Seminar at SMU Delhi
March 14, 2012 (Wednesday) ,
3:30 PM at Webinar
National Institute of Biomedical Genomics
Statistical Issues in Detecting Interactions in Genetic Association Studies
Abstract of Talk
In a genetic study of a disease, we are often interested in interaction between genes or that between
genes and environmental factors such as smoking. The case-control design and logistic regression are
traditionally used to measure marginal effects in genetic association studies. Recently, a case-only
design has been shown to be much powerful to detect interactions under the assumption of gene-
gene (G-G) or gene-environment (G-E) independence. However, we show that such an assumption
can be violated due to hidden population substructure and lead to large-scale bias in ``case-only''
methods. We discuss principal component based genome-wide methods to correct this bias using
regression or likelihood-based frameworks. These approaches allow the robust incorporation of G-G
and G-E independence assumptions for either a case-only or case-control design. Some analytical and
simulation-based comparisons with standard methods will be shown.