Fig. 3

DrImpute significantly improved the performance of PCA and t-SNE in visualizing scRNA-seq data. a The barplots of average accuracy of separating the cell subpopulations in 2D space. For Zheng and Hrvatin datasets, 1000 cells were randomly sampled from the full datasets and used for the clustering analysis for each method. Black interval represents one plus or minus standard error of the category. Wilcoxon rank sum test was utilized to compare the accuracy from different tools (***p value <0.001). b Visualization of four groups of mouse neural single cells (NP, TH, PEP, and NF) using PCA. Left and right panels, respectively, show the 2D visualization of single cells without and with preprocessing the scRNA-seq data using DrImpute. c Visualization of mouse preimplantation embryo using t-SNE. Left and right panels, respectively, show the 2D visualization of single cells without and with preprocessing the scRNA-seq data using DrImpute. The classification accuracy was computed by using the 2D coordinates of each dimension reduction results