We performed differential expression analysis using mash modeling in R. Initially, we determined the effect size and the s.e. of the effect size using limma-voom modeling as described previously12. Briefly, we filtered low-expressing genes using filterByExpr from edgeR (v.3.40.2)71,72 and normalized library size. Next, we applied voom normalization73 as a model of genetic ancestry adjusted for age and RNA quality (mitochondria mapping, gene assignment, genome mapping and rRNA mapping rates, and hidden variance using qSVA; equation (1)). After voom normalization, we fitted the model using eBayes and extracted out the effect size (log fold change) and s.e. of the effect size from the model (equation (2)) by brain region for each feature (gene, transcript, exon and junction):2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{S.E.}}=\frac{{\rm{\sigma }}}{\sqrt{n}}$$\end{document}S.E.=σn
