Revolutionizing GPCR-ligand predictions: DeepGPCR with experimental validation for high-precision drug discovery

Area of Science:

• Computational chemistry and structural biology
• Drug discovery and medicinal chemistry

Background:

• G-protein coupled receptors (GPCRs) are vital drug targets, but their membrane-bound nature complicates structure determination and hinders traditional drug interaction modeling.
• Existing computational models struggle with GPCRs due to limited, low-quality structural data and the inadequacy of generalized models trained on soluble proteins.

Purpose of the Study:

• To develop novel computational models for predicting G-protein coupled receptor (GPCR)-ligand interactions using non-structural data.
• To enable efficient and accurate large-scale virtual screening for GPCR-targeted drug discovery.

Main Methods:

• Developed two models, DeepGPCR_BC (binary classification) and DeepGPCR_RG (affinity prediction), utilizing graph convolutional networks and mol2vec.
• Represented GPCR binding pockets and ligands as graphs, processing non-structural interaction data.
• Employed graph-based deep learning to capture physical-chemical and spatial information for predictive modeling.

Main Results:

• DeepGPCR_BC achieved an AUC of 0.72, accuracy of 0.68, and TPR of 0.73, outperforming standard docking tools.
• DeepGPCR_RG demonstrated a Pearson correlation of 0.39 and RMSE of 1.34 for affinity prediction.
• Successfully screened drug candidates for GPR35 and identified active inhibitors for GLP-1R.

Conclusions:

• The developed GPCR-specific deep learning models offer an efficient and accurate approach for virtual screening.
• These models can significantly accelerate drug discovery efforts targeting G-protein coupled receptors (GPCRs).