Statistics for the analysis of molecular dynamics simulations: providing P values for agonist-dependent GPCR activation

Agustín Bruzzese,Jesús Giraldo,James A R Dalton

doi:10.1038/s41598-020-77072-4

Abstract

Molecular dynamics (MD) is the common computational technique for assessing efficacy of GPCR-bound ligands. Agonist efficacy measures the capability of the ligand-bound receptor of reaching the active state in comparison with the free receptor. In this respect, agonists, neutral antagonists and inverse agonists can be considered. A collection of MD simulations of both the ligand-bound and the free receptor are needed to provide reliable conclusions. Variability in the trajectories needs quantification and proper statistical tools for meaningful and non-subjective conclusions. Multiple-factor (time, ligand, lipid) ANOVA with repeated measurements on the time factor is proposed as a suitable statistical method for the analysis of agonist-dependent GPCR activation MD simulations. Inclusion of time factor in the ANOVA model is consistent with the time-dependent nature of MD. Ligand and lipid factors measure agonist and lipid influence on receptor activation. Previously reported MD simulations of adenosine A2a receptor (A2aR) are reanalyzed with this statistical method. TM6–TM3 and TM7–TM3 distances are selected as dependent variables in the ANOVA model. The ligand factor includes the presence or absence of adenosine whereas the lipid factor considers DOPC or DOPG lipids. Statistical analysis of MD simulations shows the efficacy of adenosine and the effect of the membrane lipid composition. Subsequent application of the statistical methodology to NECA A2aR agonist, with resulting P values in consistency with its pharmacological profile, suggests that the method is useful for ligand comparison and potentially for dynamic structure-based virtual screening.

Highlights

Molecular dynamics (MD) simulations is an established computational tool for the examination of the conformational flexibility of molecules, in particular p roteins[1]
We present a multiple-factor analysis of variance (ANOVA) with repeated measurements on the time factor, a classical statistical approach which makes use of the time-dependent nature of MD simulations and quantifies the statistical effect that several experimental conditions may have on the activation capability of a receptor
In the present article we reanalyze a recent study of ours in which the activation of the adenosine A2a Class A G protein-coupled receptors (GPCRs) (A2aR) was examined[7]

Summary

Introduction

Molecular dynamics (MD) simulations is an established computational tool for the examination of the conformational flexibility of molecules, in particular p roteins[1]. To reveal GPCR activation conformational features, MD simulations need at least μs-length trajectories and the use of several replicas to provide sufficient confidence to computational results. Importantly is that MD simulations are inherently time-dependent and, the time factor should be present in their statistical analysis. Different multivariate statistical analyses such as cluster and principal component analysis and, more recently, machine learning approaches are being a pplied[5,6]. We present a multiple-factor analysis of variance (ANOVA) with repeated measurements on the time factor, a classical statistical approach which makes use of the time-dependent nature of MD simulations and quantifies the statistical effect that several experimental conditions may have on the activation capability of a receptor. The main advantages of the approach are (1) its general practicality, as it is included in most statistical packages, and (2) the computational production of P values, which removes subjectivity from conclusions

Methods

Results

Conclusion