Saliva test coupled with AI could flag chemotherapy risk, early study results suggest

Early results from a new study suggest that a simple saliva test with powerful artificial intelligence algorithms could help doctors identify cancer patients at high risk for severe side effects from a widely used chemotherapy drug.

The drug, 5-fluorouracil (5-FU), has been in use since the 1950s — making it one of the oldest chemotherapy agents still used today. It remains a cornerstone of cancer treatment, used in nearly a third of chemotherapy treatments for colorectal, breast, head and neck, pancreatic, and stomach cancers.

5-FU disrupts how cells make and use DNA. Cancer cells divide rapidly and need to copy their DNA continuously. The drug mimics DNA building blocks and jams the process, halting cancer cell growth.

But some patients carry genetic differences that prevent their bodies from breaking down the drug properly. In those cases, 5-FU can build up to toxic levels, causing serious and sometimes life-threatening complications.

About one in three people has a reduced ability to process 5-FU. Although the federal Food and Drug Administration recommends screening, it is not widely implemented.

"This isn't just about improving outcomes, it's about saving lives," said Carla Finkielstein, a professor at the Fralin Biomedical Research Institute at VTC and senior author of the study. "For decades, 5-FU has been used with the assumption that it works the same for everyone. But we know it is not the case. Some individuals have a genetic susceptibility to this drug. Our approach adds a layer of precision to a treatment that's been around for decades but still carries serious risks for certain patients."

This research underscores the critical role of collaboration with oncologists and pathologists, whose clinical insights contributed to the research direction, Finkielstein said.

The preliminary findings were presented by the leading researchers of the project, John Janiga and Dzenis Mahmutovic, at the 2025 ASCO Gastrointestinal Cancers Symposium and also published as an abstract in the Journal of Clinical Oncology, signaling early interest in the study's clinical potential.

Scientists focused on mutations in a gene called DPYD, which produces the enzyme responsible for breaking down 5-FU.

They initially analyzed DNA from healthy volunteers and cancer patients to determine whether saliva samples could reliably detect known DPYD mutations. To expand beyond known variants, the team later employed advanced artificial intelligence tools alongside 3D protein modeling to evaluate thousands of samples.

"The AI-driven approach helped the team assess the structural and functional impact of previously unrecognized mutations in DPYD, uncovering potentially harmful variants that conventional methods might have missed," said Katherine Brown, the team's lead bioinformatician.

Several newly identified DPYD mutations in colon cancer patients were predicted to impair 5-FU metabolism. Two were labeled "pathogenic" by multiple predictive tools and confirmed by a clinical genetics database.

Some high-risk mutations were also found in healthy individuals, highlighting the need and feasibility of implementing saliva-based genetic screening for broader population-level risk assessment, according to Finkielstein, who leads the Virginia Tech Molecular Diagnostics Laboratory at the Fralin Biomedical Research Institute.

"If we can flag high-risk patients early, we can tailor their treatment plans, reduce hospitalizations, and potentially avoid fatal complications," said Finkielstein, who is also a professor of biological sciences in the College of Science. "This is a real step forward in making cancer care safer, smarter, and more personalized."

The early findings could expand current genetic testing and guide safer, more personalized chemotherapy.

More information:
John M Janiga et al, Enhancing a low-cost, minimally invasive screening test for dihydropyrimidine dehydrogenase mutations linked to 5-fluorouracil sensitivity by integrating computational mutation predictions., Journal of Clinical Oncology (2025). DOI: 10.1200/JCO.2025.43.4_suppl.226

Provided by Fralin Biomedical Research Institute at VTC