The abstract is read at the publisher; the summary is JClub's.
AbPoolRank selects mixed antibody–antigen structure pools to achieve 58.0% medium-or-better Success@1 (DockQ ≥ 0.49) on 103 cluster-disjoint post-2021 complexes from SAbDab2.
- Why it matters: Antibody–antigen prediction often over-relies on deep sampling from one co-folding predictor and model-specific confidence, which can miss better structures and wastes compute.
- What they did: AbPoolRank ranks 15-structure pools built from three outputs each of Boltz-2, Promera, Protenix-v2, ESMFold2-Fast, and OpenDDE using cross-pose agreement, within-source reproducibility, and interface geometry.
- The result: On 100-times single-predictor sampling, AbPoolRank’s 58.0% exceeds every individual predictor’s 26–52%, matches the best single-structure methods, and uses 1.8-fold less GPU time than the most accurate single-predictor setting.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.