John, I would argue with the statement that a steadily increasing Rfree denotes an unstable refinement. This could be due to the fact that your initial model is biased by information not in your data. This would include previous refinement against the TEST set, use of a high resolution MR model at lower resolution. I would define instability as the inability for the refinement to converge to a solution, where the R-factors do not significantly vary by more than ~0.1%. The most common factor for this, in my experience, is inadequate data for the parameters being refined. Having insufficient observations, and restraints, to warrant the large number of parameters being refined. This is common with low resolution structure refinements. As Ian Tickle pointed out many year's ago most refinement programs do not have sensible isotropic B-factor restrains, which is a major problem at low resolution, where the B-factors are universally very high. A problem which is partially solved by TLS refinement.
Mark On Fri, 2009-02-13 at 08:36 -0600, John Bruning wrote: > Hi, > > I would like to survey the group for causes and potential solutions to > unstable refinement (the steady and upward drift of R and Rfree in > each cycle of refinement). I am currently aware of the following > potential causes: twinning (meohedral, pseduomerohedral, or > epitaxial), anisotropy, pseudosymmetry, the lack of appropriately > tight geometric constraints, and extraordinarily high b-factors. If > anyone has run into any other causes of unstable refinement, I would > like to hear about them. > > > Thanks, > John Bruning Yours sincerely, Mark A. White, Ph.D. Assistant Professor, Dept. Biochemistry and Molecular Biology, Manager, Sealy Center for Structural Biology and Molecular Biophysics X-ray Crystallography Laboratory, Basic Science Building, Room 6.660 C University of Texas Medical Branch Galveston, TX 77555-0647 Tel. (409) 747-4747 Cell. (409) 539-9138 Fax. (409) 747-4745 mailto://wh...@xray.utmb.edu http://xray.utmb.edu http://xray.utmb.edu/~white
