Fig. 3
From: A general and flexible method for signal extraction from single-cell RNA-seq data
Using ZINB-WaVE to gain novel biological insights. a–c Lineage inference on the OE data set. Slingshot minimum spanning tree on cell clusters: a PCA with endpoint supervision (marked as red nodes in the tree). b PCA with no endpoint supervision. c ZINB-WaVE with no endpoint supervision. a, b RSEC clustering (see Methods) on the first 50 PCs of the normalized counts led to 13 clusters; c the same procedure on 50 components of W from ZINB-WaVE led to six clusters: horizontal basal cells (HBC); tran- sitional HBC (DHBC1-2); immature sustentacular cells (iSUS); mature sustentacular cells (mSUS); globose basal cells (GBC); microvillous cells (MV); immediate neuronal precursors (INP1-3); immature olfactory neurons (iOSN); mature olfactory neurons (mOSN). mOSN, MV, and mSUS are the mature cell types and should be identified as the three lineage endpoints. d, e Discovery of rare cell types for the 10× Genomics 68k PBMCs data set. d Scatterplot of first two t-SNE dimensions obtained from 10 components of W from ZINB-WaVE; cells are color-coded by cluster. Clustering was performed on the 10 components of W (see Methods). e Heatmap of expression measures for marker genes for the 18 clusters found by our procedure: columns correspond to clusters and rows to genes; the value in each cell is the average log expression measure per cluster, centered and scaled so that each row has mean zero and standard deviation one
