Hello everyone, I just collected some data on a new protein. It indexes in P3 and looks like I can scale it all the way to P 63 2 2 . My questions are: 1-I see a reasonable chi-square value 1.69 but r-merge 0.15 (0.49 at 2.75 A). It was 0.20 before I excluded some bad frames. I thought this R value is a bit high... Is it? Mosaicty was around 1.00. 2- My search model has an identity of around 40%. I tried MOLREP. Seems I got a solution LLG=486 PAK=10 TFZ ~30 RFZ ~20. My R-free before any refinement is 0.54. Two questions here, a) I think I only fully understand the meaning of LLG, what about PAK and both Z-values? b)is this R-free acceptable or is it also a little high? Thank you for any input
On Tue, Jul 12, 2011 at 6:53 PM, Yuri
1-I see a reasonable chi-square value 1.69 but r-merge 0.15 (0.49 at 2.75 A). It was 0.20 before I excluded some bad frames. I thought this R value is a bit high... Is it? Mosaicty was around 1.00.
It is on the high side, but I've seen worse. Others might have ideas on how to improve it, but some datasets are like this, and sometimes the processed data works just fine for refinement. (Some peer reviewers enjoy complaining about crystallography statistics, however.)
2- My search model has an identity of around 40%. I tried MOLREP. Seems I got a solution LLG=486 PAK=10 TFZ ~30 RFZ ~20. My R-free before any refinement is 0.54. Two questions here, a) I think I only fully understand the meaning of LLG, what about PAK and both Z-values?
Do you mean MOLREP the program, or just shorthand for molecular replacement? I'm assuming the latter since I don't think MOLREP reports statistics like those, but the answer does depend on the program. I believe PAK is the number of bad clashes between place molecules; for a 40% identical model 10 clashes isn't terribly surprising (I believe the cutoff is 15 or so). Running Sculptor on the search model will trim off some of the most divergent parts and probably decrease the clashes and improve your solution in general - this is highly recommended and will make rebuilding much easier later. RFZ and TFZ are "Z-scores" for the rotation and translation functions respectively. RFZ in Phaser isn't very relevant, but TFZ (along with non-negative LLG) is the most important indicator of whether the structure is solved. A TFZ above 8 with a single copy placed almost always means that MR worked; if you have multiple copies it might need to be higher (I'm not sure what the exact rules are). These statistics look very encouraging, anyway.
b)is this R-free acceptable or is it also a little high?
This is normal for an unrefined MR solution, especially if it isn't sequence identical - what matters is whether it decreases significantly when you refine. -Nat
On Tue, 2011-07-12 at 21:53 -0400, Yuri wrote:
I thought this R value is a bit high...
This has been discussed elsewhere and many times. The bottom line is that elevated Rmerge should never prevent you from attempting to solve the structure (especially if you have 40% identity model for MR). Rmerge is related to data quality (e.g. will be higher for a split crystal), but also depends on many factors that have nothing to do with it (e.g. redundancy, near-translational NCS, etc). It is quite astonishing that Manfred Weiss paper on this matter is celebrating it's 10th birthday and it is still commonplace for many (myself included) to report the Rmerge and not it's much more robust alternatives. Cheers, Ed. -- "I'd jump in myself, if I weren't so good at whistling." Julian, King of Lemurs
participants (3)
-
Ed Pozharski -
Nathaniel Echols -
Yuri