CARS-AMP: A Simple and Efficient Deep Learning Model for Antimicrobial Peptide Prediction

With the increase in global antibiotic resistance, there is an urgent demand for antibiotic substitutes, for which antimicrobial peptides (AMPs) have attracted a lot of attention due to their unique mechanism of action that makes it difficult for bacteria to develop drug resistance. Beyond the natural discovery of AMPs, artificial intelligence techniques have become powerful tools to explore larger chemical spaces and accelerate the process of AMPs discovery. In this study, we construct a deep learning model, CARS-AMP, containing convolutional neural networks, recurrent neural networks, and self-attention, for the classification of anti-Staphylococcus aureus peptides. The model is trained with a small but concentrated data set and exhibits near-best performance in predicting anti-S. aureus peptides. The results show that the model can achieve 92.4% accuracy in predicting anti-S. aureus peptides. In addition, we further test the ability of the model to distinguish similar sequences by comparing it with the experimental results of the ZXR-2 series of peptides, and the results suggest that the model may need improvement to discern the differences in side chain size and sequence length. The CARS-AMP model provides a simple predictor with high efficiency and could be further improved to achieve better performance and wider application in identifying AMPs.