Fig. 3

A Isomap of the leading principal components for the T-cells in the Adams dataset, shown by individuals, sex, and smoking status. The number in the insets denotes how correlated the individual-level covariate is correlated with the leading principal components. B Isomap of the cell embedding after applying eSVD-DE, demonstrating that confounding covariate and individual effects have been removed. The number in the insets denotes how correlated the individual-level covariate is correlated with the estimated eSVD-DE cell embedding. C, D Relationship between gene expression and cells’ log sequencing depth before or after eSVD-DE (shown on a log scale), respectively, for groups of 6 genes partitioned by the gene’s mean expression relationship after eSVD-DE. After eSVD-DE, each gene becomes more uncorrelated with the cell’s log sequencing depth. E Relationship between each cell’s log sequencing depth and overdispersion parameter, highlighting two genes displaying different shrinkage levels via the posterior mean