# Single cell RNAseq signatures refined with combiroc enhance identification of NK cells in blood and solid tissues

## TL;DR

A new computational method called combiroc improves the identification of NK cells in blood and tissues using RNA sequencing data.

## Contribution

The combiroc framework enhances single-cell RNA-seq analysis by identifying non-canonical marker combinations for NK cells.

## Key findings

- Combiroc identifies NK cell markers that align with the Human Protein Atlas and are validated by cytometry and functional assays.
- The method improves differential expression signatures by reducing dimensionality and increasing interpretability and transferability.
- Combiroc effectively distinguishes NK cells from other cell types in both blood and solid tumor tissues.

## Abstract

Cytotoxic CD8 T lymphocytes (CTLs) and natural killer (NK) cells share the common objective of controlling infections and detecting and removing tumor cells, albeit through distinct target recognition mechanisms. Although CTLs belong to the adaptive immune system and NK cells are innate lymphoid cells, they frequently exhibit considerable overlap in their molecular phenotypes. This overlap, as with many others in cell biology, poses challenges for distinguishing cells in the context of single cell transcriptomics. Building on a previous ROC-driven combinatorial approach, we developed a new computational framework for single-cell RNA-seq with the combiroc R package, and in this study we showed that it can identify non-canonical marker combinations for NK cells in Peripheral Blood Mononuclear Cell datasets. These combinatorial markers were in line with the Human Protein Atlas and we validated them through cytometry staining and functional assays. Markers selected with combiroc exhibit exceptional discriminatory power for identifying NK cells, both in blood and solid tumoral tissues. Besides this finding, we showed that our approach vastly optimizes standard differential expression signatures: it reduces dimensionality while improving interpretability and transferability to diagnostic applications, offering a practical solution for refining immune cell identities in complex transcriptomic datasets.

The online version contains supplementary material available at 10.1038/s41598-025-29876-5.