Designing antimicrobial peptides using deep learning and molecular dynamic simulations.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Multidrug-resistant bacteria pose a significant threat to public health.
• Traditional methods for discovering antimicrobial peptides (AMPs) are slow and expensive.
• Deep learning offers a potential solution for efficient AMP design and classification.

Purpose of the Study:

• To utilize deep learning, specifically natural language processing models, for the de novo design and identification of novel AMPs.
• To screen and validate candidate AMPs using advanced computational and experimental techniques.
• To investigate the structure-activity relationships of novel AMPs to enhance their efficacy.

Main Methods:

• Combined natural language processing models: sequence generative adversarial nets, bidirectional encoder representations from transformers, and multilayer perceptron for AMP design.
• Screened six candidate AMPs using AlphaFold2 structure prediction and molecular dynamic simulations.
• Confirmed structure and activity of a lead peptide (A-222) using nuclear magnetic resonance and performed structure-activity relationship studies on analogs.

Main Results:

• Identified a novel class of AMPs with low homology to known peptides.
• The peptide A-222 demonstrated broad-spectrum inhibition against gram-positive and gram-negative bacteria.
• Analogs of A-222 showed a 4-8 fold increase in activity against Stenotrophomonas maltophilia and Pseudomonas aeruginosa.

Conclusions:

• Deep learning models are effective in accelerating the discovery of novel AMPs.
• The identified novel AMPs, particularly A-222 and its analogs, represent a promising new class of therapeutics against resistant bacteria.
• This approach holds significant potential for developing next-generation antimicrobial agents.