Show Notes
Drost F et al., Cell Genomics - Drost et al. integrated 21 pre-trained sequence-based TCR-epitope predictors into ePytope-TCR and benchmarked them on a viral single-cell repertoire and deep mutational scans, revealing performance biases and limited generalization to rare and mutated epitopes. Key terms: T cell receptor, epitope prediction, ePytope-TCR, benchmarking, cross-reactivity.
Study Highlights:
The authors unified 21 sequence-based TCR-epitope predictors inside ePytope-TCR and evaluated them on two challenging datasets: a viral single-cell repertoire and deep mutational scans. Models predicted binding reliably for frequently observed epitopes but failed for less-represented targets. Strong biases in prediction scores between epitope classes were evident and most methods could not predict effects of single-residue epitope mutations. ePytope-TCR provides interoperable interfaces and a standardized benchmark to guide tool selection and method development.
Conclusion:
Pre-trained TCR-epitope predictors can annotate well-studied epitopes but do not generalize to rare or mutated epitopes; ePytope-TCR standardizes access and benchmarking to help researchers choose tools and accelerate development of improved models.
Music:
Enjoy the music based on this article at the end of the episode.
Article title:
Benchmarking of T cell receptor-epitope predictors with ePytope-TCR
First author:
Drost F
Journal:
Cell Genomics
DOI:
10.1016/j.xgen.2025.100946
Reference:
Drost F., Chernysheva A., Albahah M., Kocher K., Schober K., Schubert B. Benchmarking of T cell receptor-epitope predictors with ePytope-TCR. Cell Genomics. 2025;5:100946. doi:10.1016/j.xgen.2025.100946
License:
This episode is based on an open-access article published under the Creative Commons Attribution 4.0 International License (CC BY 4.0) – https://creativecommons.org/licenses/by/4.0/
Support:
Base by Base is independent and ad-free — no sponsors, no paywall. If an episode was worth your time, chip in and keep the papers audited and the original songs coming:
❤️ Support monthly: https://buy.stripe.com/cNifZhclVebvagk2JDgEg01
☕ One-time donation: https://donate.stripe.com/7sY4gz71B2sN3RWac5gEg00
More at basebybase.com
On PaperCast Base by Base you'll discover the latest in genomics, functional genomics, structural genomics, and proteomics.
Episode link: https://basebybase.com/episodes/epytope-tcrbenchmark-suite-corrupted-pdf
QC:
This episode was checked against the original article PDF and publication metadata for the episode release published on 2025-07-13.
QC Scope:
- article metadata and core scientific claims from the narration
- excludes analogies, intro/outro, and music
- transcript coverage: Audited transcript sections describing the Epitope-TCR benchmarking framework, data formats and ingestion, viral benchmarking dataset, deep mutational scanning dataset, performance metrics, bias observations, and practical implications.
- transcript topics: Benchmarking framework and Epitope-TCR architecture; Data format interoperability and unified interface; Categorical vs general predictors; Viral benchmarking dataset (638 TCRs, 14 epitopes, 5 MHC backgrounds); Deep mutational scanning dataset (epitope mutations VPSVWRSSL and NLVPMVATV); Performance metrics (AUC, recall@K, APS, F1) and results
QC Summary:
- factual score: 10/10
- metadata score: 10/10
- supported core claims: 8
- claims flagged for review: 0
- metadata checks passed: 4
- metadata issues found: 0
Metadata Audited:
- article_doi
- article_title
- article_journal
- license
Factual Items Audited:
- ePytope-TCR provides a unified interface ingesting six common data formats and 21 pre-trained models
- Two benchmarking datasets used: a viral dataset and a deep mutational scanning dataset
- Viral dataset composition: 638 distinct TCRs, 14 epitopes, 5 MHC backgrounds
- Mutation dataset comprises deep mutational scans with 132 mutations for VPSVWRSSL and 172 mutations for NLVPMVATV; evaluated on 26 TCRs
- top performance: four general and two categorical methods exceed AUC 0.6; best AUC ~0.63 (MixTCRpred)
- Recall@1 results: approximately 0.24–0.26 across evaluated models
QC result: Pass.
