We have shown above that our results were not driven by pleiotropy and reverse causality. In some cases, the relationship between a risk factor and a disease could be a mixture of multiple models. For example, we have shown above that BMI had a risk effect on T2D, which has been confirmed by RCT35, that T2D had a significant reverse effect on BMI and effect size was negative, and that there was a SNP (at the TCF7L2 gene locus) that appeared to have pleiotropic effects on T2D and BMI (Fig. 3), a mixture model of causality, reverse causality and pleiotropy. In addition, we demonstrated by the conditional GSMR analyses that the mediation effects (i.e., the effect size of a risk factor on disease mediated or driven by other risk factors) are apparently small for most risk factors except for HDL-c (Fig. 5; Supplementary Table 11).
