Researchers at UC San Diego have developed an AI model called MutationProjector that analyzes the complete genetic landscape of individual tumors to predict treatment response, trained on data from over 30,000 cancer genomes across 10 solid cancer types. The breakthrough addresses a persistent gap in precision oncology: while genetic sequencing is now routine in cancer care, doctors still struggle to interpret the thousands of mutations found in each tumor and translate that information into actionable treatment decisions.

Why Current Cancer Biomarker Approaches Fall Short?

Today, only about 8 percent of cancer cases are successfully matched to an FDA-approved therapy based on genetics, a figure that reveals the limitations of current biomarker-focused approaches. These traditional methods rely on a small handful of well-validated genetic markers, which means most patients with rare or unusual mutations fall through the cracks. MutationProjector was designed specifically to address this gap by moving beyond single-gene biomarkers and instead analyzing the full combination of genetic alterations present in a tumor.

"Genetic sequencing is already routine in cancer care, but we still struggle to fully interpret the many mutations found in a patient's tumor. 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," said Trey Ideker, genetic biologist at UC San Diego School of Medicine.

Trey Ideker, Genetic Biologist at UC San Diego School of Medicine

The model generates a compact representation of each tumor's biological state by analyzing mutation patterns holistically, then flags which molecular pathways may be disrupted and which treatments may be most effective as a result. The research was published in Cancer Discovery.

How Does MutationProjector Improve on Existing Methods?

Across independent cohorts of patients 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 surfaced both known and unexpected biomarkers associated with treatment outcomes, including KMT2D mutation in immunotherapy sensitivity and co-alteration of SMARCA4 and STK11 in immunotherapy resistance.

One key advantage of the approach is its ability to detect patterns that would be easy to miss with conventional biomarker approaches. Many cancer mutations are individually rare, making them difficult to study one at a time. By pretraining on a large collection of tumors and integrating molecular network knowledge, MutationProjector can identify meaningful patterns across the entire mutational landscape.

"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," explained JungHo Kong, postdoctoral researcher in the Department of Medicine at UC San Diego School of Medicine and first author on the study.

JungHo Kong, Postdoctoral Researcher in the Department of Medicine at UC San Diego School of Medicine

How Drug Developers Can Use This Technology

• Patient Stratification: The model helps researchers identify which patient populations a candidate therapy is most likely to benefit, enabling more biologically informed hypotheses about patient selection early in drug development.
• Trial Design Optimization: By understanding tumor biology more comprehensively, drug developers can design clinical trials that enroll the right patients, potentially improving success rates and reducing development timelines.
• Interpretability and Clinical Translation: The model was built with interpretability in mind, allowing researchers to understand why a prediction is made, not just what it is, a feature critical for translating genomic findings into actionable clinical criteria.

For oncology drug discovery teams, the appeal of MutationProjector extends beyond predicting treatment response in known cancer types. A persistent bottleneck in early-stage oncology programs is identifying which patient populations will benefit from a candidate therapy and doing so early enough to meaningfully shape trial design. Models that integrate broader mutational context, rather than relying on single-gene biomarkers, could help drug developers build more biologically grounded hypotheses about patient selection.

What's Next for Tumor Genome Foundation Models?

The UC San Diego team plans to expand MutationProjector 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. This expansion could further broaden the model's applicability and clinical utility across diverse patient populations and cancer subtypes.

"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," noted Trey Ideker.

Trey Ideker, Genetic Biologist at UC San Diego School of Medicine

The development of MutationProjector represents a significant shift in how the field approaches genomic interpretation. Rather than waiting for researchers to manually discover and validate individual biomarkers one at a time, foundation models trained on large-scale genomic datasets can learn complex patterns that connect tumor mutations to treatment outcomes. This approach has the potential to democratize precision oncology by making advanced genomic interpretation available to more patients, regardless of whether their specific mutations have been previously studied.