Title: "Locating quantitative trait loci with a modified version of BIC"
Speaker: Malgorzata Bogdan, Institute of Mathematics and Computer Science Wroclaw University of Technology, Poland
Place: Krannert (KRAN) G016; November 7, 2006, Tuesday, 4:30pm

Abstract

Locating genes responsible for quantitative traits usually relies on fitting the multiple regression model with explanatory variables defined by marker genotypes. Simulation studies demonstrate that in this genetic context most of the standard model selection criteria have a tendency to overestimate the number of regressors. In particular Bogdan et al. (2004) investigate the behavior of the well known Bayesian Information Criterion (BIC) and explain the phenomenon of the overestimation using arguments of Bayesian statistics. Bogdan et al. (2004) also propose a suitable modification of BIC (mBIC), which handles the problem of overestimation. In this talk we point at the relationship between the problem of the overestimation of the number of regressors and the classical problem of multiple testing. We also present a new, iterative version of mBIC proposed in Baierl et al (2006), as well as some new results on adjusting of mBIC to very dense markers and multiple interval mappin g (Bogdan et al. 2006). We illustrate the performance of our method with the analysis of the real data set on the morphological differences between two species of Drosophila, previously analyzed by Zeng et al. (2000).




Click here for a full schedule of BIOINFORMATICS SEMINARS, past and present.