Computational & AI Journal of Computer-Aided Molecular Design (Springer Nature)
Machine-learning genome mining of Bacillus yields novel antimicrobial peptides, tested on a few synthetic candidates
- Machine-learning mining of several thousand Bacillus genomes yielded putative antimicrobial peptides in several hundred families, sequence-distinct from human and animal AMPs yet similar in charge and hydrophobicity.
- A MIC predictor ranked candidates; a handful were synthesized, with activity varying by species and strongest against Clostridium perfringens.
- Predictions matched measurements for some species but poorly or inversely for others, so validation is partial.
A MIC predictor that tracks measurements for some species and runs inverted for others, tested on a handful of peptides, tells me the model hasn't earned a place in triage yet. I'd want to know what it was trained on, how far these sequences sit from that training data, and whether a simple charge-and-hydrophobicity baseline ranks them as well. Then synthesize a larger, blinded set.
Paper: Journal of Computer-Aided Molecular Design (Springer Nature) ↗ Genome-scale mining and experimental characterization of antimicrobial peptides from Bacillus