considered as covariates. We did principal component analysis (PCA) to extract the first three PCs for both technical (batch, RIN, and PMI) and biological (sex, age, brain weight, brain pH, left-right brain, smoking, and liver disease) confounding variables, and the obtained PC1, PC2, and PC3 were used as covariates in the model matrix design for differential expression analysis, as described in a recent article [36]. For microarray expression data from Set 2 brain tissue samples, the differential expression analysis was performed in the same way using the lmfit function in the limma package [35].
