On Friday, June 10, 2011 12:33:10 pm Matthew Franklin wrote: > Hi all - > > I seem to be encountering a problem with Refmac refinement of a > structure containing anisotropic B factors. I have been using TLS > refinement and individual anisotropic B refinement, which may not be > correct technique, but it does lower my R and Rfree versus aniso B > refinement alone (see below).
I don't think that combination makes any sense. Whatever anisotropic is being described by the TLS parameters can also be fully described by the individual anisotropic U^ij terms. So the TLS parameters are entirely redundant, leading the minimization function to be poorly defined. > I am asking Refmac to output the TLS > contribution to the ANISOU lines, presumably in addition to the aniso B > refinement contribution. I think that won't work, but it is a secondary issue. > I also reset the temperature factors to 20 at > the beginning of each refinement round. The refinement resolution is 75 > to 1.8 A, and the space group is C2, if it matters. I am virtually certain that refinement of individual anisotropic U^ij terms cannot be justified at 1.8A. Too many parameters, too few observations. I strongly suggest that you drop the idea of anisotropic refinement at this resolution, and instead concentrate on refining B_iso + TLS. On the first cycle of refinement you can reset the B factors to 20. If you want to continue refinement after that, do not tell refmac to add the TLS contribution to the output B. You want to use those, plus the output refined TLS parameters, to continue from where the first cycle left off. To the best of my knowledge, once you merged the individual Biso and the TLS contributions into one number, the program provides no way to use that number to continue with further refinement. happy refining, Ethan > > The problem comes when I take the output from such a run, model build in > Coot, and feed it back in for the next round of refinement. As you can > see from the summary table below (trimmed for brevity), the refinement > blows up rather badly, making the R factors and the geometry > substantially worse. > > Ncyc Rfact Rfree FOM -LL -LLfree rmsBOND zBOND > rmsANGL zANGL rmsCHIRAL $$ > $$ > 0 0.3376 0.3576 0.556 115671. 6318.1 0.0 > 0.0 0.0 0.0 0.0 > 5 0.2676 0.2921 0.680 109411. 6009.2 0.0 > 0.0 0.0 0.0 0.0 > 9 0.2529 0.2785 0.703 108589. 5978.4 0.0 > 0.0 0.0 0.0 0.0 > 10 0.2516 0.2787 0.710 108459. 5975.2 0.0170 > 0.829 1.534 0.731 0.097 > 15 0.2503 0.2876 0.711 108711. 6042.1 0.0241 > 0.998 1.729 0.833 0.111 > 20 0.2739 0.3152 0.672 110787. 6147.3 0.0268 > 1.103 1.837 0.885 0.119 > 25 0.2949 0.3388 0.638 112219. 6215.4 0.0255 > 1.049 1.808 0.866 0.116 > 30 0.3114 0.3575 0.609 113094. 6257.7 0.0243 > 1.006 1.794 0.854 0.114 > > This refinement also throws out an enormous number of warnings about > adjacent atoms' B factors being substantially different. Most of these > warnings appear to involve the autobuilt riding hydrogens and their > adjacent heavy atoms. > > If I use pdbcur to strip out the ANISOU lines, but otherwise keep the > file and refinement protocol unchanged, it goes along nicely: > > Ncyc Rfact Rfree FOM -LL -LLfree rmsBOND zBOND > rmsANGL zANGL rmsCHIRAL $$ > $$ > 0 0.2912 0.3233 0.667 112220. 6151.2 0.0 > 0.0 0.0 0.0 0.0 > 5 0.2476 0.2787 0.751 107853. 5934.7 0.0 > 0.0 0.0 0.0 0.0 > 9 0.2468 0.2764 0.754 107501. 5915.6 0.0 > 0.0 0.0 0.0 0.0 > 10 0.2469 0.2763 0.754 107480. 5914.2 0.0170 > 0.829 1.534 0.731 0.097 > 15 0.1933 0.2387 0.810 101501. 5723.3 0.0238 > 0.994 1.791 0.868 0.118 > 20 0.1849 0.2327 0.817 100446. 5694.7 0.0239 > 0.992 1.826 0.884 0.120 > 25 0.1818 0.2316 0.821 100034. 5681.0 0.0223 > 0.925 1.775 0.855 0.114 > 30 0.1804 0.2296 0.824 99820. 5669.4 0.0221 > 0.913 1.763 0.848 0.113 > > (Without TLS refinement, the final R and Rfree would be 0.1896 and 0.2503.) > > So, what's happening here? Does Refmac not like ANISOU lines in the > input PDB file? I don't usually work with structures at a resolution > high enough to warrant aniso B refinement, so I haven't encountered this > before. > > Thanks for any advice, > > Matt > > > -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742