Explainable deep learning and virtual evolution identifies antimicrobial peptides with activity against...

Area of Science:

• Biotechnology
• Computational Biology
• Infectious Diseases

Background:

• Antimicrobial resistance (AMR) is a growing global health threat.
• Novel antimicrobial compounds are urgently needed to combat resistant pathogens.
• Artificial intelligence (AI) offers a powerful approach for drug discovery.

Purpose of the Study:

• To develop an AI-driven platform for identifying and optimizing antimicrobial peptides (AMPs).
• To apply this platform to discover novel AMPs from under-explored microbial sources.
• To validate the efficacy of computationally designed AMPs against multidrug-resistant (MDR) bacteria.

Main Methods:

• Developed EvoGradient, an explainable deep learning model for predicting AMP potency and guiding sequence modification.
• Applied EvoGradient to virtually evolve peptides from low-abundance human oral bacteria.
• Synthesized and experimentally validated top computationally designed AMP candidates against a panel of MDR pathogens.
• Conducted in vivo studies using mouse models to assess the therapeutic potential of the most potent AMP.

Main Results:

• EvoGradient successfully identified 32 potent AMP candidates through in silico directed evolution.
• Six synthesized AMPs demonstrated significant activity against carbapenem-resistant Enterobacteriaceae (e.g., E. coli, K. pneumoniae), Acinetobacter baumannii, and vancomycin-resistant Enterococcus faecium.
• The lead compound, pep-19-mod, achieved >95% bacterial load reduction in vivo in mouse thigh infection models via systemic and local administration.

Conclusions:

• AI-driven in silico directed evolution is an effective strategy for discovering and optimizing novel AMPs.
• The developed EvoGradient platform can accelerate the identification of promising antimicrobial drug candidates.
• This approach holds significant potential for addressing the challenge of antimicrobial resistance.