Re: [ccp4bb] High Rmerge with thin frames
Am 20:59, schrieb Harry Powell: ... I think there may be issues with collecting data too finely with a Pilatus, even in shutterless mode. I don't know where the problems arise (can't be shutter/rotation axis synchronisation), but it seems to be the normal thing in XDS (which should have no problems with fine phi-slicing) to use the PATCH_SHUTTER_PROBLEM=TRUE that Martin Hallberg suggested, which looks a bit like a fudge to me (but I expect Kay to correct me on that!). ... Hi Harry, no, it is a misunderstanding that the normal thing is to use PATCH_SHUTTER_PROBLEM=TRUE in XDS; rather, it may be the last resort to try before you abandon a dataset, and if the spindle/shutter de-synchronization is so poor that this fudge needs to be used, then the BL hardware needs to be fixed before other datasets are collected. Two other points: a) one may distinguish weak data resulting in high R-factors from problems with hardware or radiation damage resulting in high R-factors by looking at (I/sigma)^asymptotic as defined in Acta D66, 733 ( http://dx.doi.org/10.1107/S0907444910014836 ). b) Marcus Müller (SLS) has shown (to me, unambiguously) that the best data are obtained from the Pilatus if delta-phi is 1/2 to 1/4 of the mosaicity (as reported by XDS). However, for the best data, other parameters of the experiment (basically the transmission and the spindle speed) have to be optimized as well. The finding of Tassos that 0.5-1 degree data gave the best result in a particular case might indicate that these other factors of his experiment were not optimal (it is indeed not trivial to get them right). best, Kay smime.p7s Description: S/MIME Cryptographic Signature
Re: [ccp4bb] High Rmerge with thin frames
Hi All Both Martin and Kay ( in a later message) have misinterpreted what I wrote - what I meant was that it seems normal in using XDS with Pilatus data, not the normal thing to do with data from other detectors. I had found a number of scripts on the web that deal specifically with processing Pilatus data with XDS that seemed to use PATCH_SHUTTER_PROBLEM=TRUE - what I didn't notice was that the commands had all been commented out, and that FALSE, the default, was in operation. So, in either case I was wrong. Apologies to Kay and Wolfgang, and all who write and use scripts for processing Pilatus data with XDS. On 5 Nov 2010, at 18:33, Martin Hallberg wrote: Hi Harry, On Nov 5, 2010, at 5:45 PM, Harry Powell wrote: I think there may be issues with collecting data too finely with a Pilatus, even in shutterless mode. I don't know where the problems arise (can't be shutter/rotation axis synchronisation), but it seems to be the normal thing in XDS (which should have no problems with fine phi-slicing) to use the PATCH_SHUTTER_PROBLEM=TRUE that Martin Hallberg suggested, which looks a bit like a fudge to me (but I expect Kay to correct me on that!). It is not the normal thing in XDS but it is perhaps a relatively common solution to shutter/spindle synchronisation problems discovered afterwards (always process directly at the beam line!). The default in XDS is indeed PATCH_SHUTTER_PROBLEM=FALSE Compensating like this is of course not the best (go and recollect!) but still way better than unusable data in the meantime. In the case Sergei originally described it would at least indicate what the problem may be. Sergei did not say which detector was used for the data collection so we don't know if it was a Pilatus or a CCD. Maybe Sergei can fill us in on the details? BTW, any views in the community on crystal lifetime with continuous data collection like on the Pilatus (or AXIOM) versus letting the crystal rest/cool/die(?) a second between frames on a CCD? Cheers, Martin Harry -- Dr Harry Powell, MRC Laboratory of Molecular Biology, Hills Road, Cambridge, CB2 0QH
Re: [ccp4bb] High Rmerge with thin frames
Dear Ronnie, we are working with weak diffracting crystals. Many tests (and taking into account Marcus Muellers results) showed that 0.3-0.5 of XDS mosaicity, (very) low dose, and high redundancy give the best results. Our crystals diffract between 7 and 4 A. At low dose I do not check diffraction really by eye. I collect 200 images (fine slicing), process and check XDS statistics. (200 images is not much when the dataset comprises 9000-12000 images). Especially for anomalous data this strategy was pivotal. We did also comparisons between CCD and PILATUS. We used one long crystal and collected on end end a MAD dataset using a CCD and afterwards at the other end a MAD dataset on PILATUS, same dose per degree. Results were absolutely clear that the fine slicing give much better results. Best, Guenter Dear Tassos, I'm interested in your third point. Do you have any explanation for why 0.5-1 degrees oscillation gave better data? Purely due to the fact that the crystals survived longer and thus yielded higher redundancy data, or also other parameters? Also do anyone know where the threshold lies for when /not/ to use fine phi slicing on the PILATUS? ie, at what level of diffraction would one need to increase the exposure (and oscillation in order to still get redundant data)? We'll be in a similar position in the coming weeks with data collection using PILATUS detectors, and would like to maximize the potential data quality from our weak diffracting crystals. Any input on this would be greatly appreciated! Cheers, Ronnie Berntsson On Nov 5, 2010, at 16:16, Anastassis Perrakis wrote: three additional points: 1. OTOH, if The diffraction is quite weak, one may be limited by counting statistics. This also cannot be overcome by processing. As JIm suggests above then, maybe you should look if the 15% Rmerge is almost reasonable given the specific I/sigI at low resolution? 2. If there is one thing I do not like in XDS, is that there is no (or I have failed to find) statistics of I/sigI and Rmerge as function of image. Have a look at the SCALA output. Maybe some images are bad? 3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. A. -- *** Priv.Doz.Dr. Guenter Fritz Fachbereich Biologie Universitaet Konstanz http://www.biologie.uni-konstanz.de/fritz e-mail: guenter.fr...@uni-konstanz.de e-mail: guenter.fr...@uniklinik-freiburg.de http://www.uniklinik-freiburg.de/neuropathologie/live/forschung/ag-g-fritz.html Tel.: +49 761 270 5078 Fax.: +49 761 270 5050
Re: [ccp4bb] High Rmerge with thin frames
Dear All, first of all, I would like to thank the many good people who have responded to my query. Yet another truly interesting discussion on this BB! As a partial summary, two points: 1. Few people suggested that our high Rmerge problem could be caused by experimental troubles like phi angle imprecision rather than the crystal itself. Thus far we could not find indications for that but maybe we did not look carefully enough. However, the frames ARE weak, and I tend to think that this is the main cause of the problem. 2. Few people argued that data processing programs (XDS in particular) 'should' handle thin frames as efficiently as thicker frames... After digesting these (very useful!) arguments, I still tend to think that the best proof would be in the pudding, i.e. trying to pool thin frames into thicker frames (but I do not have an immediate means of doing this... ) Sergei Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei -- Prof. Sergei V. Strelkov Laboratory for Biocrystallography Department of Pharmaceutical Sciences, Katholieke Universiteit Leuven ON2, Campus Gasthuisberg, Herestraat 49 bus 822, 3000 Leuven, Belgium Work phone: +32 16 330845 Fax: +32 16 323469 OR +32 16 323460 Mobile: +32 486 294132 Lab pages: http://pharm.kuleuven.be/anafar
Re: [ccp4bb] High Rmerge with thin frames
Am 20:59, schrieb Sergei Strelkov: ... I still tend to think that the best proof would be in the pudding, i.e. trying to pool thin frames into thicker frames (but I do not have an immediate means of doing this... ) Sergei, try to locate 2pck and 2pck.man which were part of former XDS distributions (there's a link to a Linux binary in article http://strucbio.biologie.uni-konstanz.de/xdswiki/index.php/2pck ). 2pck can add several images, and writes them out as one new image. After the conversion, you may only have to change the filetype, the oscillation range and the number of images in your XDS.INP . Pls tell us what you find! HTH, Kay smime.p7s Description: S/MIME Cryptographic Signature
[ccp4bb] High Rmerge with thin frames
Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei
Re: [ccp4bb] High Rmerge with thin frames
Dear Sergei, with only 3A data and 0.1 deg frame width my first guess would be radiation damage. In that case there is little you can do - the Rmerge might just be realistic. XDS has not problem dealing with thin frames (on the contrary!) and it won't help pooling frames together. Check out the statistics and plots from scala, they should tell you more about whether e.g. the scaling per frame increases with frame number, which is quite a sure sign of radiation damage. I would suggest to pick a new crystal and collect data with 0.5deg/frame. Tim On Fri, Nov 05, 2010 at 09:40:57AM +0100, Sergei Strelkov wrote: Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei -- -- 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] High Rmerge with thin frames
Dear Sergei, since your Rmerge is high at low resolution even in P1, my guess is, that there was a problem either with the crystal or with the data collection. Fine slicing should improve the data quality, because your get a better description of the reference profiles and reduce the background. Different merging of fine-sliced images into coarser-sliced images was already systematically done by Wolfgang Kabsch, with the result, that finer slices gave better results, although slices much finer than half the full mosaicity angle of a reflection didn't improve the results much further (I've heard a talk given by Wolfgang, but never found a publication about these results). There is a publication in 1999 about fine-slicing by Jim Pflugrath, that you might want to read about this (Acta Cryst D55, 1718-1725). So, merging fine-sliced images into coarse-sliced probably won't help. Anyway, good luck! Dirk. Am 05.11.10 09:40, schrieb Sergei Strelkov: Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei -- *** Dirk Kostrewa Gene Center Munich, A5.07 Department of Biochemistry Ludwig-Maximilians-Universität München Feodor-Lynen-Str. 25 D-81377 Munich Germany Phone: +49-89-2180-76845 Fax:+49-89-2180-76999 E-mail: kostr...@genzentrum.lmu.de WWW:www.genzentrum.lmu.de ***
Re: [ccp4bb] High Rmerge with thin frames
Dear Sergei how much do the refined unit cell parameters (given a fixed detector distance) vary as a function of frame number? We have been using such initial diagnostic approach to trace radiation damage issues (among other problems) for a number of crystal forms that maximally diffracted in the 3.8-4.5 angstron range. I copy/paste an excerpt from the XDS wiki that reinforces this approach: REFINE(INTEGRATE) The defaults (REFINE(INTEGRATE)=DISTANCE BEAM ORIENTATION CELL) could be modified by omitting DISTANCE, because one should assume that the distance is constant. This is particularly recommended if SPACE_GROUP_NUMBER=0 or 1. Furthermore, by fixing the distance one can better see from the results of the refinement whether the cell parameters are stable, or whether they change due to radiation damage. There are situations when one wants to reduce the number of parameters to be refined even more, see Optimization. Best regards Savvas Savvas Savvides Unit for Structural Biology @ L-ProBE Ghent University K.L. Ledeganckstraat 35, 9000 Ghent, Belgium Ph. +32 (0)472 928 519 http://www.LProBE.ugent.be/xray.html On 05 Nov 2010, at 09:40, Sergei Strelkov wrote: Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei
Re: [ccp4bb] High Rmerge with thin frames
Dear Sergei, Did you check the mean intensity as function of spindle position statistics in the CORRECT.LP file? Any (even minute) shutter problems will affect these thin frames significantly. If this is indeed the problem, you could then try to set: PATCH_SHUTTER_PROBLEM=TRUE for the CORRECT stage in XDS. Cheers, Martin On Nov 5, 2010, at 9:40 AM, Sergei Strelkov wrote: Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei
[ccp4bb] RE : [ccp4bb] High Rmerge with thin frames
Dear Sergei, It is difficult to say without looking at XDS or MOSFLM logfiles, but this sort of problem (high flat Rsym in all resolution bins) sounds like the crystal vibrating wildly in the cryostream. You could ask the data collector the following questions: 1) Was the cryostream misaligned or too far from the sample? 2) Is the crystal mounted in a long wobbly loop? 3) Was the incident beam varying (in XDS check the average background count for each image)? I wrote a web page to troubleshoot this sort of problem when I was beamline scientist on ID29 at the ESRF. Unfortunately the pages are no longer available, but I might have a backup if you are interested. Cheers, Bill De: CCP4 bulletin board de la part de Sergei Strelkov Date: ven. 05/11/2010 09:40 À: CCP4BB@JISCMAIL.AC.UK Objet : [ccp4bb] High Rmerge with thin frames Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei
Re: [ccp4bb] High Rmerge with thin frames
Hi Sergei, such fine slicing during data collection would suggest a large cell. How many reflections are you merging? And what is the redundancy (in the expected symmetry)? Rmerge tends to go up with more reflections added. Peter On 5 Nov 2010, at 08:40, Sergei Strelkov wrote: Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. Either Mosflm or XDS processes the data readily with +/- default settings but both yield a high overall Rmerge of about 0.23 in the expected symmetry. Processing in P1 yields an overall Rmerge of ~0.18, but what is especially disappointing is that Rmerge is as high as 0.15 at ~5A resolution already. The question is, how can we process the data so that the merging statistics becomes more reasonable? Apparent mosaicity turns out to be ~0.5A. My naive way of thinking is to try treating each five consecutive frames as a single 0.5 degree frame. Does anyone have experience with this? Many thanks in advance, Sergei
Re: [ccp4bb] High Rmerge with thin frames
In general, if the Rmeas or Rmerge is high in the low resolution shells, then something is not optimal with the data collection. Bill Shepard has already mentioned the loop vibrating or moving in the cryogenic gas flow. Other problems could be the goniometer head was loose, the magnet was loose, the pin was loose, etc. There could be excessive shutter shutter. The shutter and the crystal rotation could be poorly synchronized. There could be some other vibration in the system which could cause the X-ray flux to vary quite a bit. It could be as simple as the cycling of the cooling for the monochromator, the room or hutch, or the X-ray source itself. All of these would affect non-thin-images as wells. As already mentioned, combining thin-images into wider images will not overcome these problems. If Rmerge or Rmeas was 40% or more in the low resolution shells, then the diffraction pattern is probably mis-indexed, but that kind of high value was not reported here. OTOH, if The diffraction is quite weak, one may be limited by counting statistics. This also cannot be overcome by processing. Jim -Original Message- From: CCP4 bulletin board [mailto:ccp...@jiscmail.ac.uk] On Behalf Of Sergei Strelkov Sent: Friday, November 05, 2010 3:41 AM To: CCP4BB@JISCMAIL.AC.UK Subject: [ccp4bb] High Rmerge with thin frames Dear All, I am processing a dataset collected (not by me) with 0.1 degree oscillations. The diffraction is quite weak even though there is a clean diffraction pattern to about 3A. ...
Re: [ccp4bb] High Rmerge with thin frames
three additional points: 1. OTOH, if The diffraction is quite weak, one may be limited by counting statistics. This also cannot be overcome by processing. As JIm suggests above then, maybe you should look if the 15% Rmerge is almost reasonable given the specific I/sigI at low resolution? 2. If there is one thing I do not like in XDS, is that there is no (or I have failed to find) statistics of I/sigI and Rmerge as function of image. Have a look at the SCALA output. Maybe some images are bad? 3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. A.
Re: [ccp4bb] High Rmerge with thin frames
3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. Perhaps the reason for the better data was an instance of the redundancy-vs-long-exposure dilemma. Given, say, 1deg exposures: should one collect 500x1s exposures or 250x2s exposures? I think this has been examined, and while it does depend on the details of the parameters, it is often better to collect 250x2s exposures because there is a flat rate per frame noise level in the detector. I am wondering whether you just increased the number of flat rates in your data sets by increasing the number of frames while keeping the exposure/degree equal? JPK
Re: [ccp4bb] High Rmerge with thin frames
On Nov 5, 2010, at 16:57, Ronnie Berntsson wrote: Dear Tassos, I'm interested in your third point. Do you have any explanation for why 0.5-1 degrees oscillation gave better data? Purely due to the fact that the crystals survived longer and thus yielded higher redundancy data, or also other parameters? No, x-ray beam survival had nothing to do with this. The dose was the same per degree, so the damage was the same in all, at least on principle, and from what I see in practice. I simply think that the very low partiality of all reflections ends up with counts that are just above the noise (even for a well set up experiment and an excellent detector). The diffracting volume of these crystals in some orientation was really small! 1x40x70 microns, only, so the signals were very low. I think that the end integration goes wrong because images cannot be well refined with so low-count data. As soon as the reflections were stronger, things 'catch up'. Also do anyone know where the threshold lies for when not to use fine phi slicing on the PILATUS? ie, at what level of diffraction would one need to increase the exposure (and oscillation in order to still get redundant data)? In general, the slicing of the Pilatus works great for us. My only negative experience was really really small crystals. We'll be in a similar position in the coming weeks with data collection using PILATUS detectors, and would like to maximize the potential data quality from our weak diffracting crystals. Any input on this would be greatly appreciated! I would aim to be able to see nice spots, to at least 3.5 A: these would be enough to 'lock' the orientations, and I would expect either XDS or MOSFLM to integrate even very low signals. A. Cheers, Ronnie Berntsson On Nov 5, 2010, at 16:16, Anastassis Perrakis wrote: three additional points: 1. OTOH, if The diffraction is quite weak, one may be limited by counting statistics. This also cannot be overcome by processing. As JIm suggests above then, maybe you should look if the 15% Rmerge is almost reasonable given the specific I/sigI at low resolution? 2. If there is one thing I do not like in XDS, is that there is no (or I have failed to find) statistics of I/sigI and Rmerge as function of image. Have a look at the SCALA output. Maybe some images are bad? 3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. A. P please don't print this e-mail unless you really need to Anastassis (Tassos) Perrakis, Principal Investigator / Staff Member Department of Biochemistry (B8) Netherlands Cancer Institute, Dept. B8, 1066 CX Amsterdam, The Netherlands Tel: +31 20 512 1951 Fax: +31 20 512 1954 Mobile / SMS: +31 6 28 597791
Re: [ccp4bb] High Rmerge with thin frames
In the Pilatus mode these are open-shutter experiment, where the Pilatus integrates over different times - all these exposure times are slower than the frequency of the detector, as far as I understand the setup. So, the crystal gets the 'full blast' in all cases, and the blast is the same for the same rotation, as we did the experiments. btw, I have the suspicion, that for our system, it was better to do '100 sec' in the pilatus, than 100x1 sec and read for 3-4 secs with a CCD in between, but I have no rigorous proof for that. Its just an observation that should be treated with caution. Tassos On Nov 5, 2010, at 17:11, Jacob Keller wrote: 3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. Perhaps the reason for the better data was an instance of the redundancy-vs-long-exposure dilemma. Given, say, 1deg exposures: should one collect 500x1s exposures or 250x2s exposures? I think this has been examined, and while it does depend on the details of the parameters, it is often better to collect 250x2s exposures because there is a flat rate per frame noise level in the detector. I am wondering whether you just increased the number of flat rates in your data sets by increasing the number of frames while keeping the exposure/degree equal? JPK P please don't print this e-mail unless you really need to Anastassis (Tassos) Perrakis, Principal Investigator / Staff Member Department of Biochemistry (B8) Netherlands Cancer Institute, Dept. B8, 1066 CX Amsterdam, The Netherlands Tel: +31 20 512 1951 Fax: +31 20 512 1954 Mobile / SMS: +31 6 28 597791
Re: [ccp4bb] High Rmerge with thin frames
Hi I'd read Jim Pflugrath's 1999 paper in Acta D for a discussion on fine phi slicing - in general, if memory serves me correctly, he suggested using an oscillation angle ~0.5x the mosaic spread. I think there may be issues with collecting data too finely with a Pilatus, even in shutterless mode. I don't know where the problems arise (can't be shutter/rotation axis synchronisation), but it seems to be the normal thing in XDS (which should have no problems with fine phi-slicing) to use the PATCH_SHUTTER_PROBLEM=TRUE that Martin Hallberg suggested, which looks a bit like a fudge to me (but I expect Kay to correct me on that!). The important thing is to treat each data collection as a scientific experiment, and to treat each crystal as an individual. Use the strategy tools (e.g. BEST, DNA, Edna) to work out how to get the best data from each crystal - remember that the data collection itself is the last experimental step. Any of the issues mentioned earlier in this thread will be exacerbated by very short exposure times - so it may be an idea to attenuate the beam and spend five minutes on a data collection rather than one On 5 Nov 2010, at 15:57, Ronnie Berntsson wrote: Dear Tassos, I'm interested in your third point. Do you have any explanation for why 0.5-1 degrees oscillation gave better data? Purely due to the fact that the crystals survived longer and thus yielded higher redundancy data, or also other parameters? Also do anyone know where the threshold lies for when not to use fine phi slicing on the PILATUS? ie, at what level of diffraction would one need to increase the exposure (and oscillation in order to still get redundant data)? We'll be in a similar position in the coming weeks with data collection using PILATUS detectors, and would like to maximize the potential data quality from our weak diffracting crystals. Any input on this would be greatly appreciated! Cheers, Ronnie Berntsson On Nov 5, 2010, at 16:16, Anastassis Perrakis wrote: three additional points: 1. OTOH, if The diffraction is quite weak, one may be limited by counting statistics. This also cannot be overcome by processing. As JIm suggests above then, maybe you should look if the 15% Rmerge is almost reasonable given the specific I/sigI at low resolution? 2. If there is one thing I do not like in XDS, is that there is no (or I have failed to find) statistics of I/sigI and Rmerge as function of image. Have a look at the SCALA output. Maybe some images are bad? 3. making too fine slices of too weak diffraction images ends up with either too weak counting statistics or inability to 'lock' the refinement. we did that for one crystal form, collecting 0.1, 0.2, 0.35, 0.5, 0.7, 1.0 from various crystals (with the same dose per degree, at SLS using a PILATUS, mosaicity 0.4-0.6) in an attempt to get better Se signal. We miserably failed to get any useful signal at the end, but learned that for these very weak diffracting plates (submicron) collecting 0.5-1.0 degrees was actually giving at the end better data. A. Harry -- Dr Harry Powell, MRC Laboratory of Molecular Biology, MRC Centre, Hills Road, Cambridge, CB2 0QH
Re: [ccp4bb] High Rmerge with thin frames
Hello Tassos, the data you are missing are available from XDS_ASCII.HKL and can e.g. be generated with Kay Diederichs xdsstat, see http://strucbio.biologie.uni-konstanz.de/xdswiki/index.php/XDSSTAT According to the web site, it prints Rmeas instead of Rmerge, but since Rmeas should anyhow be promoted and Rmerge abandoned, that's a good thing. In case you have access to it, sadabs also generate some useful graphs from XDS_ASCII.HKL Cheers, Tim On Fri, Nov 05, 2010 at 04:16:02PM +0100, Anastassis Perrakis wrote: 2. If there is one thing I do not like in XDS, is that there is no (or I have failed to find) statistics of I/sigI and Rmerge as function of image. Have a look at the SCALA output. Maybe some images are bad? A. -- -- 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] High Rmerge with thin frames
Hi Harry, On Nov 5, 2010, at 5:45 PM, Harry Powell wrote: I think there may be issues with collecting data too finely with a Pilatus, even in shutterless mode. I don't know where the problems arise (can't be shutter/rotation axis synchronisation), but it seems to be the normal thing in XDS (which should have no problems with fine phi-slicing) to use the PATCH_SHUTTER_PROBLEM=TRUE that Martin Hallberg suggested, which looks a bit like a fudge to me (but I expect Kay to correct me on that!). It is not the normal thing in XDS but it is perhaps a relatively common solution to shutter/spindle synchronisation problems discovered afterwards (always process directly at the beam line!). The default in XDS is indeed PATCH_SHUTTER_PROBLEM=FALSE Compensating like this is of course not the best (go and recollect!) but still way better than unusable data in the meantime. In the case Sergei originally described it would at least indicate what the problem may be. Sergei did not say which detector was used for the data collection so we don't know if it was a Pilatus or a CCD. Maybe Sergei can fill us in on the details? BTW, any views in the community on crystal lifetime with continuous data collection like on the Pilatus (or AXIOM) versus letting the crystal rest/cool/die(?) a second between frames on a CCD? Cheers, Martin
Re: [ccp4bb] High Rmerge with thin frames
Compensating like this is of course not the best (go and recollect!) but still way better than unusable data in the meantime. In the case Sergei originally described it would at least indicate what the problem may be. Sergei did not say which detector was used for the data collection so we don't know if it was a Pilatus or a CCD. Maybe Sergei can fill us in on the details? The data that Sergei described were collected by yours truly at beamline ID14-2 of the ESRF, so detector is a Quantum CCD. I collected data for another protein crystal a few hours before with 0.5 degree oscillations - no abnormal statistics here, so I don't suspect cryo or loop instabilities are the underlying cause as suggested before. The choice of 0.1 degree oscillations appears somewhat unfortunate in retrospect, but this is what several programs (Best/mosflm/DNA) suggested to reduce overlaps. The data were collected for the only crystal that showed decent diffraction to ~3A out of more than 30-40 screened. -Chris
