Researchers at the University of California San Diego have developed an artificial intelligence model designed to predict how tumors may respond to treatment by interpreting their genetic profiles.
The model, called MutationProjector, was trained using genomic data from more than 30,000 tumors spanning 10 solid cancer types. Published in Cancer Discovery, the study describes how the system links complex mutation patterns to the biological pathways associated with treatment response.
Genetic sequencing is now routine in oncology, but clinicians still face challenges when interpreting the large number of mutations identified in many tumors. Current precision oncology strategies rely heavily on a relatively small set of validated biomarkers, limiting the number of patients who can be matched to FDA-approved targeted therapies on the basis of tumor genetics.
MutationProjector takes a broader approach by analyzing combinations of genetic alterations rather than focusing on single biomarkers. The system generates a compact representation of a tumor’s biological state, allowing researchers to identify disrupted molecular pathways and infer which therapies may be most effective.
The researchers validated the model using multiple independent patient cohorts, including cases of bladder cancer, lung cancer, and melanoma. Across these datasets, MutationProjector matched or outperformed existing approaches for predicting responses to immunotherapy and chemotherapy. The model also identified established and previously unrecognized biomarkers linked to treatment outcomes.
Pretraining the model on large genomic datasets, combined with molecular network analysis, enabled the model to detect patterns that conventional biomarker-based methods may overlook, particularly for rare mutations.
The researchers emphasized that the model was designed to provide biological insight alongside prediction, an important feature for clinical decision-making in precision oncology. Future plans include expanding the system to additional cancer types and integrating imaging, transcriptomic, and electronic health record data.
Newsletters
Receive the latest pathologist news, personalities, education, and career development – weekly to your inbox.
