A cancer vaccine made for one person. How does AI help?

A vaccine made from the clues in your own tumor sounds extraordinary. But it raises a practical question: which clues should it use?

That question sits at the center of personalized cancer-vaccine research. It also explains why AI has a role in the field.

This week brings two different views of that story. UCSF recently described a melanoma patient's experience entering a personalized-vaccine trial after surgery. A new review in Biotechnology Advances, published September 1, examines how generative AI might help researchers design these vaccines. The review maps possibilities and unresolved problems; it does not report a new patient trial.

Cancer cells can acquire DNA changes that produce altered proteins called neoantigens. These can provide targets for an immune response.

The aim of a cancer treatment vaccine is to help the immune system recognize features of cancer cells and respond to them. As the National Cancer Institute explains, some vaccines are custom-made around a person's tumor, while others target features shared by cancers in different people.

For an individualized vaccine, identifying a mutation is only part of the problem. Researchers must decide which potential targets deserve a place in the design.

The delivery method is another piece. In an mRNA vaccine, the RNA carries instructions that cells use to make proteins. Immune cells can then encounter those proteins and learn to respond. NCI's explanation of the underlying biology describes how this process can help teach T cells what to look for. That explains the appeal of personalization: the instructions can be chosen around features of a particular tumor.

The new review describes computational tools that rank possible targets using features such as whether a protein fragment is likely to be presented to immune cells. Think of this as helping researchers narrow a list of candidates.

Generative AI could extend that work into proposing and refining vaccine designs, including how multiple targets are combined. The authors describe it as part of a larger research process that still depends on evidence from the tumor and laboratory testing.

The hard questions are biological: will the immune system recognize the chosen targets, and can the cancer evade the response? The review says the clinical value of generative AI still needs to be established through prospective comparisons with standard approaches. A better computer-generated design has to earn its place in patient care.

Consider the difference between spotting an unusual feature and finding a useful target. A mutation may look interesting in sequencing data, but a vaccine strategy also depends on whether the relevant fragment is presented to immune cells and recognized by them. A high prediction score answers only part of that question. This is why the review emphasizes experimental validation alongside computation.

One concrete example is pVACtools, an open-source suite for identifying, prioritizing, and reviewing potential neoantigens. In a June 2026 preprint describing version 6, its developers report expanded prediction features and tools to support vaccine design.

For a research team, that means more ways to examine candidates and compare choices before committing to a design. For readers, it helps make the AI story less mysterious: the work involves specialized software, biological measurements, and decisions that researchers can inspect.

pVACtools is an example of computational research infrastructure; it should not be confused with proof that generative AI improves treatment. The June report describes software capabilities. It does not establish a survival benefit, and running the software does not produce a treatment ready for a patient.

UCSF's account puts the research in a recognizable setting: a person who had melanoma removed and was worried about it returning. His physician referred him to a clinical trial.

That study, INTerpath-001, tested an individualized vaccine alongside pembrolizumab, also called Keytruda, after surgery for high-risk melanoma. Merck and Moderna reported positive Phase 3 topline findings in August. Their announcement is about that treatment combination and setting. It does not establish that the generative methods discussed in the new review improve patient outcomes.

For a patient or family, the practical starting point is a conversation with the treating team: “Are there personalized-vaccine studies relevant to this cancer and this stage of treatment?” Whether a particular study fits requires an individual clinical assessment.

The next meaningful advance would connect the computer's choices to biology and then to patient outcomes. That means testing whether selected targets provoke the intended immune response, comparing approaches fairly, and establishing whether the resulting treatment helps people in a defined cancer setting.

Those are different achievements. A tool can become better at proposing candidates before anyone knows whether using it improves a vaccine's clinical performance.

Choosing targets is a central challenge behind a personalized cancer vaccine. AI gives researchers more ways to approach it. The test that matters to patients is whether those choices lead to treatments that help them.

— Alex

ImmunaPath provides educational information, not medical advice. Discuss treatment and clinical-trial decisions with your oncologist or qualified healthcare team.