There are many ways of performing model selection in a multiple regression framework. We use the following stepwise selection strategy to select the associated SNPs iteratively over all the SNPs across the whole genome, regardless of their P values from the meta-analysis, except for the most significant SNP, which was used for model initiation. Start with a model with the most significant SNP in the single-SNP meta-analysis across the whole genome with P value below a cutoff P value, such as 5 × 10–8.For the tth step, calculate the P values of all the remaining SNPs conditional on the SNP(s) that have already been selected in the model. To avoid problems due to colinearity, if the squared multiple correlation between a SNP to be tested and the selected SNP(s) is larger than a cutoff value, such as 0.9, the conditional P value for that SNP will be set to 1.Select the SNP with minimum conditional P value that is lower than the cutoff P value. However, if adding the new SNP causes new colinearity problems between any of the selected
