Fig. 1.

Purifying cell populations from single-cell datasets using scGate. (A) Uniform Manifold Approximation and Projection (UMAP) representation of scRNA-seq data of PBMC populations annotated by Hao et al. (2021) (B) Purification of target cell types using scGate, for B cells on the left (using marker MS4A1 [encoding CD20]) and NK on the right (using NCAM [encoding CD56] and KLRD1 as positive markers, and CD3D as a negative marker). The violin plots display normalized ADT counts for the indicated proteins on the same cells. Precision (PREC), recall (REC) and MCC are shown. (C) UMAP representation of scRNA-seq data of melanoma tumors annotated by Jerby-Arnon et al. (2018) (D) Purification of macrophages using a hierarchical GM: immune cells at the first level (left panel) and macrophages at the second level (middle panel). Macrophage gene signature (UCell) scores are shown in the right panel. (E) scGate purification of monocytes using DNA accessibility of a PBMC 10× multiomics dataset. Violin plots display coupled RNA expression values. Gene-associated accessibility values were inferred using Signac (Stuart et al., 2021). (F) PREC (Positive Predictive Value) and MCC values for five publicly available scRNA-seq datasets (derived from blood or tumors) for scGate and three other cell type classifiers