Some Old and New Tests of Disease Association with Multiple SNPs in Linkage Disequilibrium

Wei Pan
Division of Biostatistics
University of Minnesota

Wednesday, December 3rd
3:30pm
MoosT 5-125
Minneapolis Campus

Abstract:
Genome-wide association studies have become popular in detecting genetic variants associated with complex diseases. Because of often weak association strengths, it is critical to use statistical tests with high power. For the typical case-control design, we consider testing disease association with multiple SNPs in a candidate region or gene. The statistical problem in terms of its formulation is simple: we are testing on multiple parameters in one or more logistic regression models. Two most popular existing approaches are 1) to test individual or SNP-specific parameters separately in marginal/univariate regression models (with multiple test adjustment), and 2) to test multiple parameters simultaneously in a joint regression model; the parameter estimates are all (approximately) normally distributed. Two alternative approaches are discussed: the first is a compromise of the above univariate and multivariate approaches, which works well under some situations but not in others; the second is a "fix" of the first. To me, the most striking discovery from this reasearch is my rediscovery of some known results in theoretical statistics textbooks. For example, the ubiquitous use of the likelihood ratio, Wald or score test in multiple regression may not be optimal, in contrary to their perceived "optimality"; in particular, in contrast to the use of the covariance matrix in the Wald test, an alternative is to ignore the correlations and just use its diagonal elements.

A social tea will be held at 3:00 P.M. in A434 Mayo. All are Welcome.
For more details contact 612-624-4655 or see http://www.biostat.umn.edu/seminar_academic.html