Having identified six potential gene score confounders, we used step-wise multivariate linear regression to remove these confounding effects from , to generate a corrected gene score, (see Materials and Methods and Figure 1C). In this analysis the confounders are removed sequentially, accounting for the correlations between the various gene properties. Aside from the genetic distance, all five remaining properties listed in Table 1 were significant under the step-wise linear regression model (p<0.05) applied to in either about half or all of the 1,000 permuted DGI data sets (Table S3). As a result the first five properties listed in Table 1 were used for subsequent analyses (see Table S4 for model coefficients and p-values for the DGI study and the DIAGRAM+ T2D meta-analysis).
