Re: [ccp4bb] refmac5 vs phenix refine mixed up
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Pavel, dear Garib, how do you figure out automatically the correct flag? (I hope both phenix and refmac will allow to manual overwrite the software's decision) Cheers, Tim On 01/24/2013 07:47 PM, Pavel Afonine wrote: Hi, It would be nice if default setting was the same in different suites. it's a nice idea of course, but I feel it is impractical as it would require changing a lot of software, both modern and legacy. However, given array of flags it is algorithmically trivial to figure out what is test and work flags. That's what phenix.refine have been doing since its beginning (2005). And my understanding is that Refmac does this too. As always, there are corner cases here, but it's better than nothing. Plus, programs (at least phenix.refine, can't speak for others) tell which flag was actually used, and they provide option to define the flag value to use. Pavel - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAkzvUxlJ7aRr7hoRAlVfAKClRD4/JLNDcOab1HjBroQYXND3bQCfegA9 UiHvuKXg2/b3LqlbPWQpKmY= =Awum -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Dear Tim In principle if a user defines freer flag then refmac knows about that (unless freer flag is 0 then refmac assumes that it is default). In this case (if freer defined by user) then it is not altered. regards Garib On 25 Jan 2013, at 09:14, Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Pavel, dear Garib, how do you figure out automatically the correct flag? (I hope both phenix and refmac will allow to manual overwrite the software's decision) Cheers, Tim On 01/24/2013 07:47 PM, Pavel Afonine wrote: Hi, It would be nice if default setting was the same in different suites. it's a nice idea of course, but I feel it is impractical as it would require changing a lot of software, both modern and legacy. However, given array of flags it is algorithmically trivial to figure out what is test and work flags. That's what phenix.refine have been doing since its beginning (2005). And my understanding is that Refmac does this too. As always, there are corner cases here, but it's better than nothing. Plus, programs (at least phenix.refine, can't speak for others) tell which flag was actually used, and they provide option to define the flag value to use. Pavel - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAkzvUxlJ7aRr7hoRAlVfAKClRD4/JLNDcOab1HjBroQYXND3bQCfegA9 UiHvuKXg2/b3LqlbPWQpKmY= =Awum -END PGP SIGNATURE- Dr Garib N Murshudov Group Leader, MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] refmac5 vs phenix refine mixed up
I noticed that Refmac has done the 1vs0 thing correct for ages, which is very useful because mix-ups between the work set and test set used to be quite common in the reflection files at the pdb (Refmac saved me a lot of extra work with this). Dealing with this problem is very simple as the smallest set is typically the test set. Phenix however needs to deal with the CCP4 type reflection binning. Now the size of the sets cannot be used which means that you have find a smarter solution. So I wonder how this is implemented. Does Phenix use the (reasonable) assumption that the test set is labeled 1.00 or 0.00? Or does it also check the sets with other labels? Cheers, Robbie Sent from my Windows Phone From: Garib N Murshudov Sent: 2013-01-25 10:46 To: CCP4BB@JISCMAIL.AC.UK Subject: Re: [ccp4bb] refmac5 vs phenix refine mixed up Dear Tim In principle if a user defines freer flag then refmac knows about that (unless freer flag is 0 then refmac assumes that it is default). In this case (if freer defined by user) then it is not altered. regards Garib On 25 Jan 2013, at 09:14, Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Pavel, dear Garib, how do you figure out automatically the correct flag? (I hope both phenix and refmac will allow to manual overwrite the software's decision) Cheers, Tim On 01/24/2013 07:47 PM, Pavel Afonine wrote: Hi, It would be nice if default setting was the same in different suites. it's a nice idea of course, but I feel it is impractical as it would require changing a lot of software, both modern and legacy. However, given array of flags it is algorithmically trivial to figure out what is test and work flags. That's what phenix.refine have been doing since its beginning (2005). And my understanding is that Refmac does this too. As always, there are corner cases here, but it's better than nothing. Plus, programs (at least phenix.refine, can't speak for others) tell which flag was actually used, and they provide option to define the flag value to use. Pavel - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAkzvUxlJ7aRr7hoRAlVfAKClRD4/JLNDcOab1HjBroQYXND3bQCfegA9 UiHvuKXg2/b3LqlbPWQpKmY= =Awum -END PGP SIGNATURE- Dr Garib N Murshudov Group Leader, MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] refmac5 vs phenix refine mixed up
On Fri, Jan 25, 2013 at 2:24 AM, Robbie Joosten robbie_joos...@hotmail.com wrote: Phenix however needs to deal with the CCP4 type reflection binning. Now the size of the sets cannot be used which means that you have find a smarter solution. So I wonder how this is implemented. Does Phenix use the (reasonable) assumption that the test set is labeled 1.00 or 0.00? Or does it also check the sets with other labels? I forget the exact rules, but the general assumption is that if you have multiple flag values (such as 0 through 19), the test set is marked by the lowest value. If you have just two values, the test set is whichever is less common. (For SHELX files this would typically be -1, for CNS files it would be 1, but you could just as easily swap flag values and it would still pick the correct set.) I'm sure someone can figure out a way to break this (for instance, by assigning the flags with CCP4, but using 7 instead of 0 as the test set), but in practice nearly every file we've seen obeys these rules, and it can of course be overridden by the user. Anyway this is all open-source, so you can check (and re-use!) the logic for yourself here: http://cctbx.svn.sourceforge.net/viewvc/cctbx/trunk/iotbx/reflection_file_utils.py?revision=16491view=markup -Nat
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Well I am not answering your question. What is the (Wilson) B-factor of the diffraction data ? I would personally compare the average isotropic temperature factor of the model to that of the diffraction data. And further the aim of refinement is not to reduce the B-factor. The aim of refinement is to provide a model that agrees with all data available. There are structures around with very high temperature factors (both for the diffraction data set and for the model). There is nothing wrong with that. Fred. On 24/01/13 11:12, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh -- ---x With regards Rajesh K. Harijan Phd Researcher Department of Biochemistry, University of Oulu, Oulu, Finland- 90014 Off Phone: +358 85531174 Mob: +358 400408258 -- Fred. Vellieux (B.Sc., Ph.D., hdr) IBS / ELMA 41 rue Jules Horowitz F-38027 Grenoble Cedex 01 Tel: +33 438789605 Fax: +33 438785494
Re: [ccp4bb] refmac5 vs phenix refine mixed up
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. - - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAQ56UxlJ7aRr7hoRAsJmAJ9RcS1Bp7g53LwiTm1ZMAVAICHXAACfdWgD FlLHo/1euT/BIeSW7EhrvHo= =w9IY -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Yes, the wilson B-factor is comparable which is 53.6. And also same MTZ was used for refmac5 and phenix refine, which is processed one (original one). And also the reflections used in the refinement was: Phenix (46793 reflections) and refmac5 (44431 reflections). I do not know whether I answered you correctly. Thank you Rajesh On Thu, Jan 24, 2013 at 12:12 PM, rajesh harijan rsu.iitku...@gmail.comwrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh -- ---x With regards Rajesh K. Harijan Phd Researcher Department of Biochemistry, University of Oulu, Oulu, Finland- 90014 Off Phone: +358 85531174 Mob: +358 400408258 -- ---x With regards Rajesh K. Harijan Phd Researcher Prof. Rik. K. Wierenga's Group, Department of Biochemistry, University of Oulu, Oulu, Finland- 90014 Off Phone: +358 85531174 Mob: +358 417064469 http://www.biocenter.oulu.fi/projects/wierenga.html
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Dear Tim Gruene, 2013/1/24 Tim Gruene t...@shelx.uni-ac.gwdg.de -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. Is the F/SIGF columns of the output mtz of refmac5 still the same as the F/SIGF columns of the input mtz? If they are the same, why cann't I use the F/SIGF columns of the output mtz as input to the next refinement? Thanks for your reply. - - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAQ56UxlJ7aRr7hoRAsJmAJ9RcS1Bp7g53LwiTm1ZMAVAICHXAACfdWgD FlLHo/1euT/BIeSW7EhrvHo= =w9IY -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Dear Rajesh, In addition to the R/Rfree, you also need to look at issues like stereochemistry, bad contacts, clashes, the general fit into density, unmodelled ligands/waters, Ramachandran outliers, correct side chain rotamers etc etc. I would advice you to spend (a lot of) time visually inspecting your model and the density, and also make use of servers like MolProbity or WhatIF to examine the quality of your model. Fred is very right that the idea of refinement is to produce a model that agrees with all the data, and not just one with lower R values. cheers Ganesh Le 24/01/13 11:12, rajesh harijan a écrit : Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh -- ---x With regards Rajesh K. Harijan Phd Researcher Department of Biochemistry, University of Oulu, Oulu, Finland- 90014 Off Phone: +358 85531174 Mob: +358 400408258
Re: [ccp4bb] refmac5 vs phenix refine mixed up
if they are the same, there is in principle no problem. (you can quickly check using mtzdump) but, just to make sure, I always use the exact same scaled and truncated mtz-file for all refinements of any particular structure. Then there is no doubt at all...and it is in fact easer, i.e. one less file-name to edit in the GUI or script you use. Mark J van Raaij Laboratorio M-4 Dpto de Estructura de Macromoleculas Centro Nacional de Biotecnologia - CSIC c/Darwin 3 E-28049 Madrid, Spain tel. (+34) 91 585 4616 http://www.cnb.csic.es/~mjvanraaij On 24 Jan 2013, at 12:03, Qixu Cai wrote: Dear Tim Gruene, 2013/1/24 Tim Gruene t...@shelx.uni-ac.gwdg.de -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. Is the F/SIGF columns of the output mtz of refmac5 still the same as the F/SIGF columns of the input mtz? If they are the same, why cann't I use the F/SIGF columns of the output mtz as input to the next refinement? Thanks for your reply. - - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAQ56UxlJ7aRr7hoRAsJmAJ9RcS1Bp7g53LwiTm1ZMAVAICHXAACfdWgD FlLHo/1euT/BIeSW7EhrvHo= =w9IY -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear, of course you could ask Garib whether or not the output data were modified by refmac5 - often they are, at least linearly scaled (which would certainly do no harm), and unless you have read the refmac5 code or Garib assures you I would not rely on it. Further trouble is that by using the output mtz-file, which contains more data columns like the sigma-weighted coefficients for map calculations, the e.g. GUI might accidentally pick the wrong one overlooked by the user, especially if the user is less experienced. To always use the same input mtz-file you avoid such possibilities and it also points a novice user to what refinement is actually doing. Best, Tim On 01/24/2013 12:03 PM, Qixu Cai wrote: Dear Tim Gruene, 2013/1/24 Tim Gruene t...@shelx.uni-ac.gwdg.de Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. Is the F/SIGF columns of the output mtz of refmac5 still the same as the F/SIGF columns of the input mtz? If they are the same, why cann't I use the F/SIGF columns of the output mtz as input to the next refinement? Thanks for your reply. - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRARtHUxlJ7aRr7hoRAnvXAKCqUV5IHvKJShQHrN8/cCGmC4DDrACgw9gL 6MGqgIDK4DJ2vcHtuzdWPBc= =Pl4P -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
PS just checked an example, and the the refmac input and output F and SIGF are in fact NOT the same and have been subjected to something more than linear scaling. This was using refmac version 5.5.0109, admittedly not the newest one. So using the refmac output mtz as input for the next run is wrong as Tim states, although it is probable that in practice the resulting differences may not be noticeable. Mark J van Raaij Laboratorio M-4 Dpto de Estructura de Macromoleculas Centro Nacional de Biotecnologia - CSIC c/Darwin 3 E-28049 Madrid, Spain tel. (+34) 91 585 4616 http://www.cnb.csic.es/~mjvanraaij On 24 Jan 2013, at 12:03, Qixu Cai wrote: Dear Tim Gruene, 2013/1/24 Tim Gruene t...@shelx.uni-ac.gwdg.de -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. Is the F/SIGF columns of the output mtz of refmac5 still the same as the F/SIGF columns of the input mtz? If they are the same, why cann't I use the F/SIGF columns of the output mtz as input to the next refinement? Thanks for your reply. - - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRAQ56UxlJ7aRr7hoRAsJmAJ9RcS1Bp7g53LwiTm1ZMAVAICHXAACfdWgD FlLHo/1euT/BIeSW7EhrvHo= =w9IY -END PGP SIGNATURE-
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Dear all As it was already stated it is essential to use the same input file (after scaling and trancating) for all refinement sessions. Output mtz file in the absence of twinning has been scaled to account for anisotropic overall B values. It is modification of the data. In the twinning case output contains detwinned data. It is serious modification of the data and should not be used as input file for next refinement session. Output file in general is representation of the model and useful for model building but not for further refinement cycles. Regards Garib On 24 Jan 2013, at 11:30, Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Dear, of course you could ask Garib whether or not the output data were modified by refmac5 - often they are, at least linearly scaled (which would certainly do no harm), and unless you have read the refmac5 code or Garib assures you I would not rely on it. Further trouble is that by using the output mtz-file, which contains more data columns like the sigma-weighted coefficients for map calculations, the e.g. GUI might accidentally pick the wrong one overlooked by the user, especially if the user is less experienced. To always use the same input mtz-file you avoid such possibilities and it also points a novice user to what refinement is actually doing. Best, Tim On 01/24/2013 12:03 PM, Qixu Cai wrote: Dear Tim Gruene, 2013/1/24 Tim Gruene t...@shelx.uni-ac.gwdg.de Dear Rajesh, first of all, a model is not true or false, it can only be better or worse. The explanation of what you observe depends on what you did: - did you use the identical and very same mtz-file as input to all three scenarios? Some people take the output mtz and use it as input to the next refinement cycle, which is a very, very, bad thing to do. Is the F/SIGF columns of the output mtz of refmac5 still the same as the F/SIGF columns of the input mtz? If they are the same, why cann't I use the F/SIGF columns of the output mtz as input to the next refinement? Thanks for your reply. - did you ensure always the same set of reflections was used for Rfree when switching between programs? If not, your R/Rfree are meaningless. It may also be that combining phenix and refmac5 indeed resulted in a better mode - both programs have some substantial differences in how they work. Best, Tim On 01/24/2013 11:12 AM, rajesh harijan wrote: Dear All, I am working on a perfectly twinned data in space group P31. when I refine this data with phenix refine the R/Rfree is 26.6/29.4 and average B-factor is 38. I did one test now. I used phenix refined pdb and refine with refmac5 and got R/Rfree of 26.2/29.7 and average B-factor is 64. Now I used refmac5 refined pdb and refined with phenix again. Now R/Rfree is 22.1/24.8 and average B-factor is 56. My question is, why B-factor gone up now and R/Rfree reduced. In which refined model should I believe in. If last refined model is true then how should I reduce the B-factor? Thank you Rajesh - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFRARtHUxlJ7aRr7hoRAnvXAKCqUV5IHvKJShQHrN8/cCGmC4DDrACgw9gL 6MGqgIDK4DJ2vcHtuzdWPBc= =Pl4P -END PGP SIGNATURE- Dr Garib N Murshudov Group Leader, MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Most likely scenario is that Phenix by default assigns Rfree flag as 1, while ccp4/refmac - as 0. That would explain your Rfree going down - because your Rfree reflections were refined by refmac. It would be nice if default setting was the same in different suites. Best wishes.
Re: [ccp4bb] refmac5 vs phenix refine mixed up
On Thu, Jan 24, 2013 at 10:34 AM, Leonid Sazanov saza...@mrc-mbu.cam.ac.uk wrote: Most likely scenario is that Phenix by default assigns Rfree flag as 1, while ccp4/refmac - as 0. That would explain your Rfree going down - because your Rfree reflections were refined by refmac. According to Garib, the current version of Refmac will automatically switch to the proper flags, so this problem should go away. -Nat
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Hi, It would be nice if default setting was the same in different suites. it's a nice idea of course, but I feel it is impractical as it would require changing a lot of software, both modern and legacy. However, given array of flags it is algorithmically trivial to figure out what is test and work flags. That's what phenix.refine have been doing since its beginning (2005). And my understanding is that Refmac does this too. As always, there are corner cases here, but it's better than nothing. Plus, programs (at least phenix.refine, can't speak for others) tell which flag was actually used, and they provide option to define the flag value to use. Pavel
Re: [ccp4bb] refmac5 vs phenix refine mixed up
Yes, Nat is right. Starting with the latest version 5.7 (that is part of ccp4) refmac makes sure that it uses correct set for free reflections. Hopefully it will remove some of the confusions when switching from one software to another. refmac 5.8 version should definitely have this feature. This version with some bug fixes and feature additions can be found from this page; http://www2.mrc-lmb.cam.ac.uk/groups/murshudov/ This version should be available from the next ccp4 update. Garib On 24 Jan 2013, at 18:36, Nat Echols wrote: On Thu, Jan 24, 2013 at 10:34 AM, Leonid Sazanov saza...@mrc-mbu.cam.ac.uk wrote: Most likely scenario is that Phenix by default assigns Rfree flag as 1, while ccp4/refmac - as 0. That would explain your Rfree going down - because your Rfree reflections were refined by refmac. According to Garib, the current version of Refmac will automatically switch to the proper flags, so this problem should go away. -Nat Dr Garib N Murshudov Group Leader, MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
