Fig. 2

a Scatterplots between gene means and zero proportions across genes calculated from raw UMI data, clustered data, and data after sequencing-depth normalization, respectively. Fitted line is negative binomial curve. b, c Evaluations of pre-processing in Sctransform. b Distributions of sequencing depths across cells in raw UMI data vs. data cleaned by SCtransform. c Comparisons of three monocyte markers in raw UMI data vs. data cleaned by SCtransform. d, e Evaluations of pre-processing in DCA. d Log fold changes and log p values from differential expression analysis using the same data set but imputed by DCA as heterogeneous and homogeneous cell populations, respectively. e The p values comparisons between two different imputation strategies show the general deflation of biological signals of DCA when applied to heterogeneous cells