The statement that occupancy and B factor are correlated means that you can increase or decrease the occupancy while simultaneously increasing or decreasing the B factor w/o changing the R value much. This means that the R value can't tell you much about the correct settings. Since the R value is just a measure of the overall r.m.s. of the difference map (basically), the difference map isn't going to help much either. If you have high enough resolution to see features in the map when you change occupancy and B factor then those combinations are ruled out, but refinement will not generate those combinations because the goal of refinement is to flatten difference maps. You can't get around the fact that at medium resolutions this correlation exits and will cause your occupancy and B factor parameters to have large error bars. Forcing the occupancy to be equal to one can make you feel better, but from you description it is clear that that model is incorrect. You are trying to decide between an incorrect model that has the illusion of precision and a model that optimally fits your data but is imprecise. If I believe a ligand is only partially occupied I'll perform the group occupancy refinement and use the comparison of the average B factor to the neighboring B factors as a validation, but note in any mention of the final occupancy its high uncertainty. If some conclusion of you paper depends on the occupancy being 50% and not 60% you had better try some other technique - 2 Angstrom crystallography is not going to do the trick for you. Remember, when your compound is not present it is likely that water molecules will be. The strongest features in your density are likely to be places that switch between your compound and water. If you have an apo crystal model you can find these waters easily, otherwise you will have to deduce their locations from these strong regions of density. Dale Tronrud Andy Torelli wrote:
Pavel,
Thanks for the suggestions. I will try group occupancy refinement as you suggest. Consequently, I "know" that at intermediate resolution the B-factor and occupancy are coupled and my understanding was that it isn't appropriate to refine both. However, I haven't read anything that rigorously describes when it is appropriate to do so (i.e. at what resolutions or other characteristics of the data). How can one tell whether the result of refining both B-factor and occupancy simultaneously (one occupancy per group in the case of ligands) achieves a reasonable result? Obviously you'd expect potentially to see lower R/Rfree, more uniform B-factors and better electron density (i.e. no or reduced difference electron density). Conversely, what would indicate to you that it is inappropriate to refine both together?
Also, I didn't mention that I also tried to set all the isotropic B-factors to a uniform value before running TLS (recommended in Martyn Winn's manual), but that didn't help.
Thanks for your advice, -Andy
On 11/25/2009 1:30 PM, Pavel Afonine wrote:
Hi Andy,
I've been experimenting with using TLS to refine my 2.4 angstrom resolution structure. There are 2 protomers in the ASU, and each has a metal cluster and a small-molecule ligand. The model has a relatively high average B-factor for protein (~65) atoms and I also suspect that both the cluster and the ligand have partial occupancy. At this resolution however, I've modeled both groups with 100% occupancy and the average B-factors for the atoms of these groups are expectedly higher, ~100 after individual_sites and individual_adp refinement (i.e. no TLS).
Still, I would try refining group occupancy factor for ligands (one refinable occupancy per whole ligand), and before doing so I would re-set ligand's B-factors to an average value.
My problem arises when I try to use TLS. I determined the boundaries for the TLS groups using the TLSMD server, which identified 3 groups per protomer (roughly coincide with domain boundaries). After including TLS refinement, the R and R-free both improve by 1-2%, but the B-factors for the clusters and ligands end up very high. As far as I know there is still no consensus about if one needs to include water, ions or small ligands into TLS groups. I did a few experiments in the past and at that point I found that not including them into TLS groups worked better. But I was using really a few test structures so this is not conclusive. I would not include them into TLS groups, and try refining group occupancy.
Metal atoms in the cluster have B-factors >350 and the ligand atoms have B-factors >120. The average B-factor for protein atoms is slightly higher too (~70). Also, the gap between R and Rfree is larger than I'd expect (19% R and 26% Rfree), although this isn't unheard of and more importantly, the values have fully converged yet.
Try optimizing weights: "optimize_wxc=true optimize_wxu=true", it may help.
Pavel.
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb
_______________________________________________ phenixbb mailing list [email protected] http://www.phenix-online.org/mailman/listinfo/phenixbb