AI Model Links Tumor Mutations to Treatment Response

Researchers at the University of California San Diego have created an AI model named MutationProjector to interpret tumor DNA and predict treatment responses for various cancers. The model was trained using genomic data from over 30,000 tumors across 10 solid cancer types, including bladder, lung, and melanoma. Unlike current methods that rely on limited biomarkers, MutationProjector analyzes broader genetic alterations to predict how tumors might respond to immunotherapy and chemotherapy. Published in *Cancer Discovery*, the study found that MutationProjector matched or exceeded existing approaches in predicting treatment outcomes across multiple patient cohorts. It also uncovered both known and unexpected biomarkers linked to treatment responses, offering new insights for genetic testing and patient care. Trey Ideker, PhD, professor of medicine at UC San Diego and director of the Big Data Institute at the University of Oxford, led the research. He noted that while genetic sequencing is routine in cancer care, interpreting mutations remains challenging. MutationProjector aims to bridge this gap by translating complex genetic data into actionable predictions. The model works by generating a simplified representation of a tumor’s biological state, helping researchers identify disrupted molecular pathways. This approach could improve how doctors select treatments, particularly for cases where current genetic biomarkers are insufficient. JungHo Kong, PhD, the study’s first author, explained that many cancer mutations are rare and difficult to study individually. MutationProjector’s ability to detect patterns across large datasets provides a more functional understanding of tumors, moving beyond traditional mutation lists. The researchers emphasize that MutationProjector not only predicts treatment responses but also explains why certain outcomes occur, potentially advancing personalized cancer care.