The abstract is read at the publisher; the summary is JClub's.
Protein language model embeddings plus deep learning models best predict antimicrobial peptides across eight curated AMP datasets, and applying them to ancient microbiomes yields 111, 76, and 50 high-confidence candidates.
- Why it matters: Antimicrobial resistance demands new AMP discovery, but it is unclear which computational methods and feature types work best, and ancient microbiomes remain underexplored as AMP reservoirs.
- What they did: The team benchmarked six machine learning methods with twelve feature configurations (sequence, structural, and protein language model features) under one framework, then screened three ancient metagenomic datasets and used reinforcement learning to pick candidates from ancient sequence pools without modern AMP labels.
- The result: Deep learning architectures (convolutional, transformer-based, and recurrent) with protein language model features outperformed classical methods, and ancient-catalog predictions differed from modern-catalog candidates; reinforcement learning selected additional candidates with distinct properties.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.