Antimicrobial peptide prediction based on contrastive learning and gated convolutional neural network.

Area of Science:

• Biochemistry and Molecular Biology
• Computational Biology and Bioinformatics
• Microbiology and Infectious Diseases

Background:

• Excessive antibiotic use fuels antibiotic resistance and disrupts beneficial microbial communities.
• Antimicrobial peptides (AMPs) are a promising alternative to conventional antibiotics.
• Computational methods accelerate AMP discovery, reducing time and cost.

Purpose of the Study:

• To develop an efficient deep learning framework, CG-AMP, for identifying antimicrobial peptides (AMPs).
• To leverage multimodal features by integrating pre-trained language models, contrastive learning, and Convolutional Neural Networks (CNNs).
• To enhance the accuracy and efficiency of AMP identification.

Main Methods:

• Proposed CG-AMP, a dual-module deep learning framework.
• Module 1: Feature representation learning using a pre-trained language model and contrastive learning.
• Module 2: Enhanced Convolutional Neural Network (CNN) for efficient feature extraction.

Main Results:

• CG-AMP demonstrated high reliability in AMP identification.
• Achieved 0.9497 accuracy and 0.9508 F1 score on the AMPlify test set.
• Achieved 0.9403 accuracy and 0.9392 F1 score on the DAMP test set, outperforming existing methods.

Conclusions:

• CG-AMP is a robust and efficient tool for identifying antimicrobial peptides.
• The dual-module architecture effectively integrates multimodal features for improved performance.
• CG-AMP offers a valuable computational approach to accelerate the discovery of novel AMPs.