AMP-CapsNet: a multi-view feature fusion approach for antimicrobial peptide prediction using capsule networks

Area of Science:

• Biochemistry
• Computational Biology
• Pharmacology

Background:

• Antimicrobial peptides (AMPs) are crucial components of the immune system with significant potential against antibiotic-resistant bacteria.
• The rise of antibiotic resistance necessitates novel therapeutic strategies, making AMPs a key focus for drug development.
• Diabetic foot infections and other medical issues highlight the need for effective antibacterial treatments.

Purpose of the Study:

• To introduce and evaluate a novel deep learning approach, AMP-CapsNet, for precise prediction of antimicrobial peptides.
• To compare the efficacy of AMP-CapsNet against existing deep learning and baseline models for AMP identification.
• To leverage advanced computational methods for accelerating the discovery of new AMPs for therapeutic applications.

Main Methods:

• Development of a novel deep learning model, antimicrobial peptides using Capsule Neural Network (AMP-CapsNet).
• Utilization of Amino Acid Composition (AAC) and dipeptide composition (DPC) for feature encoding.
• Independent cross-validation and external testing of all models to assess performance.

Main Results:

• The AMP-CapsNet model achieved a high accuracy of 97.29% and an AUC score of 98.91% on the test set using dipeptide composition (DPC).
• AMP-CapsNet demonstrated superior performance compared to other deep learning and baseline models.
• The model using AAC achieved an accuracy of 84.42%, indicating the effectiveness of DPC for this task.

Conclusions:

• The proposed AMP-CapsNet model significantly enhances the accuracy of antimicrobial peptide prediction.
• This deep learning approach advances AMP drug discovery, paving the way for new medication development against resistant bacteria.
• The findings have implications for processing biological data and computational pharmacology.