This week's weekly Biostatistics Seminar will feature Dr. Daniel J. Schaid, Professor of Biostatistics, Health Sciences Research and Medical Genetics, Mayo Clinic, speaking on Genetic analysis of multiple correlated traits (genetic pleiotropy).
Statistical Methods for Genetic Pleiotropy: Sequential Multivariate Tests to Determine
Which Traits are Associated
The statistical association of a single trait with genetic data has revolutionized human genetics, with many genome-wide association studies providing guidance on genetic factors influencing human health. Pleiotropy – the association of more than one trait with a genetic marker – is believed to be common, yet current multivariate methods do not formally test pleiotropy. Current multivariate methods, such as multivariate regression of multiple traits on a genetic marker, or reverse regression of a genetic marker on multiple traits, test the null hypothesis that no traits are associated with a genetic marker; a statistically significant finding could result from only one trait driving the association. We developed a new formal test of pleiotropy, so that so that rejection of the null hypothesis implies at least two traits are associated with the marker. We further refined our approach to sequentially test the number of associated traits, in order to identify which traits are statistically associated, while accounting for the correlation among the traits. The new methods, with simulations illustrating it properties, will be presented, as well as application to a study of the genetics of response to small pox vaccination. The proposed sequential testing is not limited to genetic data – it can be used for any setting that attempts to evaluate which traits are associated with a predictor variable in a regression setting.