Adaptive Feature Selection With Hierarchical Learning for Drug-Target Interaction Prediction.

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions (DTIs) is crucial for advancing drug discovery and repurposing.
• Current deep learning methods for DTIs often focus narrowly on intermolecular associations, limiting representation learning and performance.
• Key features are not always leveraged effectively during interaction prediction, hindering further gains.

Purpose of the Study:

• To introduce ASHL-DTI, a novel framework designed to enhance DTI prediction.
• To improve feature quality and model generalizability by integrating hierarchical learning and adaptive feature selection.
• To overcome limitations of existing deep learning approaches in DTI prediction.

Main Methods:

• ASHL-DTI employs hierarchical learning to capture multi-level intramolecular associations for discriminative representation learning.
• An adaptive Top-k selection mechanism is incorporated to identify and retain the most predictive features.
• The framework facilitates effective interaction prediction between drugs and targets.

Main Results:

• ASHL-DTI demonstrated superior performance compared to state-of-the-art methods on multiple benchmark datasets.
• The framework achieved strong generalization capabilities in predicting novel drug-target pairs.
• Experimental results validate the effectiveness of the proposed hierarchical learning and adaptive feature selection strategies.

Conclusions:

• ASHL-DTI significantly enhances the accuracy and generalizability of drug-target interaction prediction.
• The framework holds considerable potential for accelerating drug discovery and repurposing efforts.
• The proposed approach offers a promising direction for future research in computational drug discovery.