I also should backtrack a little. I assumed that Anandhi has good data reduction with good statistics, no data problems (such as twinning), a good molecular replacement solution, and refinement stuck, but he didn't give any of those details. What are the R and Rfree values, the RMS bonds & angles? How does the Ramachandran plot look? By the way, what were the data statistics like (such as Rsym, completeness, multiplicity)? Is the electron density continuous and make sense?
Everything Dale said is well-said. You could have a perfectly good crystal structure with highly mobile surface loops and maybe a few disordered surface regions that have been over-determined and should be removed from the structure, or a protein with domains that like to rearrange, all of which could give rise to regions with high B-factors. More information is needed to know if there's a problem. If there is a problem, then I would start with Eleanor's suggestions to make sure the data is good and the MR solution is correct. Good luck. -Daniel On Sun, Nov 11, 2018 at 6:26 PM Dale Tronrud <de...@daletronrud.com> wrote: > Okay, I'm an idiot. The letter did say molecular replacement. I > have no excuse for being so blind. > > I do stand by my statement: I don't see that an average B factor of > 82 A^2, by itself, indicates any serious problem with a model at 2.65 A > resolution. > > I have generally found that the average B factor of a model is > uninformative. All refinement programs will make certain that the fall > off in intensity of the calculated F's match those of the observations, > even if the model is completely wrong. The average B factor can vary > widely between different models of the same data if the distribution of > individual B's differ. > > For example, a model built by a person in the "never truncate side > chains" camp will have a higher average B factor than the model built by > a "truncater". That just means that the high end tail of the B factor > distribution has been cut off. Atoms with B factors in the 200's only > affect the intensities of a very small number of reflections anyway and > do not have any effect on the Wilson B. > > If a structure has a wobbly domain or an unusual amount of surface > area exposed to solvent, the model will have a lot of atoms with large B > factors. The average B factor will be high, but none-the-less the > majority of the atoms will be well ordered and the Wilson B small. This > is not an indication of error. It is exactly what one would expect from > such a structure. > > We don't know what the distribution of individual B factors is in Dr. > Anandan's model. We don't know if Dr. Anandan truncates side chains or > has a liking for building protein main chain through weak density. We > don't if there are multiple copies of the protein in the asymmetric > unit, nor if some are more ordered than others. > > All we know is that the resolution of the data is 2.65 A (whatever > that means), that the B factor average is 82 A^2 (whatever that means) > and that there is one atom with a B factor of 34.57 A^2 and another with > 225.13 A^2. I just don't see that these few facts indicate a problem. > > If Dr. Anandan has some reason to distrust this model, like maybe not > being able to detect the image of the allegedly bound ligand, a great > number of details of the protein, data collection and the current model > would have to be shared. When asking such a question on this BB you > should never be concerned that your letter is too long. > > Dale Tronrud > > On 11/11/2018 1:06 PM, Dale Tronrud wrote: > > > > Did I miss a follow-up letter with more information? All I've seen > > is that Dr. Anandan said that there was a model based on 2.65 A > > resolution data with an average B factor of 82 A^2. Does this fact > > alone call for weeks of work attempting to remove model bias? Dr. > > Anandan didn't even say this structure was solved via molecular > > replacement. No indication of high R factor or stuck refinement. No > > details at all. Just a model with a fairly reasonable, if somewhat low, > > average B factor. > > > > Dale Tronrud > > > > On 11/10/2018 6:31 PM, Daniel M. Himmel, Ph. D. wrote: > >> Anandhi, > >> > >> Assuming the data reduction went well, and you're in the right space > goup, > >> there could be a lot of model bias in your structure stemming from the > >> starting model. > >> > >> There are a lot of things to try. I would > >> set all the B-factors to an artificially low B-factor to help de-mask > >> errors. Then, > >> you can generate a composite omit map and FEM maps to see if any obvious > >> model errors show up in the electron density. After correcting these, > >> you can try running > >> Autobuild and PhaseAndBuild in Phenix. Compare all the models you get > from > >> each of these, especially in regions where your original model had high > >> b-factors. > >> Use Coot to identify areas of poor agreement with electron density and > areas > >> of poor geometry. Once you spent a while correcting the model, put it > >> through > >> one or more cycles of simulated annealing at different temperatures in > >> parallel. > >> Select several sets of coordinates that give the best Rfree convergence, > >> and then > >> subject those models to individual B-factor refinement. After that, > >> check again in Coot > >> for areas of high B-factors and areas of poor geometry (try especially > >> to improve > >> your Ramachandran Plot). Use MolProbity to help in identifying errors > >> and clashes. > >> If a few rounds of simulated annealing and model building don't improve > >> things, try some > >> refinement in CCP4 Refmac. PDB_REDO, which uses Refmac, can also help > >> give you alternative models. While doing all these, don't be afraid to > >> "cut-and-paste" > >> regions from one model into another model and then correct the geometry > >> in Coot. > >> If B-factors don't come down no matter what you do, you could be in the > >> wrong space > >> group or have problems with the original data that need to be addressed. > >> > >> I hope this helps. > >> > >> Daniel > >> > >> ___________ > >> > >> Daniel M. Himmel, Ph. D. > >> > >> URL: http://www.DanielMHimmel.com/index_Xtal.html > >> <http://www.danielmhimmel.com/> > >> > >> > >> > >> > >> On Thu, Nov 8, 2018 at 7:41 PM Anandhi Anandan > >> <anandhi.anan...@uwa.edu.au <mailto:anandhi.anan...@uwa.edu.au>> wrote: > >> > >> Hello everyone, > >> > >> > >> I am trying to solve the structure of a protein with a bound ligand > >> at 2.65 A resolution. XDS was used for data reduction, phaser-MR for > >> molecular replacement and Phenix for refinement. The refinement was > >> done with the default settings ( individual B factors, occupancy and > >> TLS parameters). The resultant atomic B factors are quite high. The > >> overall B factor is 82 with a minimum value of 34.57 and maximum of > >> 225.13. I would like to know if any of the data reduction parameters > >> can affect the B factors and how best to deal with this issue. > >> > >> > >> Anandhi > >> > >> > >> > >> > >> > ------------------------------------------------------------------------ > >> > >> To unsubscribe from the CCP4BB list, click the following link: > >> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > >> > >> > >> ------------------------------------------------------------------------ > >> > >> To unsubscribe from the CCP4BB list, click the following link: > >> https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > >> > > > > ######################################################################## > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1 > ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB&A=1
