EpiRank: partner-agnostic surface-patch retrieval for epitope prediction of protein binding interfaces

EpiRank, a partner-agnostic epitope predictor built on HumanPPI predicted human dimers, ranks a real binding epitope in its top surface patch 64% of the time on 3,529 filtered target chains.

• Why it matters: Standard ROC AUC hides the designer’s question—whether the top-ranked patch for a specific target is truly a binding site—and risks overstatement from split leakage and misleading baselines.
• What they did: Using only single-chain geometry and physicochemistry (170 features; no partner, sequence search, or structural alphabet), EpiRank was evaluated as a retrieval task with homology- and partner-graph-controlled splits on 12,296 target chains and 4.18 M residues.
• The result: Random compact patches hit 41% (a 1.57× lift), 54% of predicted patches land on nothing, and per-target rankings can change substantially across models trained on 2,000-chain versus 11,669-chain superset data.

AI summary of the abstract · may contain errors · not medical advice · read the paper.