D3AI-CoV

D3AI-CoV predicts drug-target interactions and performs virtual screening using deep learning to identify potential therapeutic compounds against SARS-CoV-2.

Key Features:

• Deep Learning-Based Models: Incorporates three models—MultiDTI, MPNNs-CNN, and MPNNs-CNN-R—for predicting drug-target interactions.
• Target Prediction: MultiDTI and MPNNs-CNN achieve areas under the receiver operating characteristic curves (AUCs) of 0.93 and 0.91, respectively, compared with reported AUCs of 0.51–0.74 for other approaches.
• Virtual Screening: Enhances virtual screening by improving hit rate relative to conventional methods.
• External Validation: Predictive performance was evaluated on the external test set Test-78, containing 39 newly published active compounds and 39 inactive compounds from DrugBank.

Scientific Applications:

• Anti-COVID-19 drug discovery: Identification of potential therapeutic compounds and targets for SARS-CoV-2.
• Hit discovery and lead optimization: Prioritization of compounds for downstream experimental validation and optimization.
• Drug repurposing: Screening of existing compounds, including DrugBank entries, for potential activity against SARS-CoV-2.
• Mechanism-of-action elucidation: Support for hypothesizing drug-target interactions to aid interpretation of therapeutic mechanisms.

Methodology:

Applies deep learning models: MultiDTI integrates multiple data types for drug-target interaction prediction, while MPNNs-CNN and MPNNs-CNN-R use message-passing/graph neural network architectures combined with convolutional neural networks to model molecular structures; model performance was evaluated on the external Test-78 dataset (39 active, 39 inactive from DrugBank).