A separable temporal convolutional networks based deep learning technique for discovering antiviral medicines.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• The COVID-19 pandemic highlighted the urgent need for rapid therapeutic drug discovery.
• Artificial intelligence (AI) is increasingly utilized for in silico screening of therapeutic molecules.
• Antiviral peptides (AVPs) are natural defense mechanisms produced by organisms against viral infections.

Purpose of the Study:

• To develop and validate AI-driven models for the discovery of novel antiviral peptides (AVPs).
• To create a user-friendly web application for identifying AVPs in proteomes.
• To computationally screen natural defense proteins for potential AVPs.

Main Methods:

• Development of the Deep-AVPiden model using deep learning for AVP identification.
• Creation of a computationally efficient variant, Deep-AVPiden (DS), employing point-wise separable convolutions.
• Deployment of a web application for public access to the Deep-AVPiden model.
• Statistical comparison of model performance against state-of-the-art classifiers using Student's t-test.

Main Results:

• Deep-AVPiden achieved 90% accuracy and 90% precision; Deep-AVPiden (DS) achieved 88% accuracy and 90% precision.
• Both models significantly outperformed existing state-of-the-art classifiers.
• Identified potential AVPs in plant, mammal, and fish defense proteins with sequence similarity to known antimicrobial peptides.

Conclusions:

• The Deep-AVPiden models offer a powerful computational approach for discovering novel AVPs.
• The identified AVPs warrant further experimental synthesis and testing for antiviral efficacy.
• This AI-driven discovery platform can accelerate the development of new antiviral therapies.