Enhancing Cross-scale Feature Mutual Information via Heterogeneous Graph Contrastive Learning for Drug-Target Binding...

Area of Science:

• Computational chemistry
• Bioinformatics
• Machine learning in drug discovery

Background:

• Drug-target binding affinity prediction is essential for identifying potential drug candidates.
• Current graph neural network and self-supervised learning methods for drug-target interaction (DTI) prediction have limitations.
• Existing approaches suffer from low information density, rigid cross-scale feature integration, and biased contrastive learning strategies.

Purpose of the Study:

• To propose a novel method, EMHGCL-DTA, to address the limitations of existing drug-target binding affinity prediction models.
• To enhance the quality of representations by incorporating cross-scale mutual information and unbiased feature augmentation.
• To improve the accuracy and reliability of drug-target interaction predictions.

Main Methods:

• Developed EMHGCL-DTA, a model that reconstructs the relationship network on a macro scale, integrating molecular structure similarity information.
• Established a connection between macro- and micro-scale features using similarity derived from drug and target molecular structures.
• Introduced an unbiased feature augmentation strategy to improve drug and target representations within a heterogeneous graph contrastive learning framework.

Main Results:

• The EMHGCL-DTA model demonstrated superior performance in predicting drug-target binding affinity on two benchmark datasets.
• The proposed method effectively integrates cross-scale features and mitigates biases associated with traditional graph contrastive learning.
• Enhanced information density and representation quality were achieved through the novel approach.

Conclusions:

• EMHGCL-DTA offers a significant advancement in predicting drug-target binding affinity.
• The method's ability to leverage heterogeneous graph contrastive learning and cross-scale information improves prediction accuracy.
• This approach holds promise for accelerating the drug discovery process by providing more reliable interaction predictions.