Table 1.
Detailed information on the eight scRNA-seq datasets used to compare the performance of imputation methods.
Fig 1.
Evaluation of expression data recovery of G2S3 by down-sampling.
Performance of imputation methods measured by correlation with reference data from the first category of datasets, using gene-wise (top) and cell-wise (bottom) correlation. Box plots show the median (center line), interquartile range (hinges), and 1.5 times the interquartile (whiskers).
Fig 2.
Evaluation of G2S3 in improving cell subtype separation.
Average inter/intra-subtype distance ratio (top) and silhouette coefficient (bottom) to demonstrate cell subtype separation using the top principal components of the raw unimputed and imputed data by each method in the Chu dataset.
Fig 3.
Plots showing 2D-Visualization of the Chu dataset.
UMAP plots of the raw unimputed and imputed data by all methods. Cells are colored by true cell subtype labels. The normalized mutual information (MI) and adjusted rand index (RI) are calculated to measure the consistency between cell clustering results and true cell subtype labels.
Fig 4.
Visualization of cell trajectories in the raw and imputed data by all methods.
Cells are projected into two-dimensional space using reversed graph embedding. Pseudotemporal ordering score (POS) and Kendall rank correlation coefficient (Cor) are used to measure the consistency between the actual embryonic days and the reconstructed pseudo-time.
Fig 5.
Receiver operating characteristic (ROC) curves demonstrating improvement in differential expression analysis.
ROC curves measuring the prediction accuracy in scRNA-seq data on differentially expressed genes identified in bulk RNA-seq data from the same samples in the Chu (A) and Trapnell (B) datasets.
Fig 6.
Performance of G2S3 in recovering gene regulatory relationships.
Boxplots showing the area under the receiver operating characteristic curve (AUROC) ratios that measure the accuracy of inferred GRNs using the imputed data by different imputation methods. PIDC, GENIE3, GRNBoost2 and PPCOR are used to infer GRNs. Red line indicates the performance of a random predictor.
Table 2.
Fraction of periodic gene pairs with correct direction of correlation in the raw and imputed data by each method.
Fig 7.
Summary of performance of G2S3 and other imputation methods.
A heatmap demonstrating method performance based on the five evaluation criteria. The left five columns display performance rank using each of the five evaluation criteria. The rightmost column displays the overall performance rank based on the sum of the five ranks.