We predicted cell-type proportions of the MetaBrain Cortex-EUR and -AFR datasets using the method and single-cell profiles previously published by the PsychENCODE consortium4 (Supplementary Note). We decided to discard the developmental cell types as we expected that these cell types are very rare or not present in adult human brain and because their signatures were obtained from fetal cells. The remaining cell types included all major cell types in the brain: neurons (excitatory, inhibitory and other), oligodendrocytes, astrocytes, microglia and endothelial cells. We then predicted the cell-type proportions as previously described4 (Supplementary Note). However, to enable the joint analysis of samples, we chose to correct the log2-transformed transcript-per-million gene counts for 20 RNA-seq quality metrics using OLS as we observed that this removed dataset biases in the predictions. To maintain the information captured by relative expression differences between genes required for deconvolution, we rescaled the residuals to the original log2-transformed mean and standard deviation, and replaced negative values with zero. For the deconvolution step, we used the non-negative least squares95 implementation in SciPy (v1.4.1)96. Given that the average proportions of
