Fig. 4

Capturing cell-composition in the presence of simulated technical noise. We added simulated batch effects to the whole-blood datasets of Liu et al. [39] and Hannum et al. [38] and compared the ability of CONFINED, ReFACTor[24], PEER[41], PMA[36], and NNMF to capture cell-type composition in whole-blood. Here, we show the results of the Hannum et al. dataset; however, the results of each method were quantitatively similar across both datasets (Additional file 1: Figure S12)