Fig. 2: Applying VoPo to three mass cytometry datasets.
From: VoPo leverages cellular heterogeneity for predictive modeling of single-cell data
Single-cells across samples were projected into two dimensions using tSNE in the HSR (a), NTP (b), and LSR (c) datasets. Cells were colored by their computed differentiation score, which depicts the degree of association with the clinical outcome of interest. Cells that are brightly colored likely belong to cell populations that are strongly associated with a particular clinical outcome. The direction of frequency differences between case and control samples is shown in Supplementary Figs. 7, 10, and 15. d Integrating features extracted over the fifty metaclustering solutions with VoPo (pink boxplots) resulted in higher classification accuracy (AUC) over the baseline case where VoPo was not applied (black boxplots) across the three datasets. The boxplots show median values, interquartile range, whiskers of 1.5 times interquartile range, and all individual points.
