GraphDTI: A robust deep learning predictor of drug-target interactions from multiple heterogeneous data

Area of Science:

• Computational Biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions are complex, involving single protein binding and system-level signal transduction.
• Existing methods struggle with large, heterogeneous data like gene expression and protein-protein interactions.

Purpose of the Study:

• To develop a robust machine learning framework, GraphDTI, for accurate drug-target interaction identification.
• To improve the state-of-the-art in predicting drug mechanisms of action and system-level effects.

Main Methods:

• GraphDTI integrates molecular data (drugs, proteins, binding sites) with system-level data (gene expression, protein-protein interactions).
• A high-quality dataset and cluster-based cross-validation were used for performance evaluation.

Main Results:

• GraphDTI achieved an AUC of 0.996 on the validation dataset and 0.939 on unseen data.
• The framework significantly outperformed existing drug-target interaction predictors.

Conclusions:

• GraphDTI provides a robust and accurate method for identifying drug-target interactions.
• Applications include investigating drug polypharmacology, side effects, and drug repositioning.