Bi-Level Groupwise Multiple Features Selection Method for Identifying Novel Biomarkers in Pharmacogenetic Studies

• Activity Number: 80
• Type: Contributed
• Date/Time: Sunday, August 4, 2013 : 4:00 PM to 5:50 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Ting-Huei Chen*+ and Wei Sun
• Companies: UNC at Chapel Hill and UNC Chapel Hill
• Keywords: pharmacogenetics ; biomarker ; drug-response

Abstract

Pharmacogenetics aims to study the inherited variation in drug response to enhance drug efficacy and safety. Cancer cell lines have been widely studied to link pharmacological data with genomic factors to find therapeutic biomarkers. Most studies conduct drug-by-drug analysis to find associated biomarkers. However, since drugs can be grouped by their targets, it is of interest to find the associated biomarkers by jointly analysis of a group of drugs. To this purpose, we propose a bi-level groupwise multiple features selection method, which incorporates two levels of penalties to enhance the selection of features associated to multiple responses. Our method can identify the features associated with the sensitivity or resistance to the drugs in each target group. Applying our method to a drug response data set, we have identified several cases where bi-level selection delivered interesting results that were missed by drug-by-drug analysis. For example, we have found that drugs targeting ABL are associated with EGFR, BCR-ABL, AZU1, and CAV2, and drugs targeting PI 3-kinase are associated with YAP1, MYB, and AXL. Those identified genes are either oncogenes or related to tumorigenesis.