Deep Learning for Novel Antimicrobial Peptide Design.

Area of Science:

• Biotechnology
• Computational Biology
• Infectious Disease

Background:

• Antimicrobial resistance (AMR) is a growing global health threat, exacerbated by increased antibiotic use during the COVID-19 pandemic.
• There is an urgent need for new antibiotics, particularly against Gram-negative pathogens such as Escherichia coli, as existing treatments become less effective.
• Antimicrobial peptides (AMPs) represent a promising avenue for developing novel therapeutic agents.

Purpose of the Study:

• To design short, novel antimicrobial peptide sequences with potential activity against Escherichia coli using deep learning.
• To evaluate the efficacy of long short-term memory (LSTM) generative and bidirectional LSTM classification models for AMP design.
• To explore the potential of artificial intelligence in accelerating the discovery of new antibiotic leads.

Main Methods:

• Construction and optimization of LSTM generative and bidirectional LSTM classification models using Bayesian hyperparameter optimization.
• Generation of novel peptide sequences using the trained models.
• Classification of generated sequences as antimicrobial or non-antimicrobial.
• Analysis of predicted 3D conformations of selected AMPs.

Main Results:

• Classification models achieved validation accuracies ranging from 81.6% to 88.9%.
• Novel AMPs were successfully classified as antimicrobial with accuracies between 70.6% and 91.7%.
• Predicted structures of selected AMPs revealed characteristic alpha-helical and amphipathic properties.

Conclusions:

• Long short-term memory (LSTM) networks are effective tools for generating novel antimicrobial peptides.
• Deep learning approaches can significantly aid in the search for new antibiotic candidates against challenging bacteria like E. coli.
• This study highlights the potential of AI-driven strategies to address the critical challenge of antimicrobial resistance.