Fig. 4

Accuracy assessment with marker gene expression. econvoluted cell-type fractions compared with their marker gene expression from 634 samples of the ROSMAP bulk RNA-seq data. In the left panel, cell-type fraction estimates from seven bulk RNA-seq deconvolution runs are plotted against cell marker expression from the same samples. The cell-type markers are GFAP for astrocytes (AST), PECAM-1 for endothelial cells (END), IBA1 for microglia (MIC), NeuN for neurons (NEU), and Olig2 for oligodendrocytes (ODC). Each column shows a different deconvolution method: MuSiC, MuSiC with its “C” and “N” parameters, SCDC, CIBERSORTx, and DeTREM. Cell-type marker expression is scaled linearly from zero to one. NEU estimates were obtained by summing glutamatergic and GABAergic neuron percentages. A linear model trend line is shown for each plot. The right panel shows Pearson correlations between cell type percentage estimates and marker gene expression for non-zero estimates. The proportion of non-zero estimates are indicated in parentheses. These values were averaged for each method and shown in the bottom row