PSI Structural Biology Knowledgebase

PSI | Structural Biology Knowledgebase
Header Icons

Related Themes

GPCR modeling: any good?

PSI-SGKB [doi:10.1038/th_psisgkb.2009.35]
Technical Highlight - August 2009
Short description: Results from a competition that invited groups to predict the structure of a GPCR show some promise, but accurate modeling of these receptors remains a challenge.Nature Rev. Drug Discov. 8, 455-463 (2009)

Stefano Constanzi submitted the best overall model of the human adenosine A2A receptor.

In October 2008 a blind prediction assessment was announced. The crystallographic structure of human adenosine A2A receptor bound to the ligand ZM241385 had just been solved by Ray Stevens' group at the Scripps Research Institute. Before the coordinates were released and the structure published, groups were invited to submit their predicted models of the receptor and the ligand-binding site.

The competition ran along similar lines to CASP (Critical Assessment of methods of Protein Structure) and CAPRI (Critical Assessment of PRediction of Interactions) and was called GPCR Dock 2008. It was designed to test how good GPCR structure prediction and ligand-docking programs are by using a blind prediction assessment.

Twenty-nine groups submitted 206 structural models. They were assessed partly by measuring the ligand root mean squared deviation (RMSD) between the real structure and the predicted structure and partly by scoring the number of correct receptor–ligand contacts. These two measurements were combined to produce a z-score to rank the models.

The quality of the predictions for the receptor alone was quite good, with an RMSD of 4.2 Å (standard deviation 0.9 Å) for the alpha carbons. However, the submitted structures varied widely in their accuracy in predicting the ligand binding mode, with an average ligand RMSD of 9.5 Å (standard deviation 3.8 Å) and four (with a standard deviation of seven) ligand–receptor contacts correctly modeled. Thus, the majority of models did not predict the ligand position and the binding interactions very well.

There were good performances by several groups, and the best model overall was submitted by Stefano Costanzi. His model had a ligand RMSD of 2.8 Å from the actual structure and 34 out of 75 contacts were correctly predicted. This model accurately predicts some of the key receptor–ligand interactions.

Some of the models approximated well the real structure and captured the essence of the ligand–receptor interactions, but improvements in GPCR modeling are still needed before models can consistently be relied upon.

Related articles

A pocket guide to GPCRs

Tips for crystallizing membrane proteins

Evolving a better-expressing GPCR

Maria Hodges


  1. M. Michino et al. Community-wide assessment of GPCR structure modelling and ligand docking: GPCR Dock 2008.
    Nature Rev. Drug Discov. 8, 455-463 (2009), doi: 10.1038/nrd2877.

Structural Biology Knowledgebase ISSN: 1758-1338
Funded by a grant from the National Institute of General Medical Sciences of the National Institutes of Health