Reflection file conversion
I use d*TREK to reduce my data and phenix.reflection_file_converter to import the data. I seem to have had a bit of trouble generating the free R flags. It seems that the cap for the max number of free R reflections takes precedence over the percentage, even when the percentage is given expressly on the command line. I find myself setting the cap artificially high on the command line so that a full 10% is set aside. I was taught to set aside 10% for the free R set, only going lower if setting aside the full 10% compromised completeness. I'm just wondering what is the rationale for the numeric cap? Am I making any error(s) with my methodology? Nate N
Nate, the common wisdom seems to be that setting aside 1000-2000 reflections is enough to ensure proper statistics anything more is overkill. I am not sure about the exact reference for these numbers, but you could start with Axel Brunger's original paper about refinement with FreeR. Also a couple of Ian Tickles latest papers deal with the issue. HTH Carsten
-----Original Message----- From: [email protected] [mailto:phenixbb- [email protected]] On Behalf Of Nate Nicely Sent: Tuesday, January 19, 2010 12:29 PM To: [email protected] Subject: [phenixbb] Reflection file conversion
I use d*TREK to reduce my data and phenix.reflection_file_converter to import the data. I seem to have had a bit of trouble generating the free R flags. It seems that the cap for the max number of free R reflections takes precedence over the percentage, even when the percentage is given expressly on the command line. I find myself setting the cap artificially high on the command line so that a full 10% is set aside.
I was taught to set aside 10% for the free R set, only going lower if setting aside the full 10% compromised completeness. I'm just wondering what is the rationale for the numeric cap? Am I making any error(s) with my methodology?
Nate N
_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb
I am not sure about the exact reference for these numbers,
In addition, see phenix.refine manual, section "Relevan reading": http://www.phenix-online.org/documentation/refinement.htm#anch88 Pavel.
I use d*TREK to reduce my data and phenix.reflection_file_converter to import the data. I seem to have had a bit of trouble generating the free R flags. It seems that the cap for the max number of free R reflections takes precedence over the percentage, even when the percentage is given expressly on the command line. I find myself setting the cap artificially high on the command line so that a full 10% is set aside.
Yes, that's what you need to do.
I was taught to set aside 10% for the free R set, only going lower if setting aside the full 10% compromised completeness. I'm just wondering what is the rationale for the numeric cap? Am I making any error(s) with my methodology?
For the alpha/beta (or sigmaa) calculation you need a certain number of reflections per resolution bin to get reasonable estimates. Say you have 10 bins, 2000/10 = 200 is a sufficiently large sample per bin. I have to admit I'm uncertain about the pros and cons of setting aside a larger number of reflections. Ralf
Hi Nate,
I was taught to set aside 10% for the free R set, only going lower if setting aside the full 10% compromised completeness. I'm just wondering what is the rationale for the numeric cap? Am I making any error(s) with my methodology?
For the alpha/beta (or sigmaa) calculation you need a certain number of reflections per resolution bin to get reasonable estimates. Say you have 10 bins, 2000/10 = 200 is a sufficiently large sample per bin.
I have to admit I'm uncertain about the pros and cons of setting aside a larger number of reflections.
Test reflections are not used in map calculation. So obviously, more reflections you exclude as free reflections, less reflections will be used in map calculation. Personally, I've never seen this being a trouble but the rumor is that this may worsen the maps. I think the way PHENIX generates free-flags for you is totally fine (unless you prove otherwise). Pavel.
participants (4)
-
Nate Nicely -
Pavel Afonine -
Ralf W. Grosse-Kunstleve -
Schubert, Carsten [PRDUS]