DRaW

DRaW predicts virus-antiviral drug-target interactions using a convolutional neural network to support antiviral drug repurposing.

Key Features:

• Convolutional neural network: Implements a CNN architecture as the core predictive model.
• Prediction target: Specifically predicts drug-target interactions (DTIs) focused on virus-antiviral interactions (VAIs).
• Feature-vector representation: Uses feature vectors instead of matrix factorization for representation learning.
• Addresses matrix factorization limitations: Targets issues of data sparsity and fixed matrix-size paradigms associated with matrix factorization methods.
• Comparative benchmarking: Compared against several matrix factorization methods and other deep models.
• Dataset evaluation: Evaluated on three COVID-19 datasets and additional benchmark datasets.
• Docking validation: Validates top-ranked drug recommendations using molecular docking studies.
• Performance claim: Demonstrated superior predictive performance relative to matrix factorization approaches and existing deep models.

Scientific Applications:

• DTI prediction: Predicts novel drug-target interactions between antivirals and viral proteins.
• Drug repurposing: Prioritizes existing approved antivirals for potential new indications.
• COVID-19 candidate identification: Ranks antiviral candidates for SARS-CoV-2 using dedicated COVID-19 datasets.
• Pre-docking prioritization: Provides ranked candidates for downstream molecular docking validation.

Methodology:

Uses a convolutional neural network trained on feature-vector representations (avoiding matrix factorization), compared against matrix factorization methods and other deep models on three COVID-19 datasets and benchmark datasets, with top predictions validated by docking studies.