Fig. 2: scReadSim outperforms existing scRNA-seq and scATAC-seq read simulators.
From: scReadSim: a single-cell RNA-seq and ATAC-seq read simulator
a scReadSim outperforms the existing scRNA-seq read simulator minnow in preserving the read coverage in a mouse 10x single-cell Multiome dataset (the RNA-seq modality only)21. With chromosome 1 of the reference genome divided into consecutive, non-overlapping 1000-bp windows, the x-axis represents the windows' positions along chromosome 1, and the y-axis indicates the number of the reads overlapping each window in each pseudo-bulk sample (real or synthetic, with all cells pooled). The track height is set to 6,595,507 for all three tracks. The Spearman correlation (Cor.) and Pearson Cor. measure the similarity of read coverage between real and synthetic data across the windows. Inset: a closer view of the region chr1:86,425,858–86,443,378 using the IGV genome browser46. Each coverage track displays the depth of the reads covered at each locus and indicates its track height 136 or 438 at the left corner. b In UMAP visualizations, scReadSim outperforms the existing scATAC-seq read simulator SCAN-ATAC-Sim in mimicking real cells in the mouse 10x single-cell Multiome dataset (the ATAC-seq modality only)21. The miLISI measures the similarity of real and synthetic cells in the UMAP space: the miLISI value ranges between 1 and 2, with 2 indicating a perfect mixing of real and synthetic cells. c scReadSim converts gray areas in real data to ground-truth non-peaks in synthetic data, maintaining trustworthy (t.w.) peaks and non-peaks in real data as ground-truth peaks and non-peaks, respectively, in synthetic data.
