Identifying Patient-Specific Biomarker and Predicting Anti-Cancer Drug Sensitivity via Robust Statistical Methodology

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Heewon Park*+ and Teppei Shimamura and Seiya Imoto and Satoru Miyano
• Companies: University of Tokyo and University of Tokyo and University of Tokyo and University of Tokyo
• Keywords: Cancer biomarker ; Drug sensitivity ; Patient-specific analysis ; Robust penalized regression

Abstract

The patient-specific analysis enable to uncover individual genomic characteristics not averaged mechanisms for all patients, and thus we can effectively predict individual risk of disease and perform personalized anti-cancer therapy. Although the existing methods have been successfully uncovered crucial biomarkers, their procedures suffer from outliers, since the methods are based on non-robust manners. In practice, however, clinical and genomic alterations data contains outliers from various sources (e.g., experiment error, coding error, etc.), and outliers significantly disturb patient-specific analysis. We propose robust methodology for patient-specific analysis based on varying coefficient model and L1-type regularization. In the proposed method, outliers in high dimensional genomic data are controlled by Mahalanobis distance via principal component space. Thus, the proposed method effectively performs for patient-specific analysis without disturbance of outliers. We apply the proposed method to Sanger dataset from Cancer Genome Project for uncovering cancer biomarkers and predicting anti-cancer drug sensitivity, and show the effectiveness of our method.