By using regression analysis to adjust the gene scores for the confounding effects we identified, we provide a viable approach to determine gene association p-values in the absence of genotype data, which should prove useful for mining large GWA study meta-analyses or other types of GWA studies where only variant association statistics are available. Using the Diabetes Genetics Initiative (DGI) study, we showed that our correction accounts for most of the confounding effects on the most significant SNP score and yields gene scores that are much more accurate than those obtained without correction [31]. Notably, this regression approach and the DGI permutation system can be used to identify and adjust for confounders on other types of SNP to gene scores (e.g. considering best SNP per LD block [25], [33] or the set–based test in PLINK http://pngu.mgh.harvard.edu/~purcell/plink/anal.shtml#set). While in the current work we focus largely on developing a gene set approach following gene score correction, we envisage that the corrected gene p-values might be valuable in future gene-centric studies, allowing one to properly weigh specific genes (e.g. small genes) that may otherwise be missed.
