Prediction of Drug-Target Interactions With High- Quality Negative Samples and a Network-Based Deep Learning...

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Accurate drug-target interaction (DTI) identification is vital for efficient drug discovery.
• Computational methods accelerate DTI prediction, reducing time and cost compared to experiments.
• Existing machine learning approaches for DTI prediction suffer from poor negative sample quality and ineffective data integration.

Purpose of the Study:

• To address the challenge of high-quality negative sample selection in DTI prediction.
• To develop an advanced computational framework for accurate DTI prediction.
• To integrate diverse data sources, including network topology and biological annotations, for improved prediction performance.

Main Methods:

• A novel negative sample selection strategy based on complex network theory was developed to mitigate biases in unlabeled data.
• A new DTI prediction framework, HNetPa-DTI, was proposed, integrating drug-protein-disease network topology with protein Gene Ontology (GO) and pathway annotations.
• Heterogeneous graph neural networks were employed to extract topological features from a heterogeneous network, while graph neural networks processed GO and pathway information via various network constructions.

Main Results:

• The proposed negative sample selection method effectively addresses biases present in random selection.
• HNetPa-DTI demonstrated superior performance compared to existing baseline methods across four distinct prediction tasks.
• The integration of heterogeneous network topology and protein annotation data significantly enhances DTI prediction accuracy.

Conclusions:

• The developed negative sample selection approach improves the reliability of computational DTI prediction.
• HNetPa-DTI represents a significant advancement in DTI prediction by effectively leveraging multisource information.
• This study provides a robust computational framework that can accelerate the identification of potential drug candidates in drug discovery pipelines.