Next, we attempted to recalibrate the observed gene expression variation by considering fQTLs. In particular, our scheme described above for approximately deconvolving gene expression from heterogeneous bulk tissue (matrix B) into single-cell signatures (matrix C) and estimated cell fractions (matrix W) enables us to calculate the residual gene expression (Δ) remaining after accounting for cell fraction changes (Fig. 2). Specifically, it is the component of the bulk tissue expression variation that cannot be explained by the changing cell fractions alone: Δ = B − CW. We can subsequently use this quantity to determine “residual QTLs” by directly correlating it with genotype. In total, this results in 202,940 SNPs involved in residual eQTLs. Potentially, one can elaborate on this further by allowing the correlations to be done in a cell-type–specific fashion (fig. S35).
