Drug-Target Interaction Prediction via Graph Auto-Encoder and Multi-Subspace Deep Neural Networks

Area of Science:

• Computational biology
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-target interaction (DTI) prediction is crucial for accelerating new drug discovery.
• Deep neural networks (DNNs) are commonly used for DTI prediction but face challenges with limited data and high dimensionality.
• Existing DNNs may suffer from insufficient parameter training and underutilization of data features.

Purpose of the Study:

• To develop an advanced computational model for more effective drug-target interaction prediction.
• To overcome the limitations of traditional DNNs in handling high-dimensional and sparse DTI datasets.
• To enhance feature extraction and model training for improved DTI prediction accuracy.

Main Methods:

• A novel Graph Auto-Encoder and Multi-Subspace Deep Neural Network (GAEMSDNN) was designed.
• The model integrates a graph auto-encoder for preserving reconstruction information.
• It incorporates a subspace layer for extracting diverse feature subsets and an ensemble layer for unified optimization.

Main Results:

• The GAEMSDNN demonstrated significantly improved performance compared to existing methods in DTI prediction.
• The model effectively extracts more features from the network input.
• Enhanced training of the DNN network was achieved, leading to better predictive power.

Conclusions:

• The proposed GAEMSDNN model offers a superior strategy for computational drug-target interaction prediction.
• The integration of graph auto-encoder, subspace, and ensemble layers effectively enhances learning capabilities.
• The findings validate the effectiveness of the developed strategies for improving DTI prediction accuracy in drug discovery.