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

Abstract

G-protein coupled receptors (GPCRs), crucial in various diseases, are targeted of over 40% of approved drugs. However, the reliable acquisition of experimental GPCRs structures is hindered by their lipid-embedded conformations. Traditional protein-ligand interaction models falter in GPCR-drug interactions, caused by limited and low-quality structures. Generalized models, trained on soluble protein-ligand pairs, are also inadequate. To address these issues, we developed two models, DeepGPCR_BC for binary classification and DeepGPCR_RG for affinity prediction. These models use non-structural GPCR-ligand interaction data, leveraging graph convolutional networks and mol2vec techniques to represent binding pockets and ligands as graphs. This approach significantly speeds up predictions while preserving critical physical-chemical and spatial information. In independent tests, DeepGPCR_BC surpassed Autodock Vina and Schrödinger Dock with an area under the curve of 0.72, accuracy of 0.68 and true positive rate of 0.73, whereas DeepGPCR_RG demonstrated a Pearson correlation of 0.39 and root mean squared error of 1.34. We applied these models to screen drug candidates for GPR35 (Q9HC97), yielding promising results with three (F545-1970, K297-0698, S948-0241) out of eight candidates. Furthermore, we also successfully obtained six active inhibitors for GLP-1R. Our GPCR-specific models pave the way for efficient and accurate large-scale virtual screening, potentially revolutionizing drug discovery in the GPCR field.

Keywords: GLP-1R; GPCR; GPR35; drug screening; graph convolutional network.

Figure 1

The workflow of DeepGPCR_BC and DeepGPCR_RG models construction and evaluation. (A) illustrates the input graphic representation, model architecture, output and performance evaluation metrics. (B) depicts the process of obtaining GPCR–compound pairs and a 3D binding pocket of GPCR. (C) The detailed layer information of the GCN model. (D) briefly introduces the training label and the usage of the predicted label during application.

Figure 2

The modeled protein for the extra test set and the modeled ligands determines the pockets. Out of the 16 test cases, 6 achieved an AUC larger than 0.7, and 10 achieved an AUC larger than 0.6. It should be noted that none of the 16 proteins were included in the training set.

Figure 3

GPR35 screening pipeline and identification of active compounds. (A) Schematic representation of the stepwise screening process leading to the discovery of three active molecules. (B) Conformation of last frame from the MD simulation and 2D protein–ligand interaction diagrams for the identified active compounds.

Figure 4

Characteristics of selective candidates on GPR35. (A) Zaprinast, an endogenous ligand of GPR35, and (B–K) selective candidates of GPR35. N.D. denotes not determined.

Figure 5

Characteristics of selective candidates on GLP-1R. (A) Taspoglutide is a former experimental drug, a glucagon-like peptide-1 agonist (GLP-1 agonist) and (B–K) selective candidates of GLP-1R. N.D. denotes not determined.

Figure 6

The docking interactions of GPR35 with three active molecules and a known active control molecule in both 3D and 2D representations. (A) Residue-specific interactions between pocket residues and compound K297-0698. Interaction diagram of pocket residues with compound K297-0698. (B) Residue-specific interactions between pocket residues and compound F545-1970. Interaction diagram of pocket residues with compound F545-1970. (C) Residue-specific interactions between pocket residues and compound S948-0241. Interaction diagram of pocket residues with compound S948-0241. (D) Residue-specific interactions between pocket residues and known active compound zaprinast.

Figure 7

The predicted interactions of GLP_R1 with six active molecules from docking. (A) Residue-specific interactions between pocket residues and compound C700-1841. (B) Residue-specific interactions between pocket residues and compound G764-0921. (C) Residue-specific interactions between pocket residues and compound S954-5266. (D) Residue-specific interactions between pocket residues and known active compound V005-2405. (E) Residue-specific interactions between pocket residues and known active compound V009-0856. (F) Residue-specific interactions between pocket residues and known active compound V027-3795.