Bayesian Nonlinear EM Based Approach for Analysis of Multi-Platform Genomics Data

• Activity Number: 48 - Academics Industry Perspectives on Cancer Data Innovations: Simultaneous Inference, Inconsistency, and Clinical Response
• Type: Topic-Contributed
• Date/Time: Sunday, August 8, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Biometrics Section
• Author(s): sounak chakraborty*
• Companies: University of Missouri-Columbia
• Keywords: epigenetics; EM Algorithm; data integration; multi-platform data; non-linear effect; gene pathway

Abstract

More recently different platforms have been brought together on the same patient set. Each output from different platforms could provide a different and complementary view of the whole genome. Thus, borrowing information across platforms has become more crucial. Therefore, in a given multi-platform genomic data, identifying important genes that have significant association with the clinical outcome through integration of mRNA expression level and other output of different types of platforms is of the greatest interest. In this paper we focus on building a nonlinear model that can simultaneously incorporate the multi-platform information and identify significant linear and non-linear gene effects associated with the clinical outcome. For gene-selection from a large set we consider the EMVS methodology and Bayesian LASSO with modified NEG (Normal-Exponential-Gamma) prior, both of these method provide solution for large p scenarios. In addition to that each of these methods solve different issues as well, for the EMVS, it reduces the computation time drastically, and for the Bayesian LASSO with modified NEG prior, it is adapted to incorporate genetic grouping information.