Functional Integrative Bayesian Analysis of High-Dimensional Multiplatform Genetic Data (Functional IBAG)

• Activity Number: 350 - Bayesian Structured Variable Selection
• Type: Topic-Contributed
• Date/Time: Thursday, August 12, 2021 : 10:00 AM to 11:50 AM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Rupam Bhattacharyya* and Nicholas Henderson and Veerabhadran Baladandayuthapani
• Companies: University of Michigan and University of Michigan and University of Michigan
• Keywords: Bayesian Modeling; Hierarchical Priors; Sparse Regression; Data Integration; Precision Oncology; Variable Selection

Abstract

Rapid technological advances have made various types of genomic, epigenomic, transcriptomic and proteomic data available, each data type offering a different, partly independent, and complementary, high-resolution view of the genome. Modeling and inference in such studies is challenging, due to both high dimensionality and presence of structured dependencies but is appealing in terms of understanding the complex biological processes characterizing a disease. We propose an integrative hierarchical Bayesian framework for modeling the fundamental biological relationships underlying the cross-platform molecular features. This allows accounting for both the influences of different platforms, and their mechanistic information, in one unified model to predict patients’ clinical outcomes. Based on sparse regression-based approaches, our models allow simultaneous high-dimensional variable selection and flexible estimation of the different intrinsic biological relationships among high-throughput platforms. We exemplify our approaches using both synthetic and pan-cancer datasets and show how integrative methods have higher power to detect disease related markers than non-integrative methods.