ProT-Diff: A Modularized and Efficient Approach to De Novo Generation of Antimicrobial Peptide Sequences through Integration of Protein Language Model and Diffusion Model

Abstract

Antimicrobial Peptides (AMPs) represent a promising class of antimicrobial agents crucial for combating antibiotic-resistant pathogens. Despite the emergence of deep learning approaches for AMP discovery, there remains a gap in efficiently generating novel AMPs across various amino acid lengths without prior knowledge of peptide structures or sequence alignments. Here we introduce ProT-Diff, a modularized and efficient deep generative approach that ingeniously combines a pre-trained protein language model with a diffusion model to de novo generate candidate AMP sequences. ProT-Diff enabled the rapid generation of thousands of AMPs with diverse lengths within hours. Following in silico screening based on physicochemical properties and predicted antimicrobial activities, we selected 35 peptides for experimental validation. Remarkably, 34 of these peptides demonstrated antimicrobial activity against Gram-positive or Gram-negative bacteria, with 6 exhibiting broad-spectrum efficacy. Of particular interest, AMP_2, one of the broad-spectrum peptides, displayed potent antimicrobial activity, low hemolysis, and minimal cytotoxicity. Further in vivo assessment revealed its high effectiveness against a clinically relevant drug-resistant E. coli strain in a mouse model of acute peritonitis. This study not only presents a viable generative strategy for novel AMP design but also underscores its potential for generating other functional peptides, thereby broadening the horizon for new drug development.