irAEGIS: Pathway-guided interpretable prediction of immune-related adverse events

irAEGIS predicts immune-related adverse event risk from pre-treatment scRNA-seq and consistently outperforms seven existing methods across simulated datasets and independent ICI cohorts.

• Why it matters: irAEs are major complications of immune checkpoint inhibitor therapy, yet pre-treatment prediction and biological interpretation from single-cell data remain difficult.
• What they did: The pathway-guided, interpretable model uses gene set constrained representation learning, aggregates pathway activities within cell types, and combines cell type–specific risk signals via a hierarchical framework.
• The result: In non-small cell lung cancer, neutrophil-associated inflammatory pathways in pre-treatment cells are linked to future irAE development, while in melanoma, apoptotic and immune activation programs in CD8+ NKT-like cells are linked to severe irAEs, supported by enrichment, communication, and trajectory analyses.

AI summary of the abstract · may contain errors · not medical advice · read the paper.