AI Model Predicts Cancer Treatment Response from Tumor Genotype

Researchers at University of California, San Diego have developed a new artificial intelligence (AI) model that can translate a tumor’s complex genetic profile into predictions about how that cancer may respond to treatment. The foundation model, called MutationProjector, was trained on genomic data from more than 30,000 tumors across 10 solid cancer types, and validated through testing across multiple independent patient cohorts. Led by Trey Ideker, PhD, professor of medicine at UC San Diego School of Medicine and director of the Big Data Institute at the University of Oxford, the researchers say the model offers a new framework for connecting cancer mutations to the biological pathways that drive treatment response.

“Genetic sequencing is already routine in cancer care, but we still struggle to fully interpret the many mutations found in a patient’s tumor,” said Ideker, who also holds a second appointment at UC San Diego Jacobs School of Engineering and is a member of UC San Diego Moores Cancer Center. “Our goal with MutationProjector was to build a general-purpose model that can learn from tens of thousands of tumor genomes and turn those mutation patterns into more precise predictions about treatment response.”

Ideker is co-senior and co-corresponding author of the team’s published paper in Cancer Discovery, titled “A foundation model of cancer genotype enables precise predictions of therapeutic response,” in which the authors stated, “These results establish a unifying framework for connecting tumor genotypes to biological mechanisms and therapeutic outcomes.”

Following a cancer diagnosis, one of the next steps is often genetic testing, which helps doctors classify the tumor and decide which treatments to pursue. “DNA sequencing panels—and in particular those that broadly identify alterations in cancer-associated genes—have been widely adopted in the clinic due to their relatively low cost, rapid turnaround, and established relevance to treatment outcomes,” the authors explained.

However, while genetic testing is relatively low cost, fast, and has a strong track record in cases where validated genetic biomarkers are available, those cases remain limited, because this type of treatment stratification is currently based on only a small number of known biomarkers. Today, only about 8% of cases are successfully matched to an FDA-approved therapy on the basis of genetics and usually on the basis of a single gene, the team continued. “While this situation may reflect the incomplete scope of genes covered by current sequencing panels, it clearly also reflects a fundamental lack of knowledge about how gene mutations should be interpreted.”

They suggest that an “average” tumor has approximately 11 distinct genetic alterations identified by clinical sequencing, representing a potentially rich source of molecular information, if this information could be used to help select treatment. One of the challenges to matching cancer mutations with treatment outcomes is that most mutations are rare, the investigators pointed out. Another is that individual biomarkers do not function in isolation, but act together to influence drug response.

Unlike existing approaches that rely on a small number of biomarkers, MutationProjector analyzes the broader combination of genetic alterations present in a tumor. The model then uses this information to generate a compact representation of the tumor’s biological state, helping researchers interpret which molecular pathways may be disrupted and, by extension, which treatments may be most effective. “Foundation models, which are pre-trained on large datasets and then applied to solve diverse new challenges with relatively few samples, are especially well positioned to advance precision oncology,” Ideker and team noted.

The investigators trained their foundational model, MutationProjector, using genetic profile data from more than 30,000 tumors samples across different cancer types. They then showed that across several independent cohorts of cancer patients, including those with bladder cancer, lung cancer and melanoma, MutationProjector matched or exceeded existing methods for predicting response to common immunotherapy and chemotherapy treatments. The model also identified both known and unexpected biomarkers associated with treatment outcomes, which could help improve current approaches to genetic testing and patient stratification.

“When applied to predict immunotherapy or chemotherapy resistance across multiple cancer types and cohorts, MutationProjector achieves or exceeds state-of-the-art performance in all contexts,” the team wrote. “It identifies unexpected biomarkers, including KMT2D mutation in immunotherapy sensitivity and joint alteration of SMARCA4 and STK11 in immunotherapy resistance.”

JungHo Kong, PhD, first author of the study and a postdoctoral researcher in the department of medicine at UC San Diego School of Medicine, said, “Many cancer mutations are individually rare, which makes them difficult to study one at a time. By pretraining on a large collection of tumors and integrating molecular network knowledge, MutationProjector can detect patterns that would be easy to miss with conventional biomarker approaches. That gives us a way to move from long lists of mutations toward a more functional understanding of the tumor.”

The researchers emphasize that the model was designed not only to make predictions, but also to provide insight into why those predictions are made, which could help when refining biomarkers and treatment strategies. This interpretability is especially important in precision oncology, where clinicians need to understand how tumor genotypes relate to treatment decisions.

The team also hopes to expand the model to additional cancer types and data sources, including international cancer genome datasets and other forms of clinical information, such as imaging, transcriptomics, and electronic health records. “While 30,000+ genomes representing 10 solid tumor types were considered in our study, numerous additional tumor samples are available for expansion of MutationProjector to tumor types such as pancreatic cancer, prostate cancer or sarcomas,” the authors said. “Other future studies should explore the extent to which the MutationProject concept can be applied to further clinical tasks of interest, including application to liquid biopsies of circulating tumor DNAs for early cancer detection.”

Ideker added, “Our results suggest that tumor genome foundation models may help extend the clinical value of sequencing beyond a handful of well-known genes. This could support a more comprehensive and biologically grounded approach to precision oncology.”