As a comparison, we also applied another two methods, smallest P-value and PLINK set-based tests [40], to the two GAIN data sets. The smallest P-value method, which only used the SNP with the smallest P-value to represent a gene, detected 6 and 2 significant pathways in EA and AA data sets, respectively; only one was shared by both data sets. This showed our method could improve the power of detecting causal pathways by using multiple SNPs to represent a gene. For the set-based test method (with parameters --set-p 0.05, --set-r2 0.5, the same as that in our method), 3 significant pathways were detected in EA, and 2 in AA, but none was shared between the two data sets. One reason for the loss of power of these two methods might be their favouring of pathways containing large numbers of genes and genes with large number of SNPs, as larger pathways are expected to show more significant genes or SNPs just by chance. We checked several potential factors that might affect pathway significance: pathway size, gene size, total bp content, and
