|
Abstract:
|
Some existing approaches to measure gene-gene interaction such as the case-only approach make a crucial assumption of gene-gene independence for physically distant genes. While this strategy is known to increase power considerably, it is prone to significant bias when the independence assumption is violated. To solve this problem, we use the idea of "genetic matching," which has been proposed recently to control for stratification in a different context. Cases and controls are matched based on ancestry inferred from genome-wide null markers. Matched sets are then analyzed using extensions of conditional logistic regression that derive additional power from the gene-gene independence assumption. We compare our approach to some of the existing methods in terms of bias and efficiency, both using a GWAS data on prostate cancer from the CGEMS study and simulations.
|
