AI in cancer vaccines: from target ranking to clinical evidence

What Are Cancer Vaccines? How Can They Prevent and Treat Cancer? | Cancer Straight Talk Podcast

Memorial Sloan Kettering Cancer Center · 2024-08-01 · Source

MSK’s English-language expert discussion of preventive and therapeutic cancer vaccines. The institutional episode is dated 12 May 2023; the date shown is the YouTube publication on 1 August 2024. It describes research at that time, not current treatment or authorisation guidance.

Start with the distinction: a treatment vaccine

NCI explains that therapeutic cancer vaccines aim to strengthen immune responses against tumours in people who already have cancer. Their purpose differs from prevention vaccines. Some approaches target features of an individual tumour; others use shared antigens. “Cancer vaccine” therefore describes a field, rather than one product or a single level of evidence.

Research evidence: what does the AI predict?

The 2023 BigMHC study focused on predicting protein fragments presented through MHC-I and candidates capable of eliciting an immune response. The model learned from mass-spectrometry data, then used transfer learning with immune-response assays. Researchers reported improved prediction performance against the compared methods. This is evidence about selecting candidates, not a vaccine-efficacy result measuring a patient’s chance of recovery.

Clinical evidence: the limits of early findings

NCI’s 23 April 2025 assessment discussed small kidney and pancreatic cancer studies of personalised vaccines. Immune responses and follow-up findings were encouraging, but larger studies were needed. In particular, the absence of recurrence after surgery could not simply be attributed to vaccination. These studies should not be read as direct demonstrations of clinical benefit from BigMHC.

Editorial assessment: four questions for reading a study

Our suggested reading framework separates four questions: What does the model predict? Which independent experiment tests that prediction? What outcome was measured in patients? And which product, indication and regulatory decision support the proposed use? If a presentation leaves one of these unanswered, keep it as an open question rather than filling the gap with an assumption. Alongside the speed of generating a target list, an assessment should ask how incorrectly selected and overlooked targets will be investigated. This is our editorial framework, not a validated scoring system.

Authorisation and routine care are separate questions

NCI also describes an approved treatment vaccine, sipuleucel-T, with a defined use. That example does not confer approval on another vaccine candidate or AI model. This article supports research literacy; it does not select treatment for individuals or claim clinical effectiveness for RAI products.