DMAMP: A Deep-Learning Model for Detecting Antimicrobial Peptides and Their Multi-Activities.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Antimicrobial peptides (AMPs) are crucial in drug discovery due to their broad-spectrum antibacterial activity.
• Experimental detection of AMPs and their activities is costly; computational methods offer a more economical alternative.
• Existing computational methods often treat AMP identification and activity prediction as independent tasks, neglecting potential synergistic relationships.

Purpose of the Study:

• To develop a novel deep learning model, DMAMP, for simultaneous identification of AMPs and their activities.
• To address the limitation of independent task processing in current computational approaches.
• To leverage multi-task learning for improved feature sharing and extraction.

Main Methods:

• DMAMP employs a multi-task learning framework utilizing convolutional neural networks and residual blocks for shared feature extraction.
• Fully connected layers are used to learn task-specific information.
• Original evolutionary features are incorporated to enhance the prediction of peptide activities.

Main Results:

• DMAMP demonstrated superior performance compared to single-task models, achieving a 4.28% higher Matthews Correlation Coefficient (MCC) on AMP identification.
• The model achieved a higher average MCC for predicting five different activities compared to single-task models and existing methods.
• Feature visualization confirmed the model's ability to capture class differences, and it showed potential in identifying peptides active against SARS-CoV-2.

Conclusions:

• DMAMP offers an effective multi-task learning approach for simultaneous AMP identification and activity prediction.
• The model's ability to share and learn distinct features improves prediction accuracy.
• This method provides a promising tool for accelerating the discovery of novel antimicrobial and antiviral peptide drugs.