Dear all,
Recently I have had a pleasure to work on some high resolution X-ray data. I have solved one structure of around 270 amino acid long enzyme which has been a breeze in this scenario, and I am using this protein for the MR for another structure. Here's the scenario:
I would highly appreciate any feedback and directions on how this pickle can be solved.
I have attached the final screen of Phaser and Refine of Phenix
Best
Michal
Hi Michal Zielinski .
First of all, congrats for you for this new structure!
Everything you mention above seems reasonable. However, no one can actually help you without knowing some data you left out of your post.
For example, how good is your data? You must have a table coming out of XDS, autoPROC, DIALS, HKL3000, MOSFLM or whatever you've used, right? That table (after merging and scaling), is crucial to know how good and accurate your data are. From that, you can also check if the space group is correct (have you tried POINTLESS?). This would first question.
Then, did you tried phenix.xtriage? What's the output for twinning, space group assignment, number of molecules in the asymmetric unit, tNCS, etc? All of this is a MUST CHECK.
Regarding the phasing itself, how close are the two proteins? Sequence identity/similarity? Anything below 20-22% is marginal and eventually won't work (even giving you some meaningful statistics like the ones you present). Have you tried automatic model building? At 1.6 A it would work like a breeze.
And finally, we get to refinement. And this is a complete new world. Have you tried NCS, TLS, solvent update, etc?
If you want, we can continue discuss this offline. Just send me a message or an e-mail ([email protected]), and we can talk in more detail (in an absolute confidential, non-disclosure, non-competitive way, off course!!).
Once again, congrats.
HTH, best regards,
Jose
Yes, please go back and check that your space group is correct. You can have Phaser solve the molecular replacement in all alternative space groups. This may also be a case of pseudosymmetry, in which the true space group has lower symmetry.
Dear Michal,
Both responses are right on the money. The only missing element is placing them on a larger background. Several years back commenting in one of the streams on RG I wrote a short summary of how general notions of crystallography are related to the so called "every day users" and how there are related to the "expert" views. In this short delineation I questioned every single recommendation for casual users of X-ray crystallography such as: low R merge, low R-factor, space group, etc. Even if your data reduction program is really sophisticated like HKL the best statement about the symmetry of the lattice is established only to the accuracy of your distributed data against resolution bins. Accepted values are usually below 10% in Rmerge. Which means that on average your symmetry is violated by around 10% of your structure. So the very low resolution for ideal symmetry must be really low like around 1% in order to believe that statistical errors of data collection do not prevent you from establishing the best possible symmetry fitting the data. Anything above 4% in the lowest resolution bin (around 50-10Å) must be viewed as suspect. It must be this way because, in reality, the solvent structure, dynamic and some static disorder, solutes, and divergent temperature factors distributions always violate any symmetry including translational symmetry. Even your basic assumption of infinite crystals is violated when the symmetry is assumed because of finite size of the crystals. If your crystal is of size around 1 mmm^3 which is huge it is only around 100000 molecules along one side assuming 100Å Rg of the molecule or any cell unit size.
So your duty is to find the best symmetry defining your object. Looking at the space groups and the numbers you produced it is almost compelling that your crystal is twinned and you cannot rely on any densities produced by this defective model. First and simplest test is to reduce the data with lower symmetry and compare Rmerg. Unfortunately in my hands almost half of twinned cases did not show any obvious change. The second test must be statistical distributions of intensities in line with the recommendations of other contributors. The best test I would always recommend no matter how advanced the user is, to have the data reduced in a series of gradually lower symmetry space groups with similar or comparable Rmerge and refinement of the model in these different symmetries. Of course that the number of the molecules and multiplicity of the data would change and therefore reliability and numerical value of Rmerge but a systematic violations of repeated molecules would clearly demonstrate breaking of apparent symmetries. The correct (read best) symmetry would demonstrate itself in much better R. Alternatively if the twinning is apparent use the twin refinement but your space group would suggest that C222 should reduce to some kind of P21 or something with an angle different from 90 deg.
Cheers
Bog
- Badges
- Science topic
- Similar topics
- Physical Sciences
- Materials Science
- Plating