Identification of sex-differential gene expression

The analysis pipeline for sex-differential gene expression is deposited under doi:10.5281/zenodo.3939042. 

1. For each tissue, we selected genes to be tested for sex-differential gene expression by applying the following filters:
   1. Gene read counts were normalized between samples using TMM.
   2. Genes were selected based on expression thresholds of ≥0.1 transcripts per million (TPM) in ≥20% of samples and ≥6 reads (unnormalized) in ≥20% of samples.
   3. Expression values for each gene were kept in TPM units.
   4. Y-linked and mitochondrial genes were excluded.
2. For each tissue, voom-limma was used to compare log2-cpm gene expression values between male and female samples while adjusting for measured - at a subject and sample level - covariates including age, ischemic time, RNA Integrity Number (RIN), and  surrogate variables (SVs) defined by smartSVA. The SVs were estimated from log2-cpm gene expression values and parametrized to exclude gene expression variation linearly attributable to sex or to the measured covariates.
3. For all analyzed genes, we provided the voom-limma-quantified sex effect size and corresponding standard error matrices to MASH to model differential expression jointly across tissues. Internally, MASH filled missing values - due to genes excluded for low expression in a particular tissue - with sex effect size equal to 0 and corresponding standard error equal to 10.
4. Significant sex-differentially expressed genes were defined as LFSR ≤ 0.05, excluding genes whose effects overlap 0 at a 95% confidence interval. Positive log2 fold changes are classified as female-biased genes, whereas negative log2 fold change genes were classified as male-biased genes.

The full set of MASH summary statistics are available on the GTEx portal https://www.gtexportal.org/home/. Details of the MASH pipeline are available at doi:10.5281/zenodo.3939042.