We replicated these findings in the Cortex-AFR dataset (n = 319) as well as in two independent single-nucleus RNA-seq (snRNA-seq) datasets from ROSMAP23 (n = 39) and Bryois et al.24 (n = 196; Supplementary Figs. 17–19 and Supplementary Table 9). Across all replication datasets, we observed moderate-to-high rates of agreement, depending on the cell-type frequency and sample size (Bryois et al.24: 0.78 <Rb < 0.86, median Rb = 0.84, 0.43 < π1 < 0.83, median π1 = 0.69, 81% <AC < 90%, median AC = 0.9; Supplementary Note). Examples of replicating ieQTLs include the oligodendrocyte ieQTL genes FAM221A, NKAIN1 and STMN4, which were previously identified as oligodendrocyte-specific25, and AMPD3 and CD82, of which the SNPs were previously associated with white-matter microstructure26, suggesting a role for oligodendrocytes (Fig. 4d,e and Supplementary Fig. 20a–e). The high replication rates indicate that our approach can accurately identify the cell type for a large number of eQTLs. We note that summary statistics were available for only 54% of ieQTLs in a well-powered replication dataset (Bryois et al.24), suggesting that our approach had the power to detect ieQTLs that are not yet identified in snRNA-seq datasets.
