Inferring Drug-Target Interactions Based on Random Walk and Convolutional Neural Network.

Area of Science:

• Computational drug discovery
• Bioinformatics
• Machine learning in pharmacology

Background:

• Accurate prediction of drug-target interactions (DTIs) is crucial for efficient drug discovery.
• Existing methods struggle to deeply integrate heterogeneous data and learn comprehensive feature representations.
• There is a need for advanced computational strategies to enhance DTI prediction accuracy and reduce development costs.

Purpose of the Study:

• To develop an advanced computational method, DTIPred, for predicting drug-target interactions (DTIs).
• To deeply integrate heterogeneous data related to drugs and proteins, including similarities and interactions.
• To improve the accuracy and efficiency of identifying novel drug-target relationships.

Main Methods:

• Constructed a heterogeneous network integrating drugs, proteins, drug side effects, similarities, and associations.
• Developed DTIPred, a method combining random walk with restart and convolutional neural networks.
• Employed a two-sided model to learn deep feature representations from topological and original data.

Main Results:

• DTIPred demonstrated superior prediction performance compared to state-of-the-art methods in cross-validation experiments.
• The method successfully retrieved a higher number of actual drug-protein interactions in top predictions.
• Case studies confirmed DTIPred's capability in discovering potential drug-protein interactions for specific drugs.

Conclusions:

• DTIPred offers a powerful approach for identifying novel drug-target interactions by leveraging heterogeneous data and deep learning.
• The method enhances drug discovery pipelines by providing more accurate and relevant predictions.
• DTIPred's ability to uncover potential interactions can significantly aid biologists and accelerate drug development.