Research Project
7102830
DESCRIPTION (provided by applicant) The overarching goal of this research proposal is to devise efficient and robust statistical methods for genetic dissection of complex human traits, which are determined by a complex interplay of gene-gene and gene-environment interactions. Our basic tenet is that for the dissection of the determinants of such traits, specifically to map the underlying genes, the study of the precursor variables that modulate an end-point trait is statistically more powerful than studying the end-point trait itself, which is usually dichotomized (affected/unaffected) by defining a threshold on the frequency distribution of a quantitative trait. The major aims of this research are: (i) to develop non-parametric methods including kernel smoothing and quantile based regression techniques for linkage and association mapping multivariate phenotypes (possibly comprising a mixture of quantitative and binary variables) using data on different types of relative-sets and also unrelated individuals. (ii) to compare the proposed distribution-free methods with existing distribution based methods through extensive computer simulations, (ii) to statistically assess the advantages of using SNP markers in haplotype blocks for QTL mapping, (iv) to develop user-friendly computer programs incorporating the methodologies, (v) to modify the proposed methods to incorporate inbreeding practiced in some populations and (vi) to apply the new methods to data on different types of complex traits/disorders in disparate ethnic populations. The major statistical thrust of this research will be on development of distribution-free gene-mapping methodologies for mixed (quantitative and binary) multivariate phenotypes in the presence of epistatic and gene-environment interactions. Our past studies on univariate phenotypes (Ghosh and Majumder, American Journal of Human Genetics, 2000, 66:1046-1061) have shown that this approach is efficient and robust, especially when distributional assumptions (such as, normality) and model assumptions (such as, dominance at the QTL) are not valid.
