scDHMap can be used for trajectory interpretation and denoising counts. (A) Embedding of scDHMap on the simulated data set with three branches (M1–M3, M3–M2, and M3–M4). Dot shapes represent branches; red, blue, and green colors from shallow to deep represent ground-truth pseudotime. Ten data sets had been generated, and one example is displayed. (B) Spearman's correlation coefficient between the Poincaré pseudotime inferred by scDHMap, PoincaréMap, and the ground-truth pseudotime in the three branches. (C) Imputation errors of scDHMap pretraining, scDHMap final training, and deep count autoencoder (DCA) on the simulated data sets. (D) Area under the curve (AUC) plots of trajectory differential expression (DE) of raw counts and scDHMap-denoised counts (final training). (E) One DE gene in the branch M3–M2; plots display raw counts and scDHMap-denoised counts against the ground-truth pseudotime. Trend lines are smoothed by the LOESS regression.
