Non-redundant summary scores applied to the North American Rheumatoid Arthritis Consortium dataset

After performing a genome-wide association study, it is often difficult to know which regions to follow up, especially when no one marker reaches genome-wide significance. Researchers frequently focus on their top N findings, knowing that true associations may be buried deeper in the list. Others focus on genes or regions that have multiple markers showing evidence of association. However, these markers are often in high linkage disequilibrium with one another (r2 > 0.80), which indicates that these additional markers are providing redundant information. I propose a novel method that identifies regions with multiple lines of evidence, by down-weighting the contribution of additional markers in proportion to pairwise linkage disequilibrium. I have used this non-redundant summary score in my analysis of the North American Rheumatoid Arthritis Consortium dataset released as part of Genetic Analysis Workshop 16. Three regions were identified that had a genome-wide empirical p-value less than 0.01, including one novel region on chromosome 20 near the KCNB1 and PTGIS genes.