Hello All, I think we are all in agreement that it depends on resolution, or more specifically the amount of experimental data one has versus how many atoms one has in the model to refine.
There is also no question that structures have become better over the decades due to more and better restraints added into the refinement process, and of course people paying more attention to such things. In terms of guidance to questions, such as when to apply grouped, isotropic or anisotropic B factors, it would be nice to have something concrete to respond with. We have plenty of 'rules of thumb' or heuristics that might be valuable to the community that don't have that many years of experience under the belt. My approach, which may or may not be approved by all, is to say that for those dealing with lower resolution structures (3.0 or worse) or with data to parameter ratios less than 4, is to try something that groups B factors instead of refining every atom with its own B factor. There aren't many ways to do this and it would be nice if there were TLS type implementations that allowed one to avoid the individual B factors. There are plenty of data sets out there that fall into this category, so it might be nice to have more tools to apply in these cases. This isn't to say that adding more restraints can't help the refinement, but it isn't necessarily the best initial course of action when starting out with limited data. For very limited data, one might start with a single overall B factor and only do rigid body refinement of a model. I think the results from Pavel's experiment would be interesting and it would be useful to see the resolution bins that fell out for the various categories. cheers, tom ________________________________ From: CCP4 bulletin board <CCP4BB@JISCMAIL.AC.UK> on behalf of Pavel Afonine <pafon...@gmail.com> Sent: Sunday, January 7, 2024 11:28 AM To: CCP4BB@JISCMAIL.AC.UK <CCP4BB@JISCMAIL.AC.UK> Subject: Re: [ccp4bb] Can Refmac5 refine temperature factor residue by group? You don't often get email from pafon...@gmail.com. Learn why this is important<https://aka.ms/LearnAboutSenderIdentification> Hi All, I believe it depends on the resolution. At sufficiently low resolution, it may not be too unreasonable to assume that the main chain atoms of a residue have the same B factors, and its side chain atoms also share the same B factor (different from the main chain). Why? Simply because your low-resolution data cannot resolve the B-factor difference between, for example, CA and C atoms! This is much like expecting low to almost zero deviations from ideal bonds and angles at low resolution, because your low-res data simply cannot resolve a fraction of an Angstrom deviation in bond lengths. The bottom line, in my opinion, is that there is no black-and-white answer to the "grouped vs. individual" B-factor refinement question. Depending on a) the data resolution, and b) your experimentation with both individual and group options, you may choose one over the other. Back in the day (when working on the implementation of B-factor refinement in phenix.refine), I conducted a test where I re-refined a sample, consisting of about 50,000 models from the PDB, using 1) individual, 2) group with one B per residue, and 3) group with two B per residue (one for the main and one for the side chain). There were clear clusters of results where each of the three parameterizations outperformed the others, and this was heavily resolution-dependent. I regret not publishing that result, mostly because I thought I could always re-do it any time later.. but as Ben said "Don't put off until tomorrow what you can do today"! Regarding TLS, since someone mentioned it, I prefer to see TLS as a more physically realistic atomic vibration model, rather than a magic solution to reduce R-factors or a way to decrease the number of refinable parameters (which, in fact, is not the case because normally residual isotropic B factors are always refined on top of TLS anyway, in Phenix at least). All the best! Pavel On Sat, Jan 6, 2024 at 3:41 PM Robbie Joosten <robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>> wrote: Hi Tom, I think restraints do change the data/parameter ratio, but how much is not straightforward. In, at least, the context of the Hamilton test restraints change the degrees of freedom which translates into a change of the effective data/parameter ratio. It is treated as adding extra observations albeit with some unknown weight. Ethan Merritt's way of handeling this unknown weight (which we implemented in bselect) is setting an upper limit for the weight and thus bracketing the value. Lets assume we have glycol. Adding individual B-factors instead of one overall B adds 3 extra parameters. We also add B-factor restraints: 3 bonded atom 1-2 restraints plus 2 angled atoms 1-3 restraints (in Refmac's implementation). So 3 extra parameters, 5 extra restraints. These restraints can at best nullify the effect of the extra parameters so their effective weight is maximum 3/5, but probably less. The weight of the restraints must be somewhere between 0 (e.g when the B-factor restraint weight in Refmac is set to zero) and 0.6 (the restraint weight is huge). Assuming that the restraints do something, the degrees of freedom go up with less than 3 for our glycol refinement. How much less is related to the restraint weight we set in Refinement. Cheers, Robbie On 6 Jan 2024 23:26, Tom Peat <t.p...@unsw.edu.au<mailto:t.p...@unsw.edu.au>> wrote: Hello Robbie, Thanks for the stats and description of what has been done. I think this puts us back into the realm of restraints versus constraints and what is possible when trying to reduce the number of parameters to be refined. Although grouped B-factors don't capture the reality of side chains being more mobile, it is a constraint that reduces the number of parameters being refined and helps highlight regions of the structure which are more mobile (which a flat or average B-factor would not do). Making tighter restraints on the system as a whole doesn't change the data/parameter ratio, which can lead to its own issues, but is certainly better than just letting things go wild. As is often the outcome, we state 'it depends on your individual situation' and generally suggest looking at various possibilities until one finds some compromise which works. Not as intellectually gratifying as having a cut and dry answer to these questions that come up rather frequently. Thanks again for the stats and description. cheers, tom ________________________________ From: Robbie Joosten <robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>> Sent: Sunday, January 7, 2024 8:24 AM To: Tom Peat <t.p...@unsw.edu.au<mailto:t.p...@unsw.edu.au>>; CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK> <CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK>> Subject: RE: [ccp4bb] Can Refmac5 refine temperature factor residue by group? [You don't often get email from robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ] Hi Tom, At 3A the median number of reflections per atom is 3.4 which is indeed lower than 4. So in unrestrained refinement the data/parameter ratio is indeed worse than 1. This is where the effect of the restraints really matter and starting from a flat B-factor model is interesting. This is what pdb-redo does if there are fewer than 4 reflections per atom. In such cases first TLS model are refined, one-group-per-chain (yes, that has room for improvement) plus any user-provided grouping. The TLS model that performs best in refinement is then kept (or no TLS model at all if they don't work). Given this TLS model, the structure model is refined with isotropic B-factors and flat B-factors. Both refinement results are then tested by a program called "bselect" that performs the Hamilton test plus some fallback test. The "best" model is then chosen. If this involves isotropic B-factors, the B-factor restraint weight is then optimised. Some stats: Of the 1958 cases in the databank, 573 are refined with a flat B-factor model (2.9 reflections/atom on average), 520 with isotropic B-factors and tighter-than-default restraint weights (3.8 reflections/atom on average), 612 with isotropic B-factors and looser-than-default weights (4.1 reflections/atom on average), the rest is isotropic with default weights (3.8 ref/atom). So there is a trend given the number of reflections per atom but it is not that strong for individual cases. Testing is needed. I won't claim that this is the best protocol for each of the cases, but I guess they are decent starting points for most. Cheers, Robbie > -----Original Message----- > From: Tom Peat <t.p...@unsw.edu.au<mailto:t.p...@unsw.edu.au>> > Sent: Saturday, January 6, 2024 21:42 > To: CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK>; Robbie Joosten > <robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>> > Subject: Re: [ccp4bb] Can Refmac5 refine temperature factor residue by > group? > > It appears that Zhonghao might be worried about his data to parameter ratio. > At 3 A, one can easily be in a situation where one has fewer reflections than > four times the number of atoms (X, Y, Z plus B). > I like the idea of starting out with the average B (or even Wilson B) and then > doing TLS as that should reduce the number of parameters being refined. > Best regards, tom > > ________________________________ > > From: CCP4 bulletin board > <CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK>> on behalf of Robbie > Joosten <robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>> > Sent: Saturday, January 6, 2024 8:13 PM > To: CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK> > <CCP4BB@JISCMAIL.AC.UK<mailto:CCP4BB@JISCMAIL.AC.UK>> > Subject: Re: [ccp4bb] Can Refmac5 refine temperature factor residue by > group? > > You don't often get email from > robbie_joos...@hotmail.com<mailto:robbie_joos...@hotmail.com>. Learn > why this is important <https://aka.ms/LearnAboutSenderIdentification> > > > One wonders who those "many people" are. You may not want to use them > as your go-to reference for refinement techniques. > > Anyway, Refmac cannot do grouped B-factor refinement, but you are not > missing out on anything. As Eleanor implied, one-per-residue B-factors give > unrealistic results. You are much better off using isotropic B-factors with > tight > restraints (Refmac's default is already quite tight). Add TLS in your > refinement > to see if that helps. > If you have a really poor data/parameter ratio you could go for a flat > B-factor > model and try to capture most of the B-factor in the TLS model. This is > typically not needed at 3A, but there are exceptions (low solvent -more > atoms- or low completeness -fewer reflections- are factors to consider). If > you do go for a flat B-factor model, you need to define sensible TLS groups. > This takes some trial and error. > > pdb-redo has decent algorithms to select the B-factor model and weight for > Refmac. You could use that as a starting point for your model. > > HTH, > Robbie > > > > On 6 Jan 2024 03:13, "chenzhonghao...@163.com<mailto:chenzhonghao...@163.com>" > <chenzhonghao...@163.com<mailto:chenzhonghao...@163.com>> wrote: > > > Dear Prof. Dr. Dodson and all CCP4 community, > > > > Thanks for your reply. > > > > Just now, I used baverage. I found that it can average the B factor but > not refine it. > > This function does not fit my requirement, because my resolution is > low as 3 A. > > Many people said that Refmac5 overrefines the structure if I used > isotropic temperature refinement. > > > > Did refmac5 or other programs in CCP4 have similar functions like > one_adp_group_per_residue or two_adp_groups_per_residue in Phenix? > > > > Any help would be highly appreciated! > > > > > ________________________________ > > chenzhonghao...@163.com<mailto:chenzhonghao...@163.com> > > > From: Eleanor Dodson > <mailto:0000176a9d5ebad7-dmarc-<mailto:0000176a9d5ebad7-dmarc-> > requ...@jiscmail.ac.uk<mailto:requ...@jiscmail.ac.uk>> > Date: 2024-01-05 23:48 > To: CCP4BB <mailto:CCP4BB@JISCMAIL.AC.UK> > Subject: Re: [ccp4bb] Can Refmac5 refine temperature factor > residue by group? > > Hmmm - I am not sure about the value of this - one expects > the longer floppier side chains to have very different B values for the CB > than > the OE2.. > > The program BAVERAGE gives you a plot of mean B value > residue by residue.. > > > > > > > > baverage - averages B over main and side chain atoms > > > SYNOPSIS¶ > <https://www.ccp4.ac.uk/html/baverage.html#synopsis> > > baverage XYZIN foo_in.pdb RMSTAB foo_out1.tab XYZOUT > foo_out2.pdb > [Keyworded input > <https://www.ccp4.ac.uk/html/baverage.html#keywords> ] > > DESCRIPTION¶ > <https://www.ccp4.ac.uk/html/baverage.html#description> > > > A very simple minded program to read a PDB file, tabulate to > RMSTAB the average B values residue by residue (main chain and side chain > separately) and the RMS deviation of the B values from this mean. It also > outputs a PDB file with outlying B factors reset to lie within the given > range. > > > On Fri, 5 Jan 2024 at 03:08, > chenzhonghao...@163.com<mailto:chenzhonghao...@163.com> > <mailto:chenzhonghao...@163.com> > <chenzhonghao...@163.com<mailto:chenzhonghao...@163.com> > <mailto:chenzhonghao...@163.com> > wrote: > > > Dear CCP4 community, > > I found that Refmac5 refined the temperature factor > only by four modes (see the bottom of the attached figure). However, no > grouped B-factor (one or two per residue instead of > one per atom) was found. > > Actually, PHENIX and CNS can do it. But we are not > familiar with both software. I want to know whether Refmac5 refines one or > two group B per residue (for side and main chains) > grouped temperature factor? > > Any help would be highly appreciated > > Thanks in advance. > > best, > > > Zhonghao Chen > > > > > > > > > > ######################################################## > ################ > > To unsubscribe from the CCP4BB list, click the > following link: > https://www.jiscmail.ac.uk/cgi-bin/WA- > JISC.exe?SUBED1=CCP4BB&A=1 > > This message was issued to members of > www.jiscmail.ac.uk/CCP4BB<http://www.jiscmail.ac.uk/CCP4BB> > <http://www.jiscmail.ac.uk/CCP4BB> , a mailing > list hosted by www.jiscmail.ac.uk<http://www.jiscmail.ac.uk> > <http://www.jiscmail.ac.uk> , terms & > conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/ > <https://www.jiscmail.ac.uk/policyandsecurity/> > > > > ________________________________ > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/WA- > JISC.exe?SUBED1=CCP4BB&A=1 > > > ________________________________ > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/WA- > JISC.exe?SUBED1=CCP4BB&A=1 > > > > ________________________________ > > To unsubscribe from the CCP4BB list, click the following link: > https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 Hell ________________________________ To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 ________________________________ To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 ######################################################################## To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB&A=1 This message was issued to members of www.jiscmail.ac.uk/CCP4BB, a mailing list hosted by www.jiscmail.ac.uk, terms & conditions are available at https://www.jiscmail.ac.uk/policyandsecurity/
