Fig. 5
From: A deep generative model for multi-view profiling of single-cell RNA-seq and ATAC-seq data
scMVP facilitates trajectory inference with joint embedding latent features. a UMAP visualization of SNARE-seq mouse cerebral cortex P0 dataset for 1469 cells (214 IP–Hmgn2, 99 IP–Gadd45g, 437 IP–Eomes, 177 Ex-L2/3–Cntn2, and 542 Ex-L2/3–Cux1) with scMVP with two omics, scRNA only, scATAC only, and Monocle3 with scRNA and scATAC. The ARI scores for clustering accuracy with each embedding were labelled in the subtitle of subplots. Ex excitatory neurons, IP intermediate progenitors. b Pseudotime trajectories constructed with scMVP joint embedding in Fig. 5a. Cells are colored according to pseudotime score (top) or cellular identity (bottom). c UMAP visualization of 4619 cells (1164 αhigh CD34+ bulge, 1495 αlow CD34+ bulge, 537 Isthmus, 466 K6+ Bulge Companion Layer and 957 ORS) with scMVP with two omics, scRNA only, scATAC only, and Monocle3 with scRNA and scATAC. The ARI scores for clustering accuracy with each embedding were labelled in the subtitle of subplots. ORS outer root sheath. d Pseudotime diffusion map constructed with scMVP joint embedding in Fig. 5c. Cells are colored according to pseudotime score (top) or cellular identity (bottom). e The cell types shift during bulge cell development, referenced from SHARE-seq paper