The abstract is read at the publisher; the summary is JClub's.
In a benchmark of 16,168 antimicrobial peptide sequences, enforcing sequence-homology separation and length matching reduces MCC from 0.834 to 0.661, with length effects persisting across partitions.
- Why it matters: Published antimicrobial peptide predictors can look near-perfect because benchmarks omit key controls, so reported accuracy may reflect dataset artifacts rather than generalizable recognition.
- What they did: The authors propose a controlled evaluation protocol adding homology and length-bias controls plus a sample-size control, then test a four-model ensemble and ten baselines on a fixed benchmark of 16,168 sequences.
- The result: A length-only classifier separates far above chance; adding homology costs 0.067 MCC and length matching costs an additional 0.115, and across ten redrawn partitions the homology term shrinks while length remains, reordering methods and letting a frozen ESM-2 embedding overtake the ensemble.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.