Uncovering the Power of AMPs: A New Deep Learning Method for Prediction

Antimicrobial peptides, nicknamed AMPs, are tiny peptides that help defend against diseases. As antibiotic resistance grows thanks to misuse, finding AMPs has become a big deal. These peptides could be an alternative to traditional antibiotics. The challenge lies in accurately identifying AMPs using computers, which has been a hot topic in bioinformatics. While there are many tools for this, most only focus on a few specific activities. Predicting multiple activities would help find peptides with wide-ranging abilities. Enter deep-AMPpred, a two-stage predictor. First, it spots AMPs among other peptides. Then, it tackles the 13 most common activities AMPs can have. deep-AMPpred uses the ESM-2 model to capture the big picture of peptide sequences. It then combines CNN, BiLSTM, and CBAM models to dig deeper into local features, long-term and short-term dependencies, and attention mechanisms. This combo boosts performance in predicting both AMPs and their activities. Tests show deep-AMPpred is great at finding AMPs and predicting their activities. It proves the ESM-2 model can grab meaningful peptide features and multiple deep learning models can work together to make AMP identification and activity prediction more effective.