COMPASS: A Generalizable AI Model for Immunotherapy Response

Most patients do not respond to immune checkpoint inhibitors (ICIs), and existing biomarkers often fail to generalize across cancer types and treatments. Researchers developed COMPASS, a pan-cancer foundation model that predicts ICI response directly from bulk tumor transcriptomes using a concept bottleneck transformer.

COMPASS encodes gene expression through 44 biologically grounded immune concepts (cell states, tumor-microenvironment interactions, and signaling pathways). Trained on over 10,000 tumors, it significantly outperforms 22 existing methods across 16 external clinical cohorts spanning multiple cancers and different ICIs. It also generalizes well to previously unseen cancer types and therapies.

Beyond prediction, COMPASS generates interpretable "personalized response maps" that link gene expression patterns to specific biological programs driving response or resistance (such as TGFβ signaling and endothelial exclusion in non-responders). These maps also correlate strongly with survival outcomes.

This Ledger Entry expands how readers think about AI in precision oncology by demonstrating a concept bottleneck transformer model that integrates biological knowledge with gene expression data to achieve strong generalization across cancers and therapies while providing interpretable insights into immune response mechanisms.