The paper introduces CellPhenoX, a novel, explainable machine learning method designed to predict clinical phenotypes using single-cell multi-omics data. This computational tool integrates classification models, explainable AI (XAI) techniques, and a statistical framework to generate cell-specific Interpretable Scores that link cellular populations to clinical outcomes while accounting for covariates and interaction effects like sex and age. The authors demonstrate CellPhenoX's superior performance in identifying condition-associated cell populations, including rare ones, across simulated datasets and real-world studies, notably identifying an activated monocyte phenotype in COVID-19 severity and fibroblast changes in ulcerative colitis. Ultimately, CellPhenoX aims to bridge single-cell biological findings with tangible clinical impact by providing robust and interpretable insights into complex disease mechanisms.

References:

• Young J, Inamo J, Caterer Z, et al. CellPhenoX: An eXplainable Cell-specific machine learning method to predict clinical Phenotypes using single-cell multi-omics[J]. bioRxiv, 2025.