Re: [ccp4bb] Proper detwinning?
Hi Everyone, Thank you for your helpful input.. Below is a summary of my findings so far. In P1, P21, or P212121, translational pseudo-symmetry is not detected, as the max off-origin peak is only 3% of the origin. Pointless still suggests twinning with [L] of ~0.18 and L^2 of ~0.17 after processing and scaling in P21 with different b edges: 105.6 71.7 200.9 71.7 105.6 201.0 71.1 200.9 105.6 For the P1 unit cell, there are faint but present reflections along (0,k,0) at k = n. There are clearly no absences along (h,0,0), but I do see strong absences along (0,0,l) at l = n. I've noticed that my geometry is not so great when performing twin refinement in Refmac5 with NCS (~90.8% Ramachandran favored, ~1.7% outliers). Adding a weighting term of 0.3 (no NCS) seems to improve these (~95.0% favored, ~0.7% outliers) at the cost of Rwork/Rfree (from 15.5%/18.3% to 18.6%/21.5%). The same treatment with NCS enabled counteracts the geometry improvements (~91.1% favored, ~1.6% outliers) and does not really affect Rwork/Rfree. Thanks, Chris On 7/12/14, Isupov, Michail m.isu...@exeter.ac.uk wrote: Dear Chris, If you have pseudo-translation in your data close to (0,0,0.5), for instance, you will have origin ambiguity and, therefore, two P212121 space groups differing by a choice of origin. When ZANUDA reports P212121 as a correct space group it may be not the space group you have submitted. This is true only if you have pseudo-translation, which can be detected by calculation of the native Patterson. To check for it, in ccp4i choose 'GENERATE PATTERSON' from 'EXPERIMENTAL PHASING/HEAVY METAL LOCATION' submenu. Choose Run FFT to generate PATTERSON option and look at at the PEAKMAX output at the end of the log file If there are peaks higher than 15 % of the origin (553.48 in the table below), you have pseudo-translation, In the example below peak at (0.5,0.5,0) is around 30 % of the origin. 111 553.48 0 0 0 0. 0. 0. 0.00 0.00 0.00 2 24 14 171.73 132 132 0 0.5000 0.5000 0. 100.96 100.96 0.00 344 10.20 6 2 0 0.0231 0.0084 0. 4.66 1.69 0.00 Alternatively, many programs, including MOLREP report pseudo-translation. Origin ambiguity int the orthorhombic space group may happen when pseudo-translation is close to half of one of your cell parameters. If you do not have such pseudo-translation, your case is more complex and do not waste time on ZANUDA. If you have pseudo-translation, try to refine the model output by ZANUDA in P212121. Best, Misha From: Chris Fage [cdf...@gmail.com] Sent: 12 July 2014 00:33 To: Isupov, Michail Cc: CCP4BB@JISCMAIL.AC.UK Subject: Re: [ccp4bb] Proper detwinning? Nat and Misha, Thank you for the suggestions. Xtriage does indeed detect twinning in P1, reporting similar values for |L|, L^2, and twin fraction as in P212121. The unit cell dimensions for the 2.0-A structure (P1) are: 72.050 105.987 201.142 89.97 89.98 89.94 P 1 The unit cell dimensions for the 2.8-A structure (P212121) are: 75.456 115.154 202.022 90.00 90.00 90.00 P 21 21 21 I have been processing in HKL2000, which only recognizes one set of unit cell parameters for each Bravais lattice (does anyone know how to change this?). Specifically, for a primitive monoclinic unit cell it estimates: 104.53 71.82 200.99 89.86 91.80 91.16 This is the unit cell which refined to Rwork/Rfree ~ 27%/34%. Indexing in mosflm gives three options for primitive monoclinic: 105.6 71.7 200.9 90.0 90.1 90.0 71.7 105.6 201.0 90.0 89.9 90.0 71.7 200.9 105.6 90.0 90.3 90.0 Attempting to integrate in any of these space groups leads to a fatal error in subroutine MASKIT. I can also use the index multiple lattices feature to get a whole slew of potential space group; however, integrating reflections leads to the same fatal error. Finally, Zanuda tells me that P212121 is the best space group, according to R-factors. However, I do not believe P212121 is the correct assignment. Best, Chris On 7/10/14, Isupov, Michail m.isu...@exeter.ac.uk wrote: I would recommend to run ZANUDA in the default mode from ccp4i or on CCP4 web server. ZANUDA has resolved several similar cases for me. Misha From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Chris Fage [cdf...@gmail.com] Sent: 10 July 2014 01:14 To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] Proper detwinning? Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed
Re: [ccp4bb] Proper detwinning?
hi Chris Can you send the date-stamped log file (called something like mosflm_20140712_140235.lp, I.e. with the year, month, date, etc when you ran the program) to me directly (i.e. not to the BB, since it's likely to be of little interest to all the other subscribers...) so I can have a look at the results of indexing, and the processing that led up to the MASKIT error? Harry -- ** note change of address ** Dr Harry Powell, MRC Laboratory of Molecular Biology, Francis Crick Avenue, Cambridge Biomedical Campus, Cambridge, CB2 0QH Chairman of European Crystallographic Association SIG9 (Crystallographic Computing) On 12 Jul 2014, at 00:33, Chris Fage cdf...@gmail.com wrote: Nat and Misha, Thank you for the suggestions. Xtriage does indeed detect twinning in P1, reporting similar values for |L|, L^2, and twin fraction as in P212121. The unit cell dimensions for the 2.0-A structure (P1) are: 72.050 105.987 201.142 89.97 89.98 89.94 P 1 The unit cell dimensions for the 2.8-A structure (P212121) are: 75.456 115.154 202.022 90.00 90.00 90.00 P 21 21 21 I have been processing in HKL2000, which only recognizes one set of unit cell parameters for each Bravais lattice (does anyone know how to change this?). Specifically, for a primitive monoclinic unit cell it estimates: 104.53 71.82 200.99 89.86 91.80 91.16 This is the unit cell which refined to Rwork/Rfree ~ 27%/34%. Indexing in mosflm gives three options for primitive monoclinic: 105.6 71.7 200.9 90.0 90.1 90.0 71.7 105.6 201.0 90.0 89.9 90.0 71.7 200.9 105.6 90.0 90.3 90.0 Attempting to integrate in any of these space groups leads to a fatal error in subroutine MASKIT. I can also use the index multiple lattices feature to get a whole slew of potential space group; however, integrating reflections leads to the same fatal error. Finally, Zanuda tells me that P212121 is the best space group, according to R-factors. However, I do not believe P212121 is the correct assignment. Best, Chris On 7/10/14, Isupov, Michail m.isu...@exeter.ac.uk wrote: I would recommend to run ZANUDA in the default mode from ccp4i or on CCP4 web server. ZANUDA has resolved several similar cases for me. Misha From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Chris Fage [cdf...@gmail.com] Sent: 10 July 2014 01:14 To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] Proper detwinning? Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. P21: h, -k, -l P1: h, -k, -l; -h, k, -l; -h, -k, l Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? How do I know I'm taking the right approach? Interestingly, I solved the structure of the same protein in P212121 at 2.8 angstroms from a different crystal. Rwork/Rfree bottomed out at ~21%/26%. One unit cell dimension is 9 angstroms greater in the twinned dataset than in the untwinned. Thank you for any suggestions! Regards, Chris
Re: [ccp4bb] Proper detwinning?
Nat and Misha, Thank you for the suggestions. Xtriage does indeed detect twinning in P1, reporting similar values for |L|, L^2, and twin fraction as in P212121. The unit cell dimensions for the 2.0-A structure (P1) are: 72.050 105.987 201.142 89.97 89.98 89.94 P 1 The unit cell dimensions for the 2.8-A structure (P212121) are: 75.456 115.154 202.022 90.00 90.00 90.00 P 21 21 21 I have been processing in HKL2000, which only recognizes one set of unit cell parameters for each Bravais lattice (does anyone know how to change this?). Specifically, for a primitive monoclinic unit cell it estimates: 104.53 71.82 200.99 89.86 91.80 91.16 This is the unit cell which refined to Rwork/Rfree ~ 27%/34%. Indexing in mosflm gives three options for primitive monoclinic: 105.6 71.7 200.9 90.0 90.1 90.0 71.7 105.6 201.0 90.0 89.9 90.0 71.7 200.9 105.6 90.0 90.3 90.0 Attempting to integrate in any of these space groups leads to a fatal error in subroutine MASKIT. I can also use the index multiple lattices feature to get a whole slew of potential space group; however, integrating reflections leads to the same fatal error. Finally, Zanuda tells me that P212121 is the best space group, according to R-factors. However, I do not believe P212121 is the correct assignment. Best, Chris On 7/10/14, Isupov, Michail m.isu...@exeter.ac.uk wrote: I would recommend to run ZANUDA in the default mode from ccp4i or on CCP4 web server. ZANUDA has resolved several similar cases for me. Misha From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Chris Fage [cdf...@gmail.com] Sent: 10 July 2014 01:14 To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] Proper detwinning? Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. P21: h, -k, -l P1: h, -k, -l; -h, k, -l; -h, -k, l Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? How do I know I'm taking the right approach? Interestingly, I solved the structure of the same protein in P212121 at 2.8 angstroms from a different crystal. Rwork/Rfree bottomed out at ~21%/26%. One unit cell dimension is 9 angstroms greater in the twinned dataset than in the untwinned. Thank you for any suggestions! Regards, Chris
Re: [ccp4bb] Proper detwinning?
Chris, To change the axis ordering for e.g. changing which cell edge is the P21 B axis use an hkl matrix command. Probably can do this via the Macros during scaling but I distrust this and just edit scl.in by hand and run it as scalepack scl.in For k,l,h reindexing use hkl matrix 0 1 0 0 0 1 1 0 0 For l, h, k reindexing use hkl matrix 0 0 1 1 0 0 0 1 0 Systematic absences would be an anecdotal indicator that it is/isn't P212121. That would show strong systematic absences for (h,0,0), (0,k,0), (0,0,l) reflections. (Or reflexions if one prefers). While not impossible I would think it statistically unlikely to observe such absences if the data was really P1, P2 or P21. Going back to the images to eyeball the actual reflections on the display can be pretty illuminating. I don't remember Scalepack giving much detail in postrefinement but paying attention to the positional chi^2 values during integration might give clues about how far from 90 those unit cell axes are wandering if you try integrating in different space groups. There's also a method (or was, last time I tried it) to change the way Scalepack postrefines unit cell dimensions (value per frame or value per crystal) which might also help. More hacking of scl.in might be required. However I'm usually pretty happy if my R-free drops 12% at 2.0 Angstrom resolution when going from P21 to P1. I would look for legitimate deviations between previously identical monomers in the map and probably consider using NCS to reduce the random deviation between monomers that actually are identical by symmetry. You may have assigned the crystallographic 21 down the wrong unit cell axis in that P21 test case. Phil Jeffrey Princeton On 7/11/14 7:33 PM, Chris Fage wrote: Nat and Misha, Thank you for the suggestions. Xtriage does indeed detect twinning in P1, reporting similar values for |L|, L^2, and twin fraction as in P212121. The unit cell dimensions for the 2.0-A structure (P1) are: 72.050 105.987 201.142 89.97 89.98 89.94 P 1 The unit cell dimensions for the 2.8-A structure (P212121) are: 75.456 115.154 202.022 90.00 90.00 90.00 P 21 21 21 I have been processing in HKL2000, which only recognizes one set of unit cell parameters for each Bravais lattice (does anyone know how to change this?). Specifically, for a primitive monoclinic unit cell it estimates: 104.53 71.82 200.99 89.86 91.80 91.16 This is the unit cell which refined to Rwork/Rfree ~ 27%/34%. Indexing in mosflm gives three options for primitive monoclinic: 105.6 71.7 200.9 90.0 90.1 90.0 71.7 105.6 201.0 90.0 89.9 90.0 71.7 200.9 105.6 90.0 90.3 90.0 Attempting to integrate in any of these space groups leads to a fatal error in subroutine MASKIT. I can also use the index multiple lattices feature to get a whole slew of potential space group; however, integrating reflections leads to the same fatal error. Finally, Zanuda tells me that P212121 is the best space group, according to R-factors. However, I do not believe P212121 is the correct assignment. Best, Chris On 7/10/14, Isupov, Michail m.isu...@exeter.ac.uk wrote: I would recommend to run ZANUDA in the default mode from ccp4i or on CCP4 web server. ZANUDA has resolved several similar cases for me. Misha From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Chris Fage [cdf...@gmail.com] Sent: 10 July 2014 01:14 To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] Proper detwinning? Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. P21: h, -k, -l P1: h, -k, -l; -h, k, -l; -h, -k, l Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? How do I know I'm taking the right approach? Interestingly, I solved the structure of the same protein in P212121 at 2.8 angstroms from a different crystal. Rwork/Rfree bottomed out at ~21%/26%. One unit cell dimension is 9 angstroms greater in the twinned dataset than in the untwinned. Thank you for any suggestions! Regards, Chris
Re: [ccp4bb] Proper detwinning?
I would recommend to run ZANUDA in the default mode from ccp4i or on CCP4 web server. ZANUDA has resolved several similar cases for me. Misha From: CCP4 bulletin board [CCP4BB@JISCMAIL.AC.UK] on behalf of Chris Fage [cdf...@gmail.com] Sent: 10 July 2014 01:14 To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] Proper detwinning? Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. P21: h, -k, -l P1: h, -k, -l; -h, k, -l; -h, -k, l Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? How do I know I'm taking the right approach? Interestingly, I solved the structure of the same protein in P212121 at 2.8 angstroms from a different crystal. Rwork/Rfree bottomed out at ~21%/26%. One unit cell dimension is 9 angstroms greater in the twinned dataset than in the untwinned. Thank you for any suggestions! Regards, Chris
[ccp4bb] Proper detwinning?
Hi Everyone, Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. P21: h, -k, -l P1: h, -k, -l; -h, k, -l; -h, -k, l Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? How do I know I'm taking the right approach? Interestingly, I solved the structure of the same protein in P212121 at 2.8 angstroms from a different crystal. Rwork/Rfree bottomed out at ~21%/26%. One unit cell dimension is 9 angstroms greater in the twinned dataset than in the untwinned. Thank you for any suggestions! Regards, Chris
Re: [ccp4bb] Proper detwinning?
On Wed, Jul 9, 2014 at 5:14 PM, Chris Fage cdf...@gmail.com wrote: Despite modelling completely into great electron density, Rwork/Rfree stalled at ~38%/44% during refinement of my 2.0-angstrom structure (P212121, 4 monomers per asymmetric unit). Xtriage suggested twinning, with |L| = 0.419, L^2 = 0.245, and twin fraction = 0.415-0.447. However, there are no twin laws in this space group. I reprocessed the dataset in P21 (8 monomers/AU), which did not alter Rwork/Rfree, and in P1 (16 monomers/AU), which dropped Rwork/Rfree to ~27%/32%. Xtriage reported the pseudo-merohedral twin laws below. ... Performing intensity-based twin refinement in Refmac5 dropped Rwork/Rfree to ~27%/34% (P21) and ~18%/22% (P1). Would it be appropriate to continue with twin refinement in space group P1? It sounds like you have pseudo-symmetry and over-merged your data in P212121. I would try different indexing for P21 before giving up and using P1 (you may be able to just re-scale without integrating again, but I'm very out of date); the choice of 'b' axis will be important. If none of the alternatives work P1 may be it, but I'm curious whether the intensity statistics still indicate twinning for P1. -Nat