AI Tool Predicts Cancer Treatment Success from Tumor DNA

For cancer patients, genetic testing often reveals dozens of mutations in their tumors, but doctors still struggle to translate those findings into treatment decisions that could save lives.

Now researchers at UC San Diego have developed an AI tool called MutationProjector that bridges this gap. The model analyzes a tumor's complex genetic profile and predicts how that specific cancer will respond to chemotherapy or immunotherapy.

The team trained their AI on genomic data from more than 30,000 tumors across 10 different cancer types. Instead of looking at just one or two genetic markers like traditional approaches, MutationProjector examines the broader pattern of mutations to understand what's happening inside the tumor at a molecular level.

"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 Dr. Trey Ideker, professor of medicine at UC San Diego and lead researcher on the study. Currently, only about 8% of cancer cases get successfully matched to an FDA-approved therapy based on genetics alone.

The researchers tested their model on independent groups of patients with bladder cancer, lung cancer and melanoma. MutationProjector matched or beat existing prediction methods for both immunotherapy and chemotherapy outcomes, according to the study published in Cancer Discovery.

The model also identified unexpected biomarkers linked to treatment success that doctors hadn't previously recognized. Many cancer mutations are individually rare, making them nearly impossible to study one at a time, explained Dr. JungHo Kong, the study's first author.

Why This Inspires

This isn't just about better predictions. The AI explains why it makes each recommendation, helping doctors understand the biological reasoning behind treatment choices. That transparency matters enormously when physicians are making life-and-death decisions with their patients.

The UC San Diego team plans to expand the model to cover more cancer types and integrate additional information like medical imaging and electronic health records. The goal is making genetic testing far more actionable for the 92% of patients who currently don't benefit from it.

For families navigating a cancer diagnosis, this research offers something precious: the hope that tomorrow's treatments will be more precisely matched to each person's unique disease.