together addressed several challenges in analyzing cell type differential expression across multiple axes of meaningful biological variation: 1) accounting for correlated pseudo-bulk samples shared across individuals with the duplicateCorrelation (block = Subject.ID), 2) estimating quality weights for adjusting for cell type proportions with the function voomWithQualityWeights, and 3) calculating un-modeled variation in high-dimensional single cell data with the sva()function. To plot the gene expression profiles, we used the normalized counts per million (CPM) of each pseudobulk profile corrected for the batch effects and surrogate variables unrelated to the OUD diagnosis using the cleaning function (https://github.com/LieberInstitute/jaffelab). For heatmap visualizations of the gene expression profiles, we also z-normalized the corrected expression profile of each gene grouped by cell type, since gene expression is highly cell-type specific.
