In Computational and structural biotechnology journal
G protein-coupled receptors (GPCRs) are promising drug targets because they play a large role in physiological processes by modulating diverse signaling pathways in the human body. The GPCR-mediated signaling pathways are regulated by four types of ligands-agonists, neutral antagonists, partial agonists, and inverse agonists. Once each type of ligand is bound to the binding site, it activates, deactivates, or does not perturb signaling by shifting the conformational ensemble of GPCRs. Predicting the ligand's effect on the conformation at the binding moment could be a powerful screening tool for rational GPCR drug design. Here, we detected conformational differences by capturing the spatiotemporal residue pair pattern of the ligand-bound β2-adrenergic receptor (β2AR) using a 3-dimensional residual network, 3D-ResNets. The network was trained with the time series of protein distance maps extracted from hundreds of molecular dynamics (MD) simulation trajectories of ten β2AR-ligand complexes. The MD system was constructed with a lipid bilayer embedded in an inactive β2AR X-ray crystal structure and solvated with explicit water molecules. To train the network, three hyperparameters were tested, and it was found that the number of MD trajectories in the training set significantly affected the model's accuracy. The classification of agonists and neutral antagonists was successful, but inverse agonists were not. Between the agonists and antagonists, different residue pair patterns were spotted on the extracellular loop segment. This result demonstrates the potential application of a 3-D neural network in GPCR drug screening, as well as an analysis tool for protein functional dynamics.
Han Minwoo, Lee Seungju, Ha Yuna, Lee Jee-Young
2022
3-D Convolution Neural Network, 3D-ResNets, 3-dimensional residual networks, Artificial Intelligence, ECL, extracellular loop, GPCR, GPCRs, G protein-coupled receptors, ICL, intracellular loop, MD, molecular dynamics, Machine Learning, Molecular Dynamics Simulation, PDM, protein distance map, Pattern Recognition, TM, TMtransmembrane helix, β2-Adrenergic Receptor, β2AR, β2-adrenergic receptor
