University of California San Diego researchers unveiled MutationProjector, an AI-enabled tumor genome foundation model designed to map mutational co-dependencies and predict therapy response. The model was published in Cancer Discovery and trained on tumor genetics from more than 30,000 cases across 10 solid cancer types using data from Project GENIE and The Cancer Genome Atlas. MutationProjector compresses a tumor’s genetic variance into compact representations using UMAP-based projections, then interrogates actionable features that may reflect perturbed molecular pathways. In internal testing, it reportedly recovered masked gene mutation statuses and matched or exceeded performance of existing prediction methods across multiple patient cohorts. The clinical promise is to reduce the gap between broad sequencing coverage and actionable therapeutic matches, by interpreting combinations of mutations rather than single-gene markers. Practical deployment will require rigorous external validation and integration into clinical workflows for actionable insights.