[ccp4bb] Low B-factors
Dear all I'm refining a 33kDa protein model and I have noticed that although all other statistics seem fine, B factor values are quite low, with many around 7-10A2 and average overall values as follows. Total number of atoms in chain A 2378 Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 Average rms B for 312 main chain residues0.748 Average rms B for 273 side chain residues1.777 These are values after anisotropic B refinement, with 1 TLS group in refmac5. I get similar results if refinement is done without TLS (using model + mtz before doing TLS refinement) and with isotropic/overall/anisotropic B refinement and. Resetting B values to an arbitrary high value (50 or 70) and rerunning refinement with all possible options on/off also gives the same results. Using Phenix, after resetting B values before refinement to arbitrary values and with anisotropic or isotropic refinement, with or without occupancy refinement also gives similar results. As refinement of related molecules (but in different ligand bound states) from similar quality crystals results in models with more realistic B factors (as shown bellow), I strongly suspect that the B factors for this protein have been over refined. However, I'm not sure what other refinement settings to use to avoid this problem. Total number of atoms in chain A 2352 Average B value for main chain A 20.651 Average B value for side chain A 21.815 Average B value for whole chain A 21.199 Any help and advice is most appreciated. Best regards Paula Dr Paula Salgado Division of Molecular Biosciences Department of Life Sciences Faculty of Natural Sciences Biochemistry Building, 5th Floor Imperial College London South Kensington Campus SW7 2AZ London Tel: +44 (0)2075945464 Fax: +44 (0)20 7594 3057
Re: [ccp4bb] Low B-factors
On Tuesday, November 23, 2010, Paula Salgado wrote: Dear all I'm refining a 33kDa protein model and I have noticed that although all other statistics seem fine, B factor values are quite low, What is the resolution? What is the Wilson B? with many around 7-10A2 and average overall values as follows. Total number of atoms in chain A 2378 Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 Average rms B for 312 main chain residues0.748 Average rms B for 273 side chain residues1.777 These are values after anisotropic B refinement, with 1 TLS group in refmac5. I get similar results if refinement is done without TLS (using model + mtz before doing TLS refinement) and with isotropic/overall/anisotropic B refinement and. Resetting B values to an arbitrary high value (50 or 70) and rerunning refinement with all possible options on/off also gives the same results. Using Phenix, after resetting B values before refinement to arbitrary values and with anisotropic or isotropic refinement, with or without occupancy refinement also gives similar results. As refinement of related molecules (but in different ligand bound states) from similar quality crystals results in models with more realistic B factors (as shown bellow), I strongly suspect that the B factors for this protein have been over refined. However, I'm not sure what other refinement settings to use to avoid this problem. Total number of atoms in chain A 2352 Average B value for main chain A 20.651 Average B value for side chain A 21.815 Average B value for whole chain A 21.199 Any help and advice is most appreciated. Best regards Paula Dr Paula Salgado Division of Molecular Biosciences Department of Life Sciences Faculty of Natural Sciences Biochemistry Building, 5th Floor Imperial College London South Kensington Campus SW7 2AZ London Tel: +44 (0)2075945464 Fax: +44 (0)20 7594 3057
Re: [ccp4bb] Low B-factors
Dear Paula, I do not know about phenix, but the refmac5 output PDB file does not contain the contribution from the TLS refinement (not sure I get the terminology right). In order to get an idea of the average B-values, you should run tlsanl on the output of refmac5 with the keyword 'bresid true'. Tim On Tue, Nov 23, 2010 at 07:04:07PM +, Paula Salgado wrote: Dear all I'm refining a 33kDa protein model and I have noticed that although all other statistics seem fine, B factor values are quite low, with many around 7-10A2 and average overall values as follows. Total number of atoms in chain A 2378 Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 Average rms B for 312 main chain residues0.748 Average rms B for 273 side chain residues1.777 These are values after anisotropic B refinement, with 1 TLS group in refmac5. I get similar results if refinement is done without TLS (using model + mtz before doing TLS refinement) and with isotropic/overall/anisotropic B refinement and. Resetting B values to an arbitrary high value (50 or 70) and rerunning refinement with all possible options on/off also gives the same results. Using Phenix, after resetting B values before refinement to arbitrary values and with anisotropic or isotropic refinement, with or without occupancy refinement also gives similar results. As refinement of related molecules (but in different ligand bound states) from similar quality crystals results in models with more realistic B factors (as shown bellow), I strongly suspect that the B factors for this protein have been over refined. However, I'm not sure what other refinement settings to use to avoid this problem. Total number of atoms in chain A 2352 Average B value for main chain A 20.651 Average B value for side chain A 21.815 Average B value for whole chain A 21.199 Any help and advice is most appreciated. Best regards Paula Dr Paula Salgado Division of Molecular Biosciences Department of Life Sciences Faculty of Natural Sciences Biochemistry Building, 5th Floor Imperial College London South Kensington Campus SW7 2AZ London Tel: +44 (0)2075945464 Fax: +44 (0)20 7594 3057 -- -- Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen phone: +49 (0)551 39 22149 GPG Key ID = A46BEE1A signature.asc Description: Digital signature
Re: [ccp4bb] Low B-factors
Resolution is 1.5A, Wilson B is 11.8. Does that mean that might values are more than within expected ranges? I knew at this resolution B factors are considerably lower, but I was suspicious that some seemed, so low, specially after recent problems of over refinement of B factors with another protein at 1.7A... Must admit I never had such nice structures before, so I was being extra careful. Paula On 23 November 2010 19:27, Ethan Merritt merr...@u.washington.edu wrote: On Tuesday, November 23, 2010, Paula Salgado wrote: Dear all I'm refining a 33kDa protein model and I have noticed that although all other statistics seem fine, B factor values are quite low, What is the resolution? What is the Wilson B? with many around 7-10A2 and average overall values as follows. Total number of atoms in chain A 2378 Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 Average rms B for 312 main chain residues0.748 Average rms B for 273 side chain residues1.777 These are values after anisotropic B refinement, with 1 TLS group in refmac5. I get similar results if refinement is done without TLS (using model + mtz before doing TLS refinement) and with isotropic/overall/anisotropic B refinement and. Resetting B values to an arbitrary high value (50 or 70) and rerunning refinement with all possible options on/off also gives the same results. Using Phenix, after resetting B values before refinement to arbitrary values and with anisotropic or isotropic refinement, with or without occupancy refinement also gives similar results. As refinement of related molecules (but in different ligand bound states) from similar quality crystals results in models with more realistic B factors (as shown bellow), I strongly suspect that the B factors for this protein have been over refined. However, I'm not sure what other refinement settings to use to avoid this problem. Total number of atoms in chain A 2352 Average B value for main chain A 20.651 Average B value for side chain A 21.815 Average B value for whole chain A 21.199 Any help and advice is most appreciated. Best regards Paula Dr Paula Salgado Division of Molecular Biosciences Department of Life Sciences Faculty of Natural Sciences Biochemistry Building, 5th Floor Imperial College London South Kensington Campus SW7 2AZ London Tel: +44 (0)2075945464 Fax: +44 (0)20 7594 3057
Re: [ccp4bb] Low B-factors
On Tuesday, November 23, 2010, Paula Salgado wrote: Resolution is 1.5A, Wilson B is 11.8. Right. So there you have it. The Wilson B is telling you straight from the data that the expected B factors are on the order of 11.8 You reported your model as Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 So it looks like you're right on target. Congratulations on a nice crystal, a nice data set, and a promising model. Ethan Does that mean that might values are more than within expected ranges? I knew at this resolution B factors are considerably lower, but I was suspicious that some seemed, so low, specially after recent problems of over refinement of B factors with another protein at 1.7A... Must admit I never had such nice structures before, so I was being extra careful. Paula On 23 November 2010 19:27, Ethan Merritt merr...@u.washington.edu wrote: On Tuesday, November 23, 2010, Paula Salgado wrote: Dear all I'm refining a 33kDa protein model and I have noticed that although all other statistics seem fine, B factor values are quite low, What is the resolution? What is the Wilson B? with many around 7-10A2 and average overall values as follows. Total number of atoms in chain A 2378 Average B value for main chain A 10.684 Average B value for side chain A 12.755 Average B value for whole chain A 11.665 Average rms B for 312 main chain residues0.748 Average rms B for 273 side chain residues1.777 These are values after anisotropic B refinement, with 1 TLS group in refmac5. I get similar results if refinement is done without TLS (using model + mtz before doing TLS refinement) and with isotropic/overall/anisotropic B refinement and. Resetting B values to an arbitrary high value (50 or 70) and rerunning refinement with all possible options on/off also gives the same results. Using Phenix, after resetting B values before refinement to arbitrary values and with anisotropic or isotropic refinement, with or without occupancy refinement also gives similar results. As refinement of related molecules (but in different ligand bound states) from similar quality crystals results in models with more realistic B factors (as shown bellow), I strongly suspect that the B factors for this protein have been over refined. However, I'm not sure what other refinement settings to use to avoid this problem. Total number of atoms in chain A 2352 Average B value for main chain A 20.651 Average B value for side chain A 21.815 Average B value for whole chain A 21.199 Any help and advice is most appreciated. Best regards Paula Dr Paula Salgado Division of Molecular Biosciences Department of Life Sciences Faculty of Natural Sciences Biochemistry Building, 5th Floor Imperial College London South Kensington Campus SW7 2AZ London Tel: +44 (0)2075945464 Fax: +44 (0)20 7594 3057
Re: [ccp4bb] low B factors
Hi Jan, Did you use the TLS refinement for your data? I have had the similar problem with the multi-group TLS refinement. After the first run the resulting pdb file contained unreasonably small atomic B factors for many atoms comparing to the Wilson plot B factor. They seemed to be the TLS-B factors instead of the expected residual B factors. The TLSANL could have not produced the total nor residual B factors when given that file (plus the .tls file from that Refmac run). Moreover, I couldn't use this file as an input for the next Refmac refinement as it was not stable. Interestingly, I have had no problem when using one group only in the TLS step. The version of Refmac was 5.2.0019. Do you observe by chance a similar case? Aleks On 11 Nov 2008, at 01:36, Jan Abendroth wrote: Hi all, I have a number of low-ish resolution data sets that show a strange B-factor behaviour: All are just better than 3AA resolution, collected on a strong synchrotron beamline. Some, yet not tremendous radiation decay. Wilson scaling, obviously not very reliable at this resolution, gives me a Wilson B of about 40, already a low number. Refinement in refmac5 (5.5.0053) with individual B-factors refinement leads to an average B factor of around 16 with several individual B factors hitting the B=2 limit... When I convert Is to Fs in truncate simply using the square root, things get even a bit worse, the average B now is 14. When I try to do an overall B-factor refinement, still individual B-factors appear in the output file. refinement details: 2.8AA resolution,medium ncs for two ncs related chains, no riding hydrogens, simple scaling, MKLF target, isotropic B factors Rwork: 0.206, Rfree=0.286, bonds=0.018, angles=1.89 ... obviously I could retrain a bit more... Any ideas how to handle this? Basically, my question is: how do I get the overall B factor to realistic numbers? Thanks a lot for any hints Jan -- Jan Abendroth deCODE biostructures Seattle / Bainbridge Island, WA, USA work: JAbendroth_at_decode.com home: Jan.Abendroth_at_gmail.com -- Aleksander W. Roszak, PhD E-mail: [EMAIL PROTECTED] Protein Crystallography Web: www.chem.gla.ac.uk/~aleks University of Glasgow Fax: +44-(0)141-330 3779 Level 3 Room B 317 Tel (office): +44-(0)141-330 4476 Glasgow Biomedical Research Centre Tel (X-ray lab): +44-(0)141-330 3589 120 University PlaceMobile: +44-(0)780 9559996 Glasgow G12 8TA Scotland, UK
Re: [ccp4bb] low B factors
It might be possible that there's a problem in the low resolution portion of your data. You could check the wilson plot to make sure it looks normal at that range, and possibly change the low-resolution cutoff refmac is using. Pete Jan Abendroth wrote: Hi all, I have a number of low-ish resolution data sets that show a strange B-factor behaviour: All are just better than 3AA resolution, collected on a strong synchrotron beamline. Some, yet not tremendous radiation decay. Wilson scaling, obviously not very reliable at this resolution, gives me a Wilson B of about 40, already a low number. Refinement in refmac5 (5.5.0053) with individual B-factors refinement leads to an average B factor of around 16 with several individual B factors hitting the B=2 limit... When I convert Is to Fs in truncate simply using the square root, things get even a bit worse, the average B now is 14. When I try to do an overall B-factor refinement, still individual B-factors appear in the output file. refinement details: 2.8AA resolution,medium ncs for two ncs related chains, no riding hydrogens, simple scaling, MKLF target, isotropic B factors Rwork: 0.206, Rfree=0.286, bonds=0.018, angles=1.89 ... obviously I could retrain a bit more... Any ideas how to handle this? Basically, my question is: how do I get the overall B factor to realistic numbers? Thanks a lot for any hints Jan -- Jan Abendroth deCODE biostructures Seattle / Bainbridge Island, WA, USA work: JAbendroth_at_decode.com home: Jan.Abendroth_at_gmail.com
Re: [ccp4bb] low b factors
Are you using TLS? If so the B factors in the PDB file are relative to the TLS parameters.. Or have you lost low resolution data - you can see that in many plots - one in REFMAC gives F v Fcalc as a function of resolution? If so you probably need to use SCALE SIMPLE - in gfact that is usually the safest option.. Which version of CCP4 are you using? Eleanor German Gomez wrote: I am working with a small protein in a relatively high space group p6322 and have good data to a modest 2.5 A resolution. During the first rounds of restrained refinement however the B factors drop to an average of 15 and as low as 2 A^3. This has the effect of producing large amounts of negative density in the voids of the structure. I have tried artificially rasing the b factors and setting a higher low cutoff but nothing seems to help. I have also chosen each of the b models of refinement (anisotropic, overall, etc.) but to no avail. Any suggestions? German Gomez Ready for the edge of your seat? Check out tonight's top picks on Yahoo! TV. http://tv.yahoo.com/
[ccp4bb] low b factors
I am working with a small protein in a relatively high space group p6322 and have good data to a modest 2.5 A resolution. During the first rounds of restrained refinement however the B factors drop to an average of 15 and as low as 2 A^3. This has the effect of producing large amounts of negative density in the voids of the structure. I have tried artificially rasing the b factors and setting a higher low cutoff but nothing seems to help. I have also chosen each of the b models of refinement (anisotropic, overall, etc.) but to no avail. Any suggestions? German Gomez Ready for the edge of your seat? Check out tonight's top picks on Yahoo! TV. http://tv.yahoo.com/
[ccp4bb] Low B factors Resolved
First, Thank you all for the help and suggestions, it truly resolved the issue. First I reset the B factors to something close to the Wilson B average (53). Then making sure the B-solvent contribution flag was set and refining isotropic b-factor I reran the job. The B-factors now are reasonable and the negative density issues have been resolved. Using Babinet instead of simple scaling also brought down the R and Rfree though it did make the maps a little less defined and increased the B factors to a higher average B. Much more refining to do but I'm in a much better position now. Not to mention that my maps are a lot more interpretable. Thanks Take the Internet to Go: Yahoo!Go puts the Internet in your pocket: mail, news, photos more. http://mobile.yahoo.com/go?refer=1GNXIC
