An efficient hybrid deep learning architecture for predicting short antimicrobial peptides.

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Short antimicrobial peptides (AMPs) exhibit potent activity against diverse microbes.
• Developing novel AMPs is vital for antimicrobial drug and treatment innovation.
• Computational methods enhance AMP screening efficiency, but model improvements are needed.

Purpose of the Study:

• To propose and evaluate iAMP-DL, a hybrid deep learning model for predicting short AMPs.
• To compare iAMP-DL's performance against state-of-the-art computational methods.
• To assess the robustness and stability of the iAMP-DL framework.

Main Methods:

• Developed iAMP-DL, a hybrid deep learning architecture combining Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNNs).
• Utilized an independent test set for fair performance comparison with existing methods.
• Conducted 10 repeated experiments to evaluate model robustness and stability.

Main Results:

• iAMP-DL demonstrated superior performance compared to existing state-of-the-art methods.
• The model proved to be effective, robust, and stable in predicting short AMPs.
• Comparative analysis of negative data sampling methods confirmed the framework's general applicability for AMP prediction.

Conclusions:

• iAMP-DL is a highly effective and reliable framework for identifying promising short antimicrobial peptides.
• The proposed method advances computational approaches in antimicrobial peptide discovery.
• An accessible online web server was developed to support the research community in AMP identification.