MHADTI

MHADTI predicts drug-target interactions by embedding multiview heterogeneous information networks with hierarchical attention mechanisms to generate deep drug and target representations for DTI prediction.

Key Features:

• Multiview Heterogeneous Information Networks: Constructs multiple drug and target similarity networks from diverse data sources and integrates known DTIs to build three distinct drug-target HIN views.
• Hierarchical Attention Mechanisms: Employs hierarchical attention at node-level, semantic-level (meta-paths), and graph-level to weight nodes, meta-paths, and entire HINs.
• Deep Feature Representation: Learns comprehensive embeddings for drugs and targets that capture structural and semantic information across the multiview HINs.
• Multilayer Perceptron (MLP) Integration: Uses a multilayer perceptron on the learned embeddings to predict potential drug-target interactions.

Scientific Applications:

• Novel DTI discovery: Predicts potential novel drug-target interactions to aid identification of new therapeutic relationships.
• Drug repositioning and side-effect identification: Profiles drug-target interaction patterns to support drug repositioning efforts and identification of possible side effects.

Methodology:

Constructs multiview HINs from multisource similarity information, applies hierarchical attention (node-, semantic/meta-path-, and graph-level) to learn drug and target embeddings, and employs a multilayer perceptron for DTI prediction.