Re: [ccp4bb] Quote source inquiry

[Jessica Bruhn]

Hi Garib, Tim and James,

Thank you for the helpful information. I look forward to testing this out
on some of our data. Hopefully it helps!

To Garib, I think that electron diffraction/microED of small molecules is
actually in a pretty good position for this technique to really take off.
To your concerns:
1. Our group is personally very happy with the data coming from our
detector (CETA-D). The thicker scintillator really seems to have helped.
And there are other good detectors out there. I have some data posted in
zenodo in case you are interested (10.5281/zenodo.3905397 and
10.5281/zenodo.3937740).
2. Crystal handling is thankfully very straightforward for dry, small
molecule crystals. Just dab a TEM grid on some (crystalline) powder and in
most cases you should be ready to collect. Protein crystals are
unfortunately significantly more difficult to work with. We'll see how work
in that area progresses...
3. As for rotation, I am curious to hear what concerns you have about
rotation? Are you concerned about completeness? Or the lower data quality
in the high tilt angle frames? Or the accuracy of the goniometer with
regard to position and constant speed maintenance? If your concerns are
about completeness, I would say that we have been able to get fairly decent
completeness by combining data from multiple crystals. In our hands (18
small molecule ED structures solved in house), about half of these reached
>95% completeness, another quarter were >90% and the rest were in the
81-90% range.

You may also be interested to know that of these 18 small molecule
structures, three had to be refined in REFMAC5 because the resolution was a
little low (1.2-1.7A) or the data to parameter ratio was too poor for
SHELXL. I understand your time is limited, but I do think that electron
diffraction for small molecules is really gaining momentum. We have
collected data from almost fifty different samples from our clients all
across pharma since installing our new camera in September. Many of these
probably won't end up in public databases, but they have been hugely
impactful for these chemists.

Have a wonderful weekend.

Best wishes,
Jessica



On Fri, Jul 17, 2020 at 2:11 AM Garib Murshudov 
wrote:

> Dear Tim,
>
>
> I understand the problem. If the problem is the distance only then only
> one parameter is needed for refinement of lattice parameters.
>
> I do think that microED has good potential. However engineering problems
> need to be sorted out (detector, crystal handling, rotation etc).
>
> When the problem becomes urgent then I can coniblue working on this
> problem. I have already implemented using all data (chemistry and crystal
> data) for lattice refinement. They need to be tested properly.
> There are several issues that need to be sorted out.
>
> Regards
> Garib
>
>
> On 17 Jul 2020, at 08:29, Tim Gruene  wrote:
>
> Dear Garib,
>
> thank you very much for the details! If everything goes to plan, we are
> going to use the Dectris QUADRO in September(ish), ideally also with
> some protein crystals. In ED, distance calibration is more difficult
> than with X-rays because of instabilities in the lens system (at least
> with the older instruments), and because I do not work with a parallel
> beam, but focus the beam onto the detector surface. This is not a very
> reproducible process. In those cases where I got high resolution data,
> the cell is often quite stable, and the distance can vary by about 5%...
>
> Best regards,
> Tim
>
> On Thu, 16 Jul 2020 23:21:54 +0100
> Garib Murshudov  wrote:
>
> One correction: Model should after molecular replacement and few
> cycles of refinement (perhaps with a little bir relaxed geometry, but
> not too much).
>
> There is an option to use a model after molecular replacement but it
> is being migrated to another program that will have proper tests.
>
> Regards
> Garib
>
>
> On 16 Jul 2020, at 22:34, Garib Murshudov 
> wrote:
>
> Hi Tim,
>
> There is an option to do unit cell parameter refinement (for all
> six parameters in general which can only happen in P1). It is
> undocumented.
>
> Celrefine/lattice refine all # if you give scale instead of all
> then only one parameter is refined.
>
> Cellrefine select .  # use only atomic B value < Bmedian +
> alpha * Binterquartile_range
>
>
> The last command was added to to reduce effect of wrong atoms.
>
> These command should enable refinement all parameters with due
> account for symmetry. But I am worried about refinement of cell
> parameters using atomic models. The problem is that it affects B
> values and I think if model is not good then cells will be expanded
> to reduce non-bonding interactions. I think it will give biased
> results. For ideal case (when you deliberately change cell
> parameters) it worked perfectly when I tried. However, for real
> cases I could not convince myself to claim that it would give
> unbiased results. My current thought is that either a model after
> molecular replacement should be used or cell 

[ccp4bb] PhD position in time-resolved protein crystallography (ESRF, Grenoble (FR) and U. of Hamburg (DE))

[Antoine Royant]

Dear all,

We have an immediate opening for a PhD position at the ESRF in Grenoble 
(France) in collaboration with the CFEL in Hamburg (Germany) to work on 
methodological developments in time-resolved protein crystallography at 
synchrotrons. For further details, please see here: 
https://esrf.gestmax.eu/1438/1/phd-student-icos-lab-and-id29/en_US?backlink=search 
or below.


Best regards,

Antoine.




 Context & Job description

Time-resolved macromolecular crystallography (TR-MX) at synchrotrons has 
long been limited to the use of the Laue diffraction technique and was 
only successful for a handful of biological systems. The recent advent 
of X-ray free-electron lasers has rejuvenated the field by fostering the 
development of serial crystallography, which consists in composing a 
complete data set from single diffraction images obtained from tens of 
thousands of microcrystals passing through a pulsed X-ray beam. The 
associated development of crystal injection techniques, coupled to the 
advent of faster, noiseless X-ray detectors, has paved the way for the 
revival of TR-MX at synchrotrons. The EBSL8/ID29 beamline (beamline 
responsible, Daniele de Sanctis) currently under construction as part of 
the ESRF EBS Upgrade Project, has been designed to offer a variety of 
techniques and methods for TR-MX with a microsecond time resolution. 
Additionally, the ESRF Structural Biology Group has developed an 
oscillation-based TR-MX method, demonstrating its applicability with the 
study of a photoreceptor fragment at a time resolution of 63 ms 
(Aumonier /et al/., IUCrJ (2020) 
). We 
propose to study this target, as well as other photoreceptor fragments 
and photosensitive proteins, by various TR-MX methods (fixed-target-; 
grease injector-; tape-drive- or oscillation-based) to identify those 
which are most suitable for further development on EBSL8/ID29 and to 
make molecular movies at the lowest time resolution attainable, ideally 
100 μs. This work will be performed in co-supervision with Prof. Arwen 
Pearson (CFEL, University of Hamburg, Germany) with whom previously 
developed methods for TR-MX will be evaluated using beamlines both at 
the ESRF and PETRA III, most particularly at the T-REXX endstation on 
beamline P14, operated by EMBL-Hamburg (Mehrabi /et al/., Nat. Methods 
(2019) ).


Further information may be obtained from Antoine ROYANT (tel.: +33 (0)4 
76 88 17 46, email: roy...@esrf.fr ).



 Expected profile

 * Degree allowing enrollment for a PhD (such as MSc, Master 2 de
   Recherche, Laurea or equivalent) in biology, chemistry, physics or
   closely related science
 * A background in X-ray crystallography is highly desirable as are
   basic skills in biochemistry and molecular biology
 * An interest in spectroscopy as applied to crystals of biological
   macromolecules would be an asset.
 * The candidate should show aptitude for team working and have good
   communications and initiative skills

 * Proficiency in English (working language at the ESRF)


 Working conditions

Contract of two years employed by the ESRF, with a monthly gross salary 
of 2339 €, then of two years employed by the University of Hamburg.


The ESRF is an equal opportunity employer and encourages diversity.


--
Antoine ROYANT
Institut de Biologie Structurale
71 avenue des Martyrs
CS 10090
38044 Grenoble Cedex 9
FRANCE
Phone +33(0)4-76-88-17-46 or +33(0)-4-57-42-85-39
antoine.roy...@esrf.fr or antoine.roy...@ibs.fr




To unsubscribe from the CCP4BB list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCP4BB=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/

Re: [ccp4bb] Quote source inquiry

[Garib Murshudov]

Dear Tim,


I understand the problem. If the problem is the distance only then only one 
parameter is needed for refinement of lattice parameters.

I do think that microED has good potential. However engineering problems need 
to be sorted out (detector, crystal handling, rotation etc). 

When the problem becomes urgent then I can coniblue working on this problem. I 
have already implemented using all data (chemistry and crystal data) for 
lattice refinement. They need to be tested properly. 
There are several issues that need to be sorted out. 

Regards
Garib


> On 17 Jul 2020, at 08:29, Tim Gruene  wrote:
> 
> Dear Garib,
> 
> thank you very much for the details! If everything goes to plan, we are
> going to use the Dectris QUADRO in September(ish), ideally also with
> some protein crystals. In ED, distance calibration is more difficult
> than with X-rays because of instabilities in the lens system (at least
> with the older instruments), and because I do not work with a parallel
> beam, but focus the beam onto the detector surface. This is not a very
> reproducible process. In those cases where I got high resolution data,
> the cell is often quite stable, and the distance can vary by about 5%...
> 
> Best regards,
> Tim
> 
> On Thu, 16 Jul 2020 23:21:54 +0100
> Garib Murshudov mailto:ga...@mrc-lmb.cam.ac.uk>> 
> wrote:
> 
>> One correction: Model should after molecular replacement and few
>> cycles of refinement (perhaps with a little bir relaxed geometry, but
>> not too much).
>> 
>> There is an option to use a model after molecular replacement but it
>> is being migrated to another program that will have proper tests.
>> 
>> Regards
>> Garib
>> 
>> 
>>> On 16 Jul 2020, at 22:34, Garib Murshudov 
>>> wrote:
>>> 
>>> Hi Tim,
>>> 
>>> There is an option to do unit cell parameter refinement (for all
>>> six parameters in general which can only happen in P1). It is
>>> undocumented. 
>>> 
>>> Celrefine/lattice refine all # if you give scale instead of all
>>> then only one parameter is refined.
>>> 
>>> Cellrefine select .  # use only atomic B value < Bmedian +
>>> alpha * Binterquartile_range
>>> 
>>> 
>>> The last command was added to to reduce effect of wrong atoms.
>>> 
>>> These command should enable refinement all parameters with due
>>> account for symmetry. But I am worried about refinement of cell
>>> parameters using atomic models. The problem is that it affects B
>>> values and I think if model is not good then cells will be expanded
>>> to reduce non-bonding interactions. I think it will give biased
>>> results. For ideal case (when you deliberately change cell
>>> parameters) it worked perfectly when I tried. However, for real
>>> cases I could not convince myself to claim that it would give
>>> unbiased results. My current thought is that either a model after
>>> molecular replacement should be used or cell parameters should be
>>> refined outside using data only (then you can only make cell
>>> parameters consistent with each other). 
>>> 
>>> If you still would want to use refmac to do this calculations then
>>> you should do it iteratively. Refine cell, then change mtz file
>>> (different cell parameters) then refine cell again. Please also
>>> note that if you are refining cell parameters then the resolution
>>> of the data will also change (if you are refining all six
>>> parameters then changes will be anisotropic)
>>> 
>>> 
>>> In general it is a trivial but extremely trivial problem.
>>> 
>>> Regards
>>> Garib
>>> 
>>> 
>>> 
 On 16 Jul 2020, at 18:31, Tim Gruene >>> >> wrote:
 
 Hi Jessica,
 
 Jens Luebben wrote cellopt for this purpose. It is available from
 github, https://github.com/JLuebben/CellOpt 
 
 >
 
 It is based on the idea available in whatcheck, i.e. to optimise
 the unit cell parameters based on geometry restraints DFIX/DANG.
 Those need to be three-dimensional: I've had cases where the
 restraints do not span all three dimensions. In this case one of
 the cell parameters can refine to unrealistic values. 
 
 We are quite slow on the manuscript for its reference, but for the
 time being, please quote the nanoArgovia (www.nanoscience.ch 
 
 >) project A12.01 
 (A3EDPI).
 
 cellopt is called liked a shelxl job, i.e. like 'cellopt name'
 where name.ins and name.hkl are present in the directory. You can
 add some constraints to the lattice type. 
 
 Refmac5 can also refine the unit cell parameters (Max Clabbers has
 made use of this feature), but as far as I understand, refmac5
 only scales the unit cell volume isotropically - I am happy to be
 corrected.
 
 When you resolution is quite high, say 

Re: [ccp4bb] Quote source inquiry [SEC=UNCLASSIFIED]

[Harry Powell - CCP4BB]

Hi Tom

Welcome to the old folks club!

There are a few points in your post that I think are incredibly relevant here, 
and worth picking out for the casual reader - 

> it takes so little time to calibrate using powder diffraction 

Exactly! For a user collecting on a modern beamline that can collect a dataset 
in seconds, the time spent on this step falls below their conscious threshold, 
so why not do it?

> It can be so hard for users to grow crystals and we want to make sure the 
> data quality for our users is the best possible. 

Again, “exactly”. I haven’t been on a data collection and processing course in 
the last decade where _every_ data processing developer hasn’t said something 
along the lines of “data collection is the _last_ experimental step (often of a 
long and painful process) - everything after this is computing, and can be 
repeated ad nauseam”. 

> I also remember getting really frustrated in the old days as a user with 
> incorrect beam position and/or detector distance in image headers and didn't 
> want our users to have to deal with that.

This was really brought home to me when I had a dataset given to me by a user 
which had been collected at a well-known source (which shall remain nameless, 
to protect the guilty…) where _every_ useful piece of metadata in the image 
header was incorrect - wavelength, beam centre, crystal-to-detector distance. 
Fortunately, the user had noted which (fixed wavelength) beamline they had 
actually used, and the data were rescued (after trials of different crystal to 
detector distances & beam centre in the data processing).

In the Mosflm coding I had (for many years) a set of “Trusted” detectors where 
the header values were (at the very least) good approximations to “true” - all 
other detector headers were considered unreliable. Of course, XDS (uniquely?) 
has always ignored the header information, but this has its own problems.

Harry

> On 16 Jul 2020, at 22:49, CARADOC-DAVIES, Tom  wrote:
> 
> Hi Harry,
> 
> I laughed when I read your question below "Or is this considered just 
> something that old folk do?".
> 
> At the Australian Synchrotron MX beamlines (MX1 and MX2) we also go 
> old-school and collect Lanthanum hexaboride powder diffraction data during 
> each user setup. We collect a single 180 degree image of Lab6 at minimum 
> detector distance and then up to 500mm in 100mm steps. They we can analyse 
> the data (using automated fit2d scripts) to refine direct beam position on 
> the detector and the detector distance offset from reported distance. The 
> lab6 pin is stored in the robot dewar in a staff puck and mounting and 
> collecting the 6-8 images is quick (maybe 2 minutes?) as we have automated 
> collection and processing that runs via our user setup GUI.
> This process means that values for distance and direct beam in the image 
> headers for each user experiment have been experimentally determined that 
> morning. This accuracy helps the reliability of our auto-processing 
> (especially for dodgy crystals). Energy is calibrated each user run or if the 
> LaB6 data suggests a change in "distance". 
> 
> This is probably overkill but dates from the early days when we were building 
> user confidence in the beamlines. I also remember getting really frustrated 
> in the old days as a user with incorrect beam position and/or detector 
> distance in image headers and didn't want our users to have to deal with 
> that. I also spent many hours trying to refine beamX/Y from ice rings in 
> imosflm from beamlines with a postit-note stuck to the side of the monitor 
> with cryptic beamX/Y values.
> 
> We do get teased by some of our beamline scientist colleagues for our very 
> extensive beamline setup for each user (we could probably go to monthly for 
> some checks and weekly for others) but it takes so little time to calibrate 
> using powder diffraction we have just kept doing it.
> 
> If you think that is not pedantic enough we also collect a full dataset of 
> cubic insulin for every setup and the user setup report shows the data 
> processing statistics (the users also get the raw data and auto-processing 
> files). If we cannot get really nice data (low Rmerge at low resolution and 
> good low res anomalous signal at 13keV on sulfur) on cubic insulin won't 
> release the beamline to users until we can investigate and fix the issue. It 
> is also quick as we have Eigers and collections are so fast now.
> 
> I guess it is just part of our user support philosophy. It can be so hard for 
> users to grow crystals and we want to make sure the data quality for our 
> users is the best possible. Most users don't get the attention to detail that 
> goes on "under the hood" at the beamlines and this is how it should be. It is 
> our job to handle that stuff and success means users just expect good data 
> from good crystals and they get it. If you don't look for problems in the 
> beamlines you don’t find them. I think lot of MX 

Re: [ccp4bb] Quote source inquiry

[Tim Gruene]

Dear Garib,

thank you very much for the details! If everything goes to plan, we are
going to use the Dectris QUADRO in September(ish), ideally also with
some protein crystals. In ED, distance calibration is more difficult
than with X-rays because of instabilities in the lens system (at least
with the older instruments), and because I do not work with a parallel
beam, but focus the beam onto the detector surface. This is not a very
reproducible process. In those cases where I got high resolution data,
the cell is often quite stable, and the distance can vary by about 5%...

Best regards,
Tim

On Thu, 16 Jul 2020 23:21:54 +0100
Garib Murshudov  wrote:

> One correction: Model should after molecular replacement and few
> cycles of refinement (perhaps with a little bir relaxed geometry, but
> not too much).
> 
> There is an option to use a model after molecular replacement but it
> is being migrated to another program that will have proper tests.
> 
> Regards
> Garib
> 
> 
> > On 16 Jul 2020, at 22:34, Garib Murshudov 
> > wrote:
> > 
> > Hi Tim,
> > 
> > There is an option to do unit cell parameter refinement (for all
> > six parameters in general which can only happen in P1). It is
> > undocumented. 
> > 
> > Celrefine/lattice refine all # if you give scale instead of all
> > then only one parameter is refined.
> > 
> > Cellrefine select .  # use only atomic B value < Bmedian +
> > alpha * Binterquartile_range
> > 
> > 
> > The last command was added to to reduce effect of wrong atoms.
> > 
> > These command should enable refinement all parameters with due
> > account for symmetry. But I am worried about refinement of cell
> > parameters using atomic models. The problem is that it affects B
> > values and I think if model is not good then cells will be expanded
> > to reduce non-bonding interactions. I think it will give biased
> > results. For ideal case (when you deliberately change cell
> > parameters) it worked perfectly when I tried. However, for real
> > cases I could not convince myself to claim that it would give
> > unbiased results. My current thought is that either a model after
> > molecular replacement should be used or cell parameters should be
> > refined outside using data only (then you can only make cell
> > parameters consistent with each other). 
> > 
> > If you still would want to use refmac to do this calculations then
> > you should do it iteratively. Refine cell, then change mtz file
> > (different cell parameters) then refine cell again. Please also
> > note that if you are refining cell parameters then the resolution
> > of the data will also change (if you are refining all six
> > parameters then changes will be anisotropic)
> > 
> > 
> > In general it is a trivial but extremely trivial problem.
> > 
> > Regards
> > Garib
> > 
> > 
> >   
> >> On 16 Jul 2020, at 18:31, Tim Gruene  >> > wrote:
> >> 
> >> Hi Jessica,
> >> 
> >> Jens Luebben wrote cellopt for this purpose. It is available from
> >> github, https://github.com/JLuebben/CellOpt
> >> 
> >> 
> >> It is based on the idea available in whatcheck, i.e. to optimise
> >> the unit cell parameters based on geometry restraints DFIX/DANG.
> >> Those need to be three-dimensional: I've had cases where the
> >> restraints do not span all three dimensions. In this case one of
> >> the cell parameters can refine to unrealistic values. 
> >> 
> >> We are quite slow on the manuscript for its reference, but for the
> >> time being, please quote the nanoArgovia (www.nanoscience.ch
> >> ) project A12.01 (A3EDPI).
> >> 
> >> cellopt is called liked a shelxl job, i.e. like 'cellopt name'
> >> where name.ins and name.hkl are present in the directory. You can
> >> add some constraints to the lattice type. 
> >> 
> >> Refmac5 can also refine the unit cell parameters (Max Clabbers has
> >> made use of this feature), but as far as I understand, refmac5
> >> only scales the unit cell volume isotropically - I am happy to be
> >> corrected.
> >> 
> >> When you resolution is quite high, say 0.7A like what we get for
> >> zeolites and some organic compounds, you can refine the cell and
> >> the distance simultaneously, only the BEAM correlates heavily with
> >> the distance. DIALS can produce plots for the correlation between
> >> refined parameters, which is very handy for electron diffraction
> >> data.
> >> 
> >> Best wishes,
> >> Tim
> >> 
> >> 
> >> On Thu, 16 Jul 2020
> >> 08:20:11 -0700 Jessica Bruhn
> >> <450e5de75376-dmarc-requ...@jiscmail.ac.uk
> >> > wrote: 
> >>> As someone working with continuous rotation electron diffraction,
> >>> mostly with small molecules, I am often very concerned about the
> >>> accuracy of my cell dimensions. I have to heavily restrain my
> >>> experimental geometry, including the detector distance, because
> >>> they are so unusual compared to X-ray setups. I