🛰️ Daily AI Frontier
‹ back to 2026-08-07

EpiBench: Can LLMs Understand Epitopes for Antibody Drug Discovery?

Research Bioinformatics AI

Ranking

Overall 68
Content 80
Popularity 39

Observed public metrics from 1 member.

Merged summary

TL;DR - EpiBench is a closed-book, sequence-only benchmark of 1,609 curated samples testing whether LLMs can reason about antibody epitopes, and it shows current general-purpose models fall short of reliable epitope understanding for antibody drug discovery.

  • Data is grounded in structural antibody–antigen contacts, curated functional B-cell assays, and deep mutational scanning escape measurements; scoring is automatic.
  • Covers five linked tasks spanning the development workflow: targetable region discovery, antibody-conditioned epitope identification, epitope binning, functional epitope assessment, and antibody escape assessment, with controlled sampling to limit shortcut exploitation.
  • Nine general-purpose LLMs were evaluated with task-specific baselines, antigen length stratification, explicit-reasoning comparison, and failure-mode inspection.
  • Findings: models capture partial epitope signal but are weak at antibody-specific sequence grounding, long-context residue localization, and biologically grounded reasoning.

Sources (1)

EpiBench: Can LLMs Understand Epitopes for Antibody Drug Discovery?

arXiv cs.CL Zirui Wang, Jiaqi Wang, Qinghan Wang, Yuzhi Xu, Gang Du, Tingjun Hou, Odin Zhang 2026-08-06 arXiv:2608.06022
Public signals Semantic Scholar citations 0 · Semantic Scholar influential citations 0
Providers: Hugging Face · N/A OpenAlex · N/A Publisher · N/A Semantic Scholar · Citations 0 · Influential citations 0 X · N/A Fetched 2026-08-24 14:25:13.114149 UTC

TL;DR - EpiBench is a closed-book, sequence-only benchmark of 1,609 curated samples testing whether LLMs can reason about antibody epitopes, and it shows current general-purpose models fall short of reliable epitope understanding for antibody drug discovery.

  • Data is grounded in structural antibody–antigen contacts, curated functional B-cell assays, and deep mutational scanning escape measurements; scoring is automatic.
  • Covers five linked tasks spanning the development workflow: targetable region discovery, antibody-conditioned epitope identification, epitope binning, functional epitope assessment, and antibody escape assessment, with controlled sampling to limit shortcut exploitation.
  • Nine general-purpose LLMs were evaluated with task-specific baselines, antigen length stratification, explicit-reasoning comparison, and failure-mode inspection.
  • Findings: models capture partial epitope signal but are weak at antibody-specific sequence grounding, long-context residue localization, and biologically grounded reasoning.
item →