rethinkPeptides Search
Menu
Study breakdown

Why Predicting Which Peptides Will Trigger Anti-Cancer Immune Responses Is Still Hard

ReviewN/A (Review) evidence
The takeaway

Computer prediction of tumor peptide antigens is useful but unreliable — nearly one-third of validated cancer antigens would be missed by standard binding cutoffs.

~33% missed

Nearly one-third of validated tumor antigens have MHC binding below the standard 500 nM cutoff

What the researchers found

Nearly one-third of experimentally validated T cell tumor antigens have MHC binding affinities (IC50 > 500 nM) below standard prediction cutoffs, highlighting the limitations of in silico approaches.

Why it matters

Personalized cancer vaccines depend on accurately predicting which tumor peptides the immune system will recognize. If algorithms miss one-third of real antigens, patients may receive suboptimal vaccines.

The numbers in context

~33% of validated antigens had IC50 >500 nM; multi-step processing pathway; patient-specific validation recommended

How the study worked

Literature review of antigen processing, presentation, epitope discovery, and computational prediction methods. Validated tumor antigens assessed against predicted MHC class I binding scores.

Who was studied

N/A (review of validated T cell cancer antigens)

What this study cannot tell us

Review article — no new experimental data generated; computational prediction tools have continued to improve since publication.

How to read the evidence

N/A — narrative review synthesizing existing literature on antigen processing and prediction accuracy.

When this study was published

Published in 2020; AI-based antigen prediction tools have advanced significantly since then.

The bigger picture

Personalized cancer vaccines (neoantigen vaccines) are a rapidly growing immunotherapy approach. Improving antigen prediction accuracy is critical to making these vaccines effective for more patients.

Questions still open

  • Can machine learning approaches trained on validated antigens improve prediction accuracy beyond traditional binding affinity cutoffs?
  • Should clinical neoantigen vaccine pipelines use lower IC50 thresholds to avoid missing valid antigens?
  • How much does antigen processing (not just MHC binding) contribute to prediction failures?

Common questions

What are tumor antigens?
Peptide fragments from mutated or abnormal proteins in cancer cells that the immune system can recognize and attack.
Why can't computers just predict the right vaccine targets?
Antigen processing involves many steps beyond just binding to MHC molecules, and current algorithms oversimplify this complexity, missing about a third of real targets.

Read the original research

Antigen processing and presentation in cancer immunotherapy.

Journal for immunotherapy of cancer, 8(2)

Citation

Lee, Maxwell Y; Jeon, Jun W; Sievers, Cem; Allen, Clint T. (2020). Antigen processing and presentation in cancer immunotherapy.. Journal for immunotherapy of cancer, 8(2). https://doi.org/10.1136/jitc-2020-001111