Ana Brandusa Pavel, Bin Li, Andrew Krueger. Novel network predictor for drug sensitivity in cell line response data [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 5549. doi:10.1158/1538-7445.AM2017-5549