iAFPs-Mv-BiTCN: Predicting antifungal peptides using self-attention transformer embedding and transform evolutionary...

Area of Science:

• Computational biology and bioinformatics
• Drug discovery and development
• Mycology and infectious diseases

Background:

• Fungal infections pose a significant global health threat, exacerbated by factors like COVID-19.
• Current antifungal drug development is costly, time-consuming, and can cause adverse effects.
• Peptide therapeutics show promise due to their high specificity and targeted action.

Purpose of the Study:

• To develop a highly accurate computational model for predicting antifungal peptides.
• To overcome limitations of existing prediction methods by integrating diverse feature encoding techniques.
• To provide a valuable tool for accelerating the discovery of novel antifungal peptide therapeutics.

Main Methods:

• Utilized word embedding (skip-gram) and transformer-based bidirectional encoder representations for peptide encoding.
• Integrated transform-based evolutionary features using Pseudo Position-Specific Scoring Matrix with Discrete Wavelet Transform (PsePSSM-DWT).
• Employed Shapley Additive exPlanations (SHAP) for optimal feature selection and trained the model using a bi-directional temporal convolutional network (BiTCN).

Main Results:

• The iAFPs-Mv-BiTCN model achieved high predictive accuracy (98.15% on training data, 94.11% on independent data) and AUC (0.99 on training, 0.98 on independent).
• Demonstrated superior performance compared to existing models, with approximately 4-5% higher accuracy.
• The model's feature selection approach reduced training costs effectively.

Conclusions:

• The iAFPs-Mv-BiTCN model is a reliable and effective tool for predicting antifungal peptides.
• This computational approach can significantly aid pharmaceutical design and accelerate antifungal drug discovery.
• The study highlights the potential of integrated computational methods in addressing critical health challenges like fungal infections.