DeepMGT-DTI

DeepMGT-DTI predicts drug–target interactions by integrating multilayer graph representations of drug molecular structure with target sequence residue features using transformer and convolutional neural network architectures.

Key Features:

• Transformer network: Integrates multilayer graph information to capture global and atom-level features of drug molecular structures.
• Multilayer graph representation: Encodes intra-molecular relationships and multi-layer structural attributes of compounds for model input.
• Convolutional Neural Network (CNN): Captures local residue information within target protein sequences.
• Feature fusion across GCN layers: Fuses feature information between layers in a graph convolutional neural network to extract comprehensive interaction-relevant representations.

Scientific Applications:

• Drug discovery: Prioritizes candidate drug–target interactions by combining detailed molecular and sequence features.
• Drug repositioning: Identifies potential therapeutic candidates for COVID-19 and Alzheimer's disease through DTI predictions.

Methodology:

Combines a transformer network applied to multilayer molecular graphs, a graph convolutional neural network with inter-layer feature fusion, and a CNN for target residue features; evaluated on the DrugBank dataset.