Fig. 2: Effect of signature choice on deconvolution accuracy.
From: Comprehensive evaluation of deconvolution methods for human brain gene expression
A Scatterplots of true and CIBERSORT-estimated proportions in VL-derived in silico mixtures, for nine different signatures. Matched: the signature and mixture were derived from the same dataset. Plot titles represent the signature used in deconvolution, as follows: VL Velmeshev, NG Nagy, CA Human Cell Atlas, LK Lake, TS Tasic, F5 FANTOM5, IP immuno-paned, MM mouse immuno-panned, DM darmanis. See the “Methods” section for further details about signatures. Dotted line: y = x. B Barplots of Pearson correlation (r) for all cell-types and signatures presented in (A). Dotted line: r = 0.8. C Barplots of normalised mean absolute error (nmae) for all cell-types and signatures presented in (A). OPC oligodendrocyte precursor cells. Dotted line: nmae = 1.
