Combiroc: when 'less is more' in bulk and single cell marker signatures.

Abstract

Here we present the combiroc R package, for signatures refinement in high throughput omics. Based on a ROC-driven marker selection, it can be used to find powerful smaller sub-signatures from scRNAseq experiments and to annotate cells using fewer markers. Trained on PBMC dataset, combiroc found NK marker combinations with high cell-discriminating power, in agreement with human protein atlas and that were validated both computationally and experimentally on independent datasets.