Bayesian Polynomial Regression Models to Fit Multiple Genetic Models for Quantitative Traits

Abstract: Typically in a genetic association study, the association between each single nucleotide polymorphism (SNP) and a quantitative trait is tested using linear regression under a specific genetic model that can assume a genotypic (2 degrees of freedom), dominant, recessive, co-dominant, or additive mode of inheritance of each tested SNP. However, the inheritance pattern is rarely known, and using a suboptimal model can lead to a loss of power. In this work, I present a coherent Bayesian framework for selection of the most likely model from the five genetic models (genotype, additive, dominant, co-dominant, and recessive). This approach uses a polynomial parameterization of genetic data to simultaneously fit the five models and save computations. I show that the five genetic models are special cases of the proposed polynomial parameterization. There is a convenient transformation between the polynomial model (2 degrees of freedom model) and other four specific genetic models (1 degree of freedom model) by utilizing the fact that the parameters in the four specific models are constrained by a linear contrast of the parameters in the polynomial model. Given that all genetic models are equally likely a prior, I propose to use marginal likelihood to select the most likely genetic model. I provide a closed-form expression of the marginal likelihood for normally distributed data, and evaluate the performance of the proposed method and existing method through simulated and real genome-wide data sets. Polynomial parameterization of different genetic models provides a coherent theoretical framework within which the most likely model is chosen based on a Bayesian model selection approach using marginal likelihood.

Harold Bae graduated from Dartmouth College with a Bachelor of Arts in applied mathematics in 2005. Then, he receivved a Master of Science in Health Services Research from The Dartmouth Institute for Health Policy and Clinical Practice, in which he was first exposed to the world of Biostatistics. After his masterʻs training, he joined the Veterans Affairs Outcomes Research Group based in White River Junction, VT as a project coordinator. He started his doctoral training in Biostatistics at Boston University in 2009. His research areas of interest include statistical genetics and Bayesian statistics, and his applied research has focused on sickle cell disease, longevity and aging.

This is an event. It was posted Apr 2, 2014 at 2:00pm and last updated Apr 2, 2014 at 2:00pm.