On Tuesday, 26 July, 2016 19:13:50 Harry Powell wrote:
> hi Ethan
>
> I think this is absolutely wrong. The default behaviour of Mosflm is
> to refine the cell parameters via post-refinement before integration
> and gives accurate values;
Yes. This is what I meant when I inaccurately said "during indexing".
When you reach the step of "Cell Refinement" in the imosflm interface
you can keep tweaking the choice of specific frames and the number of
cycles until you accept the residuals. If for some reason you have
problems and decide to proceed anyhow, you can do that easily enough
even though the residuals are larger than they should be.
So far as I know, the imperfect cell parameters are then held fixed
throughout the subsequent integration run. Do I have that wrong?
> HKL refines the cell parameters accurately following integration,
> in the scaling step, which should give equally correct refined values.
Yes, but unlike mosflm, HKL by default continues to refine the cell parameters
incrementally as it proceeds through the full set of frames being integrated.
So even if your initial values are poor, they are gradually corrected as
you go. This mode of operation is necessitated by the fact that HKL does
its initial indexing using only the first few frames, whereas mosflm uses
data from frames well separated in reciprocal space. So mosflm is more
likely to start out with well-refined cell parameters, but HKL makes up for
this by using a default protocol that continues to refine them as it goes.
Typically you see the fit of predicted spots to observed spots improve as
the integration proceeds.
> Why do you say the default behaviour of Mosflm would be more likely to give incorrect refinement?
I didn't say it would be more likely to give incorrect refinement.
My intended point was that if for some reason the initial refined values were
not ideal for describing later frames in the data collection run, they will
not (so far as I know) be adjusted when those later frames are eventually
encountered. This is in contrast to HKL, where so long as the change is
gradual enough to be tracked, the cell parameters and the corresponding
spot predictions will be continually optimized as the integration proceeds.
Whichever program you are using you can go back and do additional post-refinement
if necessary after integration and initial scaling, which I pointed out was better
than expecting the model refinement program to detect and fix things later using
only the merged data. People do not typically do this, and I think that is
usually fine. The current discussion comes from asking whether you would
get a better final model if more effort were spent in refining the cell parameters.
I am inclined to doubt it, but have only anecdotal evidence to back that up.
cheers,
Ethan
> > Scenario 2)
> > Suppose the true cell parameters do not vary but for some reason they
> > were incorrectly refined during indexing and not adjusted during data
> > integration. I think this is more likely for mosflm than for HKL given
> > the default behaviour of the two programs.
>
> Harry
> --
> Dr Harry Powell
> Chairman of International Union of Crystallography Commission on Crystallographic Computing
> Chairman of European Crystallographic Association SIG9 (Crystallographic Computing)
>
> > On 26 Jul 2016, at 19:04, Ethan A Merritt <[log in to unmask]> wrote:
> >
> >> On Tuesday, 26 July, 2016 17:20:01 Keller, Jacob wrote:
> >> It seems, Robbie, that you did a similar manual exercise to Ethan's? How hard would it be to actually implement a cell refinement stage in refinement? Seems actually pretty obvious to try, so it's probably been done, but I wonder why it's not routine. Maybe not worth the trouble (although it doesn't seem to be much trouble, really: 3-6 extra parameters).
> >
> > In principle these parameters may vary over the course of a data collection.
> > To refine them properly you need access to the unmerged data so that the
> > frame index or time of collection is attached to the observation.
> > The data integration program can do this (HKL or mosflm or xds or whatever),
> > and that's where you can now look for cell parameter error estimates.
> >
> > Scenario 1)
> > Suppose the true cell parameters do vary and the reported esds from
> > integration reflect this. How would you use this information in subsequent
> > refinement against the merged intensity data?
> >
> > Scenario 2)
> > Suppose the true cell parameters do not vary but for some reason they
> > were incorrectly refined during indexing and not adjusted during data
> > integration. I think this is more likely for mosflm than for HKL given
> > the default behaviour of the two programs. In this case it is possible
> > that you could detect and correct the error during model refinement,
> > but that seems less promising than going back to the integration
> > program and doing additional post-refinement before merging the data.
> >
> > Ethan
> >
> >>
> >> JPK
> >>
> >>
> >> *******************************************
> >> Jacob Pearson Keller, PhD
> >> Research Scientist
> >> HHMI Janelia Research Campus
> >> email: [log in to unmask]
> >> *******************************************
> >>
> >> ________________________________________
> >> From: CCP4 bulletin board [[log in to unmask]] on behalf of Robbie Joosten [[log in to unmask]]
> >> Sent: Tuesday, July 26, 2016 1:03 PM
> >> To: [log in to unmask]
> >> Subject: Re: [ccp4bb] Why don't we quote errors on unit cell constants in MX
> >>
> >> Hi Ethan,
> >>
> >> WHAT IF/ WHAT_CEHCK indeed still checks for systematic deviations in bond
> >> lengths, but the output deformation matrix is not constrained to the space
> >> group which doesn't make it easy to fix your cell dimensions. For refinement
> >> I have seen some, but not a lot, effect when you change your cell dimensions
> >> based on WHAT_CHECK's suggestions. This is partly due to the way I refine
> >> models: refining for the best likelihood with reasonable bond length rmsZ,
> >> gives enough slack to get good R-factors provided the cell dimensions are
> >> not terribly far off. The low rmsd fetishists will benefit more from
> >> correcting the cell dimensions because over-restraining your bond lengths
> >> will lead to worse fit with the data.
> >>
> >> Cheers,
> >> Robbie
> >>
> >>> -----Original Message-----
> >>> From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of
> >>> Ethan Merritt
> >>> Sent: Tuesday, July 26, 2016 18:38
> >>> To: [log in to unmask]
> >>> Subject: Re: [ccp4bb] Why don't we quote errors on unit cell constants in
> >> MX
> >>>
> >>>> On Tuesday, 26 July 2016 03:32:52 PM Thomas, Leonard M. wrote:
> >>>> I have had this discussion with my small molecule colleague many times
> >>> and have wondered myself why they are not reported. They are determined
> >>> in HKL2000 and XDS, I don't recall if they are in moslfm and there is a
> >> mmCIF
> >>> line for them. But once converted to an mtz they are basically ignored.
> >>>
> >>> The uncertainty in cell parameters determined by the data collection
> >>> program is confounded by uncertainty in many other experimental
> >>> parameters; notably wavelength and detector distance but also detector x/y
> >>> positional readout accuracy. Worse yet, they may drift over the course of
> >>> data collection (the wavelength certainly, but also the cell parameters
> >>> themselves).
> >>>
> >>> This is largely masked by the discrete nature of the Bragg indices hkl.
> >>> A 1% error in cell parameter or detector distance might prevent you from
> >>> indexing the crystal at all, but it does not translate into uncertainty in
> >> the
> >>> index values per se. That is, you don't get index uncertainties h = 1 ±
> >> 0.01, k
> >>> = 2 ± .02 and so on.
> >>> Instead an error in cell parameter or wavelength shows up in the
> >> calculated
> >>> phases. But the uncertainty in phases is already large due to our
> >> imperfect
> >>> models and the magnitude of this added error is tiny, albeit systematic.
> >>>
> >>> What would you do with such numbers if they were reported?
> >>> I can imagine that recurring large errors on a particular beamline might
> >>> indicate a fixable problem with the detector positioning or monochromator
> >>> stability, but I am dubious it tells you anything about the reliability of
> >>> structures determined from those data sets.
> >>>
> >>> Stepping back from theoretical arguments to anecdotal evidence...
> >>> Years ago WHATIF used to (maybe still does?) analyze systematic deviation
> >>> from ideal geometry in a refined model as a function of projection along
> >> the
> >>> crystal axes, thereby suggesting an error in that cell parameter.
> >>> But my experience has been that after changing the cell parameters
> >>> accordingly, re-refining the structure, and re-analyzing in WHATIF the
> >>> systematic deviation was only shifted, not improved.
> >>> In my hands repeated cycles of cell parameter adjustment and re-refinement
> >>> neither converged nor improved the overall model statistics.
> >>> So I chalked this up as an interesting idea that didn't work out in
> >> practice,
> >>> and I have not gone back to it since then.
> >>>
> >>> Ethan
> >>>
> >>>>
> >>>>
> >>>> Leonard M. Thomas Ph.D.
> >>>> Macromolecular Crystallography Laboratory Oklahoma COBRE in Structural
> >>>> Biology Price Family Foundation Institute of Structural Biology
> >>>> Department of Chemistry and Biochemistry University of Oklahoma
> >>>> Stephenson Life Sciences Research Center
> >>>> 101 Stephenson Parkway
> >>>> Norman, OK 73019
> >>>> 405-325-1126
> >>>> [log in to unmask]
> >>>> http://barlywine.chem.ou.edu
> >>>> http://structuralbiology.ou.edu
> >>>>
> >>>> ________________________________________
> >>>> From: CCP4 bulletin board <[log in to unmask]> on behalf of Keller,
> >>>> Jacob <[log in to unmask]>
> >>>> Sent: Tuesday, July 26, 2016 10:22:42 AM
> >>>> To: [log in to unmask]
> >>>> Subject: Re: [ccp4bb] Why don't we quote errors on unit cell constants
> >>>> in MX
> >>>>
> >>>> I've wondered a couple of things on this score:
> >>>>
> >>>> --What is the general range in uncertainties of cell parameters? I
> >> assume
> >>> there's a huge range as a function of resolution esp., but it would be
> >> nice to
> >>> have a general feeling for it.
> >>>>
> >>>> --Has anyone experimented with implementing a cell parameter step in
> >>> refinement? I would think the effect would be small but still perhaps
> >>> significant, especially at lower resolution and/or when cell parameters
> >> are
> >>> not known very precisely.
> >>>>
> >>>>
> >>>> *******************************************
> >>>> Jacob Pearson Keller, PhD
> >>>> Research Scientist
> >>>> HHMI Janelia Research Campus
> >>>> email: [log in to unmask]
> >>>> *******************************************
> >>>>
> >>>> ________________________________________
> >>>> From: CCP4 bulletin board [[log in to unmask]] on behalf of Tim
> >>>> Gruene [[log in to unmask]]
> >>>> Sent: Tuesday, July 26, 2016 11:15 AM
> >>>> To: [log in to unmask]
> >>>> Subject: Re: [ccp4bb] Why don't we quote errors on unit cell constants
> >>>> in MX
> >>>>
> >>>> Dear Graeme,
> >>>>
> >>>> XDS reports standard uncertainties of the unit cell parameters in
> >>>> CORRECT.LP and I recently started to include this information in
> >>>> publications. So at least one of us does ;-)
> >>>>
> >>>> I also include this information in the SHELXL ZERR keyword - as fas as
> >>>> I understand it affects the s.u. for coordinates and other things you
> >>>> can calculate with SHELXL.
> >>>>
> >>>> The information is not too obvious to access, and it only have limited
> >>>> information content. Maybe that's why it isn't quoted.
> >>>>
> >>>> Best,
> >>>> Tim
> >>>>
> >>>>
> >>>>> On Tuesday, July 26, 2016 03:07:03 PM Graeme Winter wrote:
> >>>>> Dear CCP4BB
> >>>>>
> >>>>> Does anyone know why we don't quote standard uncertainties on unit
> >>>>> cell constants in the way that the small molecule community do? It
> >>>>> would seem in the new world of multi-crystal data sets and serial
> >>>>> crystallography some idea of the measure of ignorance would be
> >>> particularly valuable.
> >>>>>
> >>>>> I'm not worried about whether they are "right" or "true" just
> >>>>> interested in why we don't quote them...
> >>>>>
> >>>>> An example for thaumatin may look like this, for example:
> >>>>>
> >>>>> Unit cell (with estimated std devs):
> >>>>> 57.7841(1) 57.7841(1) 149.9963(3)
> >>>>> 90.0000(0) 90.0000(0) 90.0000(0)
> >>>>>
> >>>>> (in other news, there is no place to store this information in an
> >>>>> MTZ
> >>>>> file...)
> >>>>>
> >>>>> Thanks & best wishes Graeme
> >>>>>
> >>>>> --
> >>>>> This e-mail and any attachments may contain confidential, copyright
> >>>>> and or privileged material, and are for the use of the intended
> >>>>> addressee only. If you are not the intended addressee or an
> >>>>> authorised recipient of the addressee please notify us of receipt by
> >>>>> returning the e-mail and do not use, copy, retain, distribute or
> >>>>> disclose the information in or attached to the e-mail. Any opinions
> >>>>> expressed within this e-mail are those of the individual and not
> >>>>> necessarily of Diamond Light Source Ltd. Diamond Light Source Ltd.
> >>>>> cannot guarantee that this e-mail or any attachments are free from
> >>>>> viruses and we cannot accept liability for any damage which you may
> >>>>> sustain as a result of software viruses which may be transmitted in
> >>>>> or with the message. Diamond Light Source Limited (company no.
> >>>>> 4375679). Registered in England and Wales with its registered office
> >>>>> at Diamond House, Harwell Science and Innovation Campus, Didcot,
> >>>>> Oxfordshire, OX11 0DE, United Kingdom
> >>>> --
> >>>> --
> >>>> Paul Scherrer Institut
> >>>> Dr. Tim Gruene
> >>>> - persoenlich -
> >>>> OFLC/102
> >>>> CH-5232 Villigen PSI
> >>>> phone: +41 (0)56 310 5297
> >>>>
> >>>> GPG Key ID = A46BEE1A
> >>>
> >>> --
> >>> mail: Biomolecular Structure Center, K-428 Health Sciences Bldg
> >>> MS 357742, University of Washington, Seattle 98195-7742
> >>
> >
>
--
Ethan A Merritt
Biomolecular Structure Center, K-428 Health Sciences Bldg
MS 357742, University of Washington, Seattle 98195-7742
|
