The abstract is read at the publisher; the summary is JClub's.
GRABS consolidates 4,380 canonical antibody–antigen records with experimentally measured Kd values, verified by content keys to achieve 100% structure-backed integrity.
- Why it matters: Binding-affinity ML is hindered by scattered, incompatible public labels that mix genuine Kd values with censored or predicted data and can create batch or alignment leaks.
- What they did: The dataset merges nine public sources into structure-ready artifacts, using md5(VH+VL+antigen)[:10] content addressing and a regeneration check that matches node features to sequences for 4,282/4,282 records.
- The result: On a leak-free, antigen-cluster-grouped benchmark, cross-target variant ranking collapses to the noise floor (Spearman rho ≈ 0 for every model family), and structure provides no advantage.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.