To assess sharing across brain regions and to increase our power to detect main and interacting eQTL effects within admixed BA-only individuals, we used the multivariate adaptive shrinkage framework as described previously12. We extracted the effect sizes and s.e. for these effect sizes from the nominal results for either the main or interacting cis-eQTL. To specify a correlation structure across brain regions (that is, overlapping sample donors), we used the estimate_null_correlation_simple function before fitting the mash model. The mash model included both the canonical covariance matrices and the data-driven covariance matrices learned from our data.
