A Two-Step Modeling Strategy for Testing and Estimating Genetic Susceptibility to the Ill-Effects of Adiposity: Illustration in an Outbred F2 Mice Population

• Activity Number: 566
• Type: Topic Contributed
• Date/Time: Wednesday, August 3, 2011 : 2:00 PM to 3:50 PM
• Sponsor: ENAR
• Author(s): Guo-Bo Chen*+ and Katherine H. Ingram and Gustavo de los Campos and Nengjun Yi and Xiang-Yang Lou and Daniel Pomp and David B. Allison
• Companies: University of Alabama at Birmingham and University of Alabama at Birmingham and University of Alabama at Birmingham and University of Alabama at Birmingham and University of Alabama at Birmingham and The University of North Carolina and University of Alabama at Birmingham
• Keywords: Prediction ; Bayesian Lasso ; insulin resistance ; Linear regression ; Interaction ; Adiposity

Abstract

We explored the possibility of using genome-wide data to achieve the types of predictions that would be useful in personalized medicine. We examined an outbred F2 mouse population derived from an intercross between the line M16 and the line ICR. Specifically, we asked "Can we identify mice for whom their degree of fatness will strongly influence their degree if insulin resistance and those for whom their degree of fatness will be unrelated to their degree if insulin resistance". To illustrate how the variation in our phenotype (Y) in question, insulin sensitivity, is determined by body composition (P) and genes (G) as well as their interaction, a two-step statistical model was proposed. In step one, the genetic values associated with a joint measure of P and Y are inferred by regressing phenotypes on all markers concurrently, with their effects estimated using the Bayesian Lasso. In step two, Y is regressed on the predicted genetic values (from step one) and percent fat and their interaction (B). The p-value for the interaction effect is most significant (p< 0.001). These results underscore the importance of both body composition and genetic influence on insulin resistance.