Fig. 4

RMSE (left) and its mean ranks (right) are shown for each classifier/deconvolution method and for the Leukocyte 1 (A–B), 2 (C–D), and 3 (E–F) data sets. Methods are ranking from top-down regarding the best performance (lowest RMSE). For the RMSE mean rank plots, the best method, and its 95% confidence interval (Friedman and Nemenyi post hoc test) is highlighted in gray. Methods whose average RMSE does not overlap at all with the highlighted area, are significantly worse than the top performing method. CimpleG.10 is based on 10 CpGs per cell type, while ENMix.min and IDOLmin are versions of IDOL and ENMix using the lowest possible number of CpGs