Hi Kendall, we've been disusing this off-list between developers and some users... I'm going to copy-paste one of my latest comments: """ Given the poll result I guess there is no way to make everyone happy with a unique solution, so having multiple options (keep/trim/do something else) is clearly the way to go. I guess another poll (or proper research) about "what's distinguishable" (or what you call poor density) would be of help too. Because I bet once we start doing trimming there will be always someone screaming "No! you are trimming/not building at too low/high CC/sigma!" Personally, I'm not leaning towards one or another option. I believe all of them have approximately equal amount of clear advantages and disadvantages. I was just thinking of a longer term research-like solution that might (or might not) bring a novel idea... By the way, another alternative way to model them is to define a probability distribution mask around a side chain where its atoms are expected, and use that mask as a contribution to Fcalc. I guess (not 100% sure) this is what you can do in BUSTER-TNT (at least this is what advertised in their paper). That would account for these missing atoms to some degree without actually including the ATOM records for them into final PDB. One wrinkle though is that in this case you would need to deposit that mask along with your PDB file, since you will be having a mixed model - atomic model + nonatomic model. (I wonder how many users who ever used this option actually did deposit the masks? -:) ) """ All the best! Pavel. On 3/29/11 6:43 PM, Kendall Nettles wrote:
We have been doing a lot of parallel refinements where we are checking out the new options in PHENIX refinement, and one of the things we have observed is that sometimes the sidechains with no clear electron density end up in the main chain density, and distort the model. There is no clear pattern as to which options lead to this phenotype, as different combinations give different results. Until we can sort out what is causing this, it seems clear to me that it is better to delete the side chains. If you want to leave a side chain with no clear electron density, you have to make sure each one is not distorting the model. So leaving the side chains with no clear electron density requires much work, with a benefit that is not clear to me.
Kendall Nettles
On Mar 28, 2011, at 1:04 PM, Ed Pozharski wrote:
Pavel,
- what you mean by "no density", Lack of confidence in placement of the side chain. Everyone would have somewhat different take on it, but the question is more about what to do, not how to decide if the side chain is disordered.
Therefore this raises another item for your questionnaire: There is "other" option, feel free to use it
refine group occupancy for these atoms (one occupancy per all atoms in question - the occupancy typically will refine to something less than 0.5 or so). This raises an entirely different question regarding reliability of occupancy refinement in general due to its correlation with the B-factors. Another can of worms.
This trick with smearing out an atom by B-factor may only work for isolated (single) atoms such as waters because they are not bonded to anything through restraints. Certainly, presence of restraints makes the B-factor increase less steep. I just looked at an instance of a disordered arginine (no density above 1 sigma for any side chain atoms), and B-factors jump from 30 at the backbone to 90 at the tip of the side chain. This would reduce the density level ~5x, which is probably quite sufficient for blending it into the solvent. There could be a bit of a problem in the middle, where B-factors are inflated/deflated, but it does take care of density reduction.
Things like atom-specific restraints and modified restraint target may be of some help, but the effect on the final model may be too small to validate the effort.
-- "I'd jump in myself, if I weren't so good at whistling." Julian, King of Lemurs
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb