In the first stage of virtual ligand screening, we will apply structure-based GPCR-specific QSAR model trained on the GPCR-ligand bound complexes. A distinguishing feature of the QSAR model is the graph neural network architecture coupled with Behler-Parrinello symmetry functions in the representation of protein-ligand complexes [1]. The derived QSAR model will be applied to billion-size REAL Enamine to obtain focused chemical library as the input for a more computationally expensive screening, as it follows.
In the second stage, we will compose a representative conformational ensemble of the MCHR1 target. We will consider available experimental structures as well as construct theoretical models using AlphaFold2 and MSA subsampling strategy [2] followed by the machine learning filter for the antagonist-bound conformations [3]. Next, we will analyze the obtained conformations using deep learning-based binding profiling approach [4], resulting in the most promising bound-like conformations for the molecular docking.
In the third stage, given the focused chemical library and representative conformations obtained in the first and second stages, respectively, we will run molecular docking pipelines, comprising physics-based and deep learning-based docking engines. We will use our deep learning-based scoring function [5] to estimate the binding affinity from the docked compounds, and apply consensus scoring to obtain the ranked list of the compounds, from which we will select the 100 hit candidates based on the score, chemical scaffold, and manual inspection of the binding poses.
[1] Karlov, Dmitry S., et al. "graphDelta: MPNN scoring function for the affinity prediction of protein–ligand complexes." ACS omega 5.10 (2020): 5150-5159.
[2] Wayment-Steele, Hannah K., et al. "Prediction of multiple conformational states by combining sequence clustering with AlphaFold2." BioRxiv (2022): 2022-10.
[3] Buyanov, Ilya, and Petr Popov. "Characterizing conformational states in GPCR structures using machine learning." Scientific Reports 14.1 (2024): 1098.
[4] Kozlovskii, Igor, and Petr Popov. "Spatiotemporal identification of druggable binding sites using deep learning." Communications biology 3.1 (2020): 1-12.
[5] Mqawass, Ghaith, and Petr Popov. "graphLambda: Fusion Graph Neural Networks for Binding Affinity Prediction." Journal of Chemical Information and Modeling (2024).
