The abstract is read at the publisher; the summary is JClub's.
On the CryptoBench test fold, cryptic-pocket detectors achieve 14.5-point coverage but only 36.7-point conversion to top-five recovery, and rankers differ by 10.6 points at equal coverage.
- Why it matters: Top-n benchmarks mix two distinct steps—proposing the true site and ranking it highly—so it is unclear where detector consensus actually improves performance.
- What they did: The study retained per-candidate overlaps across five candidate-generation methods in six configurations, then tested ranking sensitivity by comparing two rankers on the same candidate sets and injecting synthetic competitors from each target’s wrong-answer score distribution.
- The result: Union coverage reached 92.2% (98.6% for sites ≥ eight residues), but synthetic competitors reduced top-five recovery by 16.8 on training and 17.0 on test; pooling detectors helped little (0 at budget five, 11.8 at twenty).
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.