AFP-MFL: accurate identification of antifungal peptides using multi-view feature learning

Area of Science:

• Biochemistry
• Computational Biology
• Drug Discovery

Background:

• Peptide-based drugs show promise as antifungal agents due to high efficacy and low toxicity.
• Experimental identification of antifungal peptides (AFPs) is costly and time-consuming.
• Accurate computational prediction of AFPs is crucial for efficient drug development.

Purpose of the Study:

• To develop a novel deep learning model, AFP-MFL, for predicting antifungal peptides (AFPs) based solely on amino acid sequences.
• To enhance the speed and accuracy of identifying potential antifungal peptide drug candidates.

Main Methods:

• Developed AFP-MFL, a deep learning model utilizing peptide sequences without structural information.
• Integrated contextual semantic information from protein language models, evolutionary data, and physicochemical properties.
• Employed a co-attention mechanism for feature integration and the SHAP method for feature importance analysis.

Main Results:

• AFP-MFL demonstrated superior performance compared to existing state-of-the-art models across four independent test datasets.
• The model effectively leverages sequence-derived features for accurate AFP prediction.
• Feature analysis using SHAP provided insights into prediction drivers.

Conclusions:

• AFP-MFL offers a powerful and efficient computational tool for rapid screening and identification of novel antifungal peptides.
• The developed web server provides accessible functionality for researchers in antifungal drug discovery.
• This approach significantly advances the development pipeline for peptide-based antifungal therapeutics.