Semi-inductive dataset construction and framework optimization for practical drug target interaction prediction with...

Area of Science:

• Computational Biology
• Pharmacology
• Artificial Intelligence

Background:

• Deep learning methods for drug-target interaction (DTI) prediction show promise but face limitations in data diversity and model complexity.
• Existing benchmarks often lack sufficient data volume and balance, hindering real-world applicability.

Purpose of the Study:

• To develop a unified framework, SCOPE-DTI, that addresses limitations in DTI prediction by integrating a large-scale, balanced dataset with advanced deep learning models.
• To enhance the accuracy and efficiency of DTI prediction for accelerating drug discovery.

Main Methods:

• Constructed SCOPE-DTI from 13 public repositories, creating a dataset up to 100-fold larger than common benchmarks.
• Employed a SCOPE model integrating 3D protein and compound representations, graph neural networks, and bilinear attention mechanisms.
• Developed a user-friendly interface and database for SCOPE-DTI.

Main Results:

• The SCOPE model demonstrated superior performance across various DTI prediction tasks, outperforming state-of-the-art methods.
• Experimental validation identified anticancer targets for two bioactive natural compounds, confirming the framework's effectiveness.
• The integrated dataset and model significantly improved the prediction of cross-domain interaction patterns.

Conclusions:

• SCOPE-DTI offers a comprehensive solution for DTI prediction by providing extensive data, sophisticated modeling, and accessible tools.
• The framework effectively accelerates drug discovery research by overcoming data scarcity and model complexity challenges.
• This work facilitates the identification of novel drug targets and the development of new therapeutics.