Efficient Deep Model Ensemble Framework for Drug-Target Interaction Prediction.

Area of Science:

• Computational chemistry
• Pharmacology
• Bioinformatics

Background:

• Accurate prediction of drug-target interactions (DTI) is essential for efficient drug development.
• Existing deep learning models for DTI prediction face challenges with performance and false negatives.

Purpose of the Study:

• To propose EADTN, a novel ensemble model for enhanced DTI prediction.
• To improve the reliability and interpretability of DTI prediction models.
• To validate the model's effectiveness through experimental drug-target interaction identification.

Main Methods:

• Developed EADTN, an ensemble model featuring innovative feature adaptation for local weight extraction.
• Implemented clustering-enhanced parameter fine-tuning to mitigate false negatives.
• Utilized Shapley value-based analysis for identifying key drug substructures and enhancing model interpretability.

Main Results:

• EADTN demonstrated superior performance compared to state-of-the-art models across diverse datasets.
• The model successfully identified potential interactions between NQO1 targets and drugs SIRT-IN-1 and LY2183240.
• Wet-lab experiments validated the predicted drug-target interactions, confirming EADTN's reliability.

Conclusions:

• EADTN offers a significant advancement in DTI prediction, enhancing accuracy and reducing false negatives.
• The model's interpretability features aid in understanding drug mechanisms and substructure importance.
• EADTN shows promise for applications in drug repositioning and accelerating drug discovery pipelines.