Model bias in macromolecular crystal structures

Reduction of model bias in macromolecular crystallography through various omit-map techniques has been investigated. The two cases studied were the p21 protein complexed with GDP at 2.25 A resolution and the AN02 Fab fragment of an anti-dinitrophenyl-spin-label murine monoclonal antibody complexed with its hapten at 2.9 A resolution. In the former case, the correct model was compared to a partially incorrect model consisting of an exchanged pair of β strands along with rearrangement of the connecting loops whereas, in the latter case, the correct placement of an active-site tryptophan side chain was compared to an incorrect rotamer conformation. Partial structures were created by omission of spherical regions around the incorrect region. Omit maps without refinement of the partial structure showed a large degree of model bias. Model bias could be reduced significantly by refinement of the partial structure. Simulated-annealing refinement of the partial structure showed the best results, followed by conjugate-gradient minimization with or without prior randomization of the partial structure. To avoid compensation for missing atoms during simulated-annealing refinement of the partial structure, a suitable `boundary' region was restrained to the starting coordinates. Model bias removal by iterative density modification was not successful in that it reduced density for both the correct and incorrect conformations.