Reinforced Metapath Optimization in Heterogeneous Information Networks for Drug-Target Interaction Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Graph neural networks (GNNs) are effective for drug-target interaction (DTI) prediction.
• Heterogeneous information networks (HINs) with metapaths enhance DTI prediction performance.
• Existing methods struggle with fixed metapath selection and limited node information exploitation.

Purpose of the Study:

• To develop a novel method for DTI prediction by optimizing metapaths in HINs.
• To overcome limitations of fixed metapath selection and insufficient node information utilization.
• To enhance the performance of GNN-based DTI prediction models.

Main Methods:

• Metapath optimization formulated as a Markov decision process with reinforcement learning.
• Iterative training of a reinforcement learning agent to learn high-quality metapaths.
• Construction of node-based subgraphs along metapaths, processed by graph convolutional neural networks at varying depths.

Main Results:

• The proposed method achieved significant advantages over traditional DTI prediction approaches.
• Experimental validation on standard heterogeneous biological benchmark datasets confirmed effectiveness.
• Learned metapaths enhanced downstream network performance, leading to improved predictions.

Conclusions:

• Optimizing metapaths dynamically improves GNN-based DTI prediction.
• Leveraging node information through subgraph construction enhances model performance.
• The novel approach offers a more effective strategy for predicting drug-target interactions.