in the UW and Merck studies, respectively (Figure 1B). At more stringent thresholds, validation rates exceeded 80%, albeit with fewer included genes (Figure 1B). We also note that the sex-associated gene set was strongly enriched for genes on the X and Y chromosomes (Figure 1C; X chromosome, hypergeometric test, p-value = 1.72×10−14, Y chromosome, p-value<2×10−16), as would be expected for genes with sex-associated expression levels. Effects due to age were less reproducible: 13.2% and 21.5% of significantly age associated genes (UC age t-test FDR<5%) replicated in the UW and Merck studies, respectively (Figure 1E; an example of a replicated age-associated gene, TMEM22, is displayed in Figure 1F). Effect sizes for both sex and age were correlated across studies (Figure S4; Spearman's rho, UC-UW sex = 0.597, UC-Merck sex = 0.720, UC-UW age = 0.333, UC-Merck age = 0.159), underscoring the reproducibility of demographic effect estimates.
