Developing Polygenic Risk Prediction Models for Cancer Subtypes Incorporating Multivariate Disease Characteristics

• Activity Number: 679 - Variable Selection and Prediction Models for Genomic Data
• Type: Contributed
• Date/Time: Thursday, August 2, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Genomics and Genetics
• Author(s): Haoyu Zhang* and Thomas U. Ahearn and Ni Zhao and Montserrat García-Closas and Nilanjan Chatterjee
• Companies: Johns Hopkins University and National Cancer Institute and Johns Hopkins University and National Cancer Institute and Johns Hopkins University
• Keywords: polygenic risk score; tumor heterogeneity; two-stage model

Abstract

Recent advance in genetic research has allowed the identification of hundreds of genetic loci related to cancer risk. Many efforts have been spent on developing polygenic risk prediction models for disease preventions. However, currently available prediction models do not account for tumor heterogeneity. Many cancers, including breast cancer, are intrinsically heterogeneous with different subtypes having different etiology, clinically behaviors and genetic risk factors. In this talk, we present a two-stage model that estimates polygenetic risk scores utilizing multiple correlated tumor characteristics. Our method can not only model the risk of different cancer subtypes, but also efficiently handle missing tumor characteristics. It allows the analysis to borrow information from cancer cases with incomplete tumor characteristics observations. Several existing polygenetic risk prediction models for cancer subtypes are compared with our proposed method to evaluate the performance. The proposed method is implemented in a large-scale breast cancer GWAS study with 96,317 cases and 111,357 controls, with multiple clinically relevant tumor characteristics included in the analysis.