We evaluated the extent to which eQTLs are dependent on cell-type proportions by determining cell-type interaction eQTLs (ieQTLs)3,18,19. In the Cortex-EUR subset, we predicted cell-type proportions using single-cell RNA-seq-derived signature profiles4 (Supplementary Note and Supplementary Fig. 16). The cell-type proportions and reconstruction accuracy of our predictions (87%) were comparable to a previous study that used this reference profile on a subset of the Cortex-EUR samples4. We observed low-to-moderate correlations between predicted cell types (0.01 < Pearson’s correlation coefficient (r) < 0.55; Fig. 4a) and high positive correlations with immunohistochemistry (IHC) counts from the ROSMAP cohort20 (overall Pearson’s r = 0.89 and per cell-type Pearson’s r > 0.1; Fig. 4b). However, we note that the exact proportion for each cell type remains uncertain21,22.
