Errors and reproducibility in electron-density map interpretation

Three investigators, with varying levels of experience, independently built and refined the structure of Escherichia coli ribokinase at 2.6 Å resolution. At the end of the refinement/rebuilding processes the models had essentially converged, although each had its own particular pattern of remaining errors. The subsequent refinement of the same structure at 1.8 Å resolution allowed an overall quality check of each of the lower resolution models, and an analysis of which graphics-based tools were generally most efficient in locating these errors. Criteria which are useful in the application of Ramachandran, main-chain and side-chain database and real-space fit analyses are presented.