We defined the data-driven covariance matrices as the top four principal components from the PCA performed on the ‘strong’ signals. For gene-level analysis, we defined a set of ‘strong’ tests running a simple condition-by-condition (mash_1by1) analysis as described in ‘Global ancestry-associated differential expression analysis’. For the isoform-level analysis (that is, transcripts, exons and junctions), we defined a set of ‘strong’ tests using either the results from permutation or the eigenMT analyses. Specifically, for the main effect analysis, the set of ‘strong’ tests was selected if a feature–variant pair was present in at least one brain region within the permutation or conditional analyses. For the interaction analysis, we selected the set of ‘strong’ tests if a feature–variant pair was present in at least one brain region from the eigenMT top associations.
