New curated antimicrobial peptide dataset and deep learning tool target toxicity prediction
- Toxicity is a major obstacle to turning antimicrobial peptides into medicines, and existing prediction approaches suffer from scattered data and inconsistent annotation.
- The authors present CHAMP, a machine-readable collection of antimicrobial peptides, and AMPToxPred, a deep learning model for predicting peptide toxicity.
- Detail on model design, dataset size and benchmarking is not available from the public abstract, so this note covers only what is stated.
💡 Points Based on the abstract
The dataset may matter more than the model. Peptide toxicity labels mix hemolysis, cytotoxicity and in vivo readouts across assays and cell types, so harmonizing them is real work. What I'd want to know is how sequence-similar peptides were split between train and test, and whether it beats a simple baseline on my own series. Without that, a good score tells me little.
📄 Paper · Digital Discovery (RSC) ↗ CHAMP and AMPToxPred: A Machine-Readable Dataset of Antimicrobial Peptides and Deep Learning-Based toxicity Prediction