We then loaded aggregate (mean per cluster) data for each cell type and selected the genes as described above for dendrogram construction analysis. We then intersected the selected genes from aggregate data and quality filter on energy voxel data. We calculated the correlation between each voxel and each cell-type, where voxel data was normalized as above and the aggregate data was normalized in a similar way ((X-m)/s) after log2+1 transform. Finally, we calculated the regional fold enrichment: for each cell-types take the top 100 pixels (across the whole brain) and calculate the fold-enrichment of the anatomical region IDs that are among them by normalizing to frequency within the 100 to the overall frequency of each region ID.
