AMPGP: Discovering Highly Effective Antimicrobial Peptides via Deep Learning

Antimicrobial peptides (AMPs) have emerged as vital candidates in the fight against antibiotic resistance. The traditional processes for AMP design and discovery are often time-consuming and inefficient. Here, we propose the AMPGP model, which employs deep learning algorithms for both generation and prediction. The generation model incorporates an attention mechanism into the seqGAN framework to generate high-quality AMPs. The prediction model is structured into four distinct feature channels to address the limitations of relying on a single source of information. The evaluation on the independent test set achieved an accuracy of 98.46%, surpassing several advanced models. Ultimately, we identified 10 candidate AMPs, and the experiment indicated that peptide No. 1 (LITHLFRFKNSGRILM) and No. 2 (FKLSVLYLGRGNIMKAYYGIKIARAG) exhibited broad-spectrum antibacterial and cellular viability, with no significant hemolytic activity observed. The AMPGP model thus presents a promising approach for discovering effective peptides and enhances the potential for clinical applications.