A unified framework combining linear and 3D molecular features for robust drug-protein interaction prediction

Area of Science:

• Computational Biology
• Drug Discovery
• Bioinformatics

Background:

• Accurate prediction of drug-protein interactions (DPIs) is crucial for drug discovery.
• Existing methods often rely heavily on 3D structures, limiting their applicability.
• There is a need for methods that integrate diverse molecular features for robust DPI prediction.

Purpose of the Study:

• To develop PointDPI, a novel computational method for predicting drug-protein interactions (DPIs).
• To leverage both linear and 3D structural features of drugs and proteins for enhanced prediction accuracy.
• To identify key regulatory sites involved in DPIs and validate predicted interactions experimentally.

Main Methods:

• Developed PointDPI, a model that simultaneously utilizes linear and 3D molecular features.
• Employed feature alignment for stereoscopic recognition of molecular properties, reducing reliance on complete 3D structures.
• Preserved local topological relationships to prevent data distortion during analysis.
• Utilized model gradients to predict critical regulatory sites.

Main Results:

• PointDPI demonstrated superior performance compared to several state-of-the-art (SOTA) methods.
• Achieved increased accuracy, particularly in predicting interactions involving previously unseen molecules.
• Successfully predicted four drug-protein interactions (DPIs) that were subsequently validated experimentally at both mRNA and protein levels.

Conclusions:

• PointDPI offers a robust and accurate approach for predicting drug-protein interactions.
• The method's ability to handle unseen molecules and integrate diverse features holds significant promise for drug discovery.
• Experimental validation confirmed the therapeutic potential of predicted interactions for inflammatory, neurological diseases, and neuroprotection.