Bayesian Inference of Posterior Error Probabilities for Disease Mutation Association

Abstract

Associations between single-nucleotide polymorphisms (SNPs) and phenotype are an important research paradigm in genomics and have been extensively studied, especially for the purpose of bettering our understanding of he emergence and development of disease. Various methodologies accommodating different data types, data qualities and genomic properties have been developed. In this work, we focus on a specific aspect of the analysis, namely the computation of an association score using a statistical procedure. We argue that Bayesian inference using a mixture prior that is learned from the data leads to better predictions than existing approaches, and suggest its incorporation into existing tools. To demonstrate its power, we implement the new procedure and compare its performance to popular tools on simulated data. We also detect associations between disease and SNPs using a real, modest-size RNA-Sequencing dataset, showing that the method can produce useful insights and has broad applicability.