GraphormerDTI: A graph transformer-based approach for drug-target interaction prediction.

Area of Science:

• Computational chemistry
• Artificial intelligence in drug discovery
• Machine learning for bioinformatics

Background:

• Traditional drug discovery is time-consuming and expensive.
• Artificial Intelligence (AI) accelerates drug screening by predicting drug-target interactions (DTIs).
• Existing AI methods struggle with limited labeled DTI data and underutilize molecular structural information.

Purpose of the Study:

• To develop an advanced machine learning approach for accurate DTI prediction.
• To improve the generalization capability of AI models for out-of-sample molecules.
• To leverage molecular topological structures for enhanced DTI prediction.

Main Methods:

• Proposed GraphormerDTI, utilizing a Graph Transformer neural network to model molecular structures.
• Employed iterative Transformer-based message passing for embedding molecular graphs.
• Integrated Graph Transformer with 1D-Convolutional Neural Network (1D-CNN) and attention mechanisms for DTI prediction.

Main Results:

• GraphormerDTI demonstrated superior performance in out-of-molecule DTI prediction across three benchmark datasets.
• Achieved exceptional performance, outperforming five state-of-the-art baselines.
• Performed comparably to the best baseline for transductive DTI prediction.

Conclusions:

• GraphormerDTI effectively models molecular structures, enabling accurate DTI prediction.
• The approach shows strong generalization capabilities for novel drug molecules.
• GraphormerDTI offers a powerful tool for accelerating AI-driven drug discovery.