Published online by Cambridge University Press: 05 December 2011
In order to investigate and demonstrate objective modelling of proteins as a basis for drug design, we have sought to model several proteins in particularly persuasive circumstances. This is either (a) by filing the results of the model with an independent institution prior to X-ray determination of their structure, or (b) by using wholly automatic, general and reproducible methods, or (c) most often by both. Results suggest the ability to predict the core of the protein to an accuracy of about 1 Å rms deviation between predicted and experimental all-atom coordinates, and of surface loops in the range 1-4 Å rms deviation. Although the upper end of the latter scale seems disturbing, it turns out that many of the surface loops show such large variations for the same protein as studied by different crystallographic groups, particularly when no common protein is used as a starting point for refinement in both cases. Recognising the dynamic nature of some loops on enzymes, and including in the calculation the ability to handle dynamics over long timescales, allows analysis and refinement of enzyme inhibitors as pharmaceuticals. Here we analyse these aspects, particularly by reference to X-ray crystallography data.