Predicting drug-protein interactions by preserving the graph information of multi source data

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Predicting drug-target interactions (DTIs) is crucial for drug discovery and repurposing.
• Existing computational methods often analyze network connections or topology in isolation, limiting prediction accuracy.
• A comprehensive approach is needed to leverage both node connections and network structure for accurate DTI prediction.

Purpose of the Study:

• To develop a novel computational method, TTGCN, for accurate prediction of drug-target interactions (DTIs).
• To improve upon existing methods by integrating heterogeneous graph convolutional neural networks (GCN) and graph attention networks (GAT).
• To enhance DTI prediction by considering both network topology and node features.

Main Methods:

• Proposed TTGCN, a method combining graph attention networks (GAT) and residual graph convolutional networks (R-GCN) for feature extraction.
• Employed a two-tiered feature learning strategy to extract drug and target embeddings from a heterogeneous network.
• Utilized inductive matrix completion for DTI prediction, preserving network node connectivity and topological structure.

Main Results:

• TTGCN demonstrated superior performance in predicting DTIs compared to existing methods.
• Achieved higher area under the curve (AUC) and area under the precision-recall curve (AUPRC) in experimental evaluations.
• Case studies validated TTGCN's capability in identifying potential drug-target interactions.

Conclusions:

• The proposed TTGCN method offers a significant advancement in computational DTI prediction.
• Integrating GAT and R-GCN effectively captures complex network features for improved accuracy.
• TTGCN shows promise for accelerating drug discovery and repurposing efforts.