A benchmarking study on virtual ligand screening against homology models of human GPCRs

Abstract

G-protein-coupled receptor (GPCR) is an important target class of proteins for drug discovery, with over 27% of FDA-approved drugs targeting GPCRs. However, being a membrane protein, it is difficult to obtain the 3D crystal structures of GPCRs for virtual screening of ligands by molecular docking. Thus, we evaluated the virtual screening performance of homology models of human GPCRs with respect to the corresponding crystal structures. Among the 19 GPCRs involved in this study, we observed that 10GPCRs have homology models that have better or comparable performance with respect to the corresponding X-ray structures, making homology models a viable choice for virtual screening. For a small subset of GPCRs, we also explored how certain methods like consensus enrichment and sidechain perturbation affect the utility of homology models in virtual screening, as well as the selectivity between agonists and antagonists. Most notably, consensus enrichment across multiple homology models often yields results comparable to the best performing model, suggesting that ligand candidates predicted with consensus scores from multiple models can be the optimal option in practical applications where the performance of each model cannot be estimated.

Performance of virtual screening against GPCR homology models: Impact of template selection and treatment of binding site plasticity

“…In this work, we explored the use of homology model ensembles derived from one or several templates as well as ensembles of MD snapshots. Our finding that ensembles rarely outperform the best single model agrees with results obtained in other benchmarks [32,34]. However, it should be noted that there are several scenarios where it can be advantageous to use ensembles of models in prospective screens.…”

“…Large sets of models with different side chain conformations must hence be considered to evaluate the quality of a template. This finding makes it difficult to compare our results to previous studies that were mainly based on a single homology model per template [24,32,34]. In agreement with expectations, the most accurate predictions of the 5-HT 2A R binding site were obtained based on closely related 5-HT 2 subtypes (66-75% TM sequence identity) and these models showed excellent virtual screening performance.…”

“…A common rule of thumb is to select the template with the highest sequence identity to construct a homology model. However, benchmarks of homology modeling have not found any clear correlation between the sequence identity of the template and virtual screening performance [32] and, intriguingly, GPCR models based on distant templates have occasionally resulted in better ligand enrichment than closely related ones [33,34]. Second, prospective docking screens against homology models are generally carried out against a single rigid structure of the binding site.…”

Differential long‐term regulation of TAS2R14 by structurally distinct agonists

“…It thus appears that the biasing is dependent on 1 or more of these deep‐pocket interactions. Despite the exploratory nature of these modeling results and the limitations of docking to GPCR crystal structures and homology models (45), they point to potential unique interactions of DPD and ORP with the receptor. These compounds may lock the receptor‐β‐arrestin complex in some way to mitigate internalization.…”

G Protein–Coupled Receptors in Asthma Therapy: Pharmacology and Drug Action

“…In the absence of crystal structures, homology modeling is the most accurate and practical tool for the predictions of new GPCR structures. The predicted homology models can yield comparable success rates in structure-based drug design with respect to the corresponding experimentally solved structures, especially for those that have been through refinement/selection with experimental data (Carlsson et al, 2011;Langmead et al, 2012;Mysinger et al, 2012;Lim et al, 2018). Compounds that have been discovered through structure-based drug design using GPCR homology models have also entered clinical trials (Langmead et al, 2012;Huang et al, 2018).…”