dsAMP and dsAMPGAN: Deep Learning Networks for Antimicrobial Peptides Recognition and Generation

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Antibiotic resistance poses a significant global health threat.
• Antimicrobial peptides (AMPs) offer a promising alternative due to their non-specific mechanisms, reducing resistance development.
• Computational methods, particularly deep learning, are increasingly vital for identifying and designing new AMPs.

Purpose of the Study:

• To develop a novel deep learning framework for the classification, function prediction, and generation of AMPs.
• To create a robust AMP predictor and a generative model for novel AMP candidate synthesis.
• To address the urgent need for new therapeutic agents against antibiotic-resistant pathogens.

Main Methods:

• Development of discoverAMP (dsAMP), a Convolutional Neural Network (CNN) Attention Bidirectional Long Short-Term Memory (BiLSTM) model incorporating transfer learning for AMP classification.
• Implementation of dsAMPGAN, a Generative Adversarial Network (GAN), for the de novo generation of AMP candidates.
• Rigorous evaluation of dsAMP's performance using metrics like sensitivity, specificity, accuracy, and AUC, and assessment of dsAMPGAN-generated peptides' properties.

Main Results:

• The dsAMP model demonstrated superior performance compared to existing classifiers, achieving over 95% classification accuracy with transfer learning.
• dsAMP exhibited high sensitivity, specificity, Matthew correlation coefficient, precision, F1 score, and AUC.
• dsAMPGAN successfully generated novel AMP candidates with physical and chemical properties comparable to natural AMPs.

Conclusions:

• The developed deep learning framework, dsAMP and dsAMPGAN, provides a reliable and effective tool for identifying and designing novel AMPs.
• These computational tools can accelerate the discovery of new AMPs for clinical applications, aiding in the fight against antibiotic resistance.
• The study highlights the potential of deep learning in advancing peptide-based drug discovery and combating antimicrobial resistance.