Fig. 1: An overview of the VoPo pipeline for robust clinical outcome prediction.

VoPo is an end-to-end bioinformatics pipeline for prediction and visualization of high-throughput single-cell data. a Patient samples are collected and the immune system is profiled at a single-cell level (image created with BioRender.com). b Cells from individual samples are first assigned to within-sample clusters. c A collection of cell populations (metaclusters) common to all samples is defined through repeated metaclustering. Each uniquely colored graph is a schematic representation of a unique metaclustering solution, with each node representing a metacluster. d Unsupervised Laplacian Score-based feature selection is applied independently to each metaclustering solution to reduce feature redundancy. This results in a collection of features acquired collectively across the independent metaclustering solutions. The color of a node (e.g., retained metacluster) schematically represents the metaclustering solution where it was produced (from c). e The features retained across independent metaclustering solutions are integrated and used to classify patients according to clinical phenotype.