CellPhenoX: An Explainable Machine Learning Method for Identifying Cell Phenotypes To Predict Clinical Outcomes from...

Area of Science:

• Computational biology
• Genomics
• Artificial intelligence

Background:

• Single-cell technologies reveal disease heterogeneity but struggle to link cell phenotypes to clinical outcomes.
• Existing methods lack interpretability and struggle with complex interaction effects (e.g., sex, age, disease).

Purpose of the Study:

• To develop an explainable machine learning method, CellPhenoX, for identifying cell-specific phenotypes and interaction effects associated with clinical outcomes.
• To enhance the clinical translation of single-cell data by providing interpretable, cell-specific insights.

Main Methods:

• CellPhenoX integrates classification models, explainable artificial intelligence (AI), and statistical frameworks.
• It generates interpretable, cell-specific scores to identify condition-associated cell populations.
• The method was benchmarked using simulations and real-world patient cohorts.

Main Results:

• CellPhenoX successfully identified cell-specific phenotypes and interaction effects across diverse single-cell studies.
• It detected an activated monocyte phenotype in COVID-19 correlated with disease severity.
• The method uncovered fibroblast state transitions predicting inflammation and identified therapy-induced T cell changes in the tumor microenvironment.

Conclusions:

• CellPhenoX provides a powerful, interpretable framework for translating single-cell findings into clinical impact.
• The method effectively addresses challenges in linking cell-level data to clinical outcomes and interaction effects.
• It demonstrates broad applicability in understanding disease heterogeneity and identifying biomarkers.