Benchmarking result on eight mixture of cell line scRNA data. (A) The imputation consistency measured by RMSE and Pearson correlation (Cor.). (B) UMAP plot. Cells from the same group share the same annotation color. From left to right are datasets p3cl, sc_10x_5cl, sc_10x, sc_celseq2_5cl_p1, sc_celseq2_5cl_p2, sc_celseq2_5cl_p3, sc_celseq2 and sc_dropseq with the dropout rate of 90%, 80%, 80%, 65%, 65%, 75%, 65% and 75%, respectively. From top and bottom are tools: Raw, Dropout, I-Impute, SAVER, MAGIC, kNN-smoothing, SMURF_CV, SMURF_F and SMURF_V. (C) The clustering accuracy measured by adjusted rand index (ARI), adjusted mutual information (AMI), normalized mutual information (NMI), homogeneity score (Homogeneity), completeness score (Completeness) and V-measure score (V-measure) of HVG600. SMURF_CV, SMURF_V and SMURF_F stand for SMURF with variance mode constant coefficient of variance (CV), constant Fano (F) and constant variance (V), respectively. SMURF models with suffix ‘H’ mean the clustering was conducted with the cell embedding yielded by SMURF.