Geometry-Enhanced Multiscale Joint Representation Learning for Drug-Target Interaction Prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are crucial for drug efficacy and development.
• Current DTI prediction methods often neglect spatial and network information, limiting accuracy.
• Integrating molecular structure and network data is essential for improved DTI prediction.

Purpose of the Study:

• To propose a novel Geometry-enhanced Multiscale Joint Representation Learning (GMJRL) method for accurate DTI prediction.
• To address limitations of existing methods by incorporating both macro-scale network and micro-scale geometric information.
• To develop an effective fusion strategy for multiscale representations in DTI prediction.

Main Methods:

• GMJRL extracts global network information (macro-scale) and geometric details (micro-scale) from drugs and targets.
• Incorporates drug bond angle and target atomic coordinate information.
• Employs a self-attention-based joint representation learning for effective fusion of multiscale data.
• Utilizes a negative sampling algorithm for reliable negative sample selection.

Main Results:

• GMJRL effectively integrates macro-scale network and micro-scale geometric information.
• The self-attention mechanism successfully fuses different scale representations.
• The negative sampling algorithm enhances the reliability of training data.
• Extensive experiments demonstrate GMJRL's promising performance in DTI prediction.

Conclusions:

• GMJRL offers a novel and effective approach for predicting drug-target interactions.
• The method's ability to integrate multiscale information improves prediction accuracy.
• GMJRL has the potential to accelerate the drug development process by reducing experimental screening costs.