Finally, whereas the analysis presented here relies on decomposing the variance into two terms, a genetic relatedness component and a component representing residual effects, future studies may need to account for additional variance components to more precisely model the heterogeneous phenotypic variance. In expression quantitative trait loci mapping, for example, one may want to add additional variance components to account for technical bias44. When multiple variance components are involved, one would need to make use of algorithms such as PROC MIXED implemented in SAS, as EMMA is developed for two variance components only; this would increase the running time of the first step of our procedure. However, because the same variance components estimated from the null hypothesis would be used across the genome-wide markers, the overall computational time should still be acceptable.
