To further assess the utility of homology models for predicting ligand-receptor interactions, GPCR Network organized a community-wide assessment of GPCR modeling and docking. More than 30 computational groups participated in the assessment, which included modeling of dopamine D3 receptor and CXCR4 complexes with antagonists; the CXCR4 assessment included both small molecule and peptide antagonists. The analysis of the blind modeling results suggests that reliable prediction of ligand-receptor interactions is feasible when the target receptor homology to a structural template exceeds 30-35% in the TM helical bundle (like in the DRD3 case), approaching accuracy of a high-resolution crystal structure. Modeling based on less than 30% homology, as in the case of CXCR4, is substantially less accurate and often will have limited applicability to prediction of ligand binding interactions.
Further information and assessment results are available here >>