Title: Structure Refinement at Low Resolution
Studies of very large proteins or large protein assemblies are a big challenge in structural biology since high resolution data are often difficult to obtain. Once an approximate model is built, its refinement at low resolution is particularly prone to over-fitting, since the number of parameters is much larger than the number of independent observables.
We present the Deformable Elastic Network (DEN) approach which combines low resolution data with prior structural knowledge, e.g. from homology models, and which effectively reduces the number of parameters that need to be fitted. Its application to refinement in X-ray crystallography in the 4-5 A resolution range drastically improves the quality of the obtained structures. Furthermore, we combine the DEN approach with efficient geometry-based conformational sampling techniques and use this to flexibly fit atomic models into density maps obtained by cryo-electron microscopy.