Re: [ccp4bb] ice rings
Dear Chen, you should only exclude resolution ranges when you actually have ice-ring contamination in your data: did you decide on that by hand or is it based in your case on some automatic analysis? If you have this situation then the completeness will indeed go down - after all, there are possible reflections that you didn't observe (since you excluded the resolution ranges these reflections they occur in). If you are worried about this, you could e.g. try the automatic detection and treatment of ice-rings during processing in e.g. autoPROC (see [1] and [2]): it will only exclude ice-ring resolution ranges that are detected and it will use an excluded resolution range as narrow as possible/necessary to avoid rejecting otherwise good data when processing data in XDS. Cheers Clemens [1] www.globalphasing.com/autoproc/ [2] www.globalphasing.com/autoproc/manual/autoPROC7.html#step1_spotnohkl On Fri, Jul 20, 2018 at 10:49:48AM -0400, CPMAS Chen wrote: > Hi, All CCP4 users, > > This might be a little off-topic, but I cannot find a mail list for XDS. > During XDS processing, we can exclude the resolution range to "remove" ice > ring. > In my case, when I excluded these range, the completeness at this range > will be much lower, ~45%. > N 1/d^2DmidNmeas NrefNcent %poss C%poss Mlplct AnoCmp > AnoFrc AnoMlt $$ $$ > 16 0.0582 4.1525565 4013 54 95.2 98.96.4 > 94.6 99.23.2 > 17 0.0619 4.0226217 4167 54 95.6 98.66.3 > 94.7 99.03.2 > 18 0.0657 3.9011393 1973 23 44.4 94.25.8 > 41.2 92.53.1 > 19 0.0694 3.7931643 4556 60 99.0 94.66.9 > 98.7 99.73.5 > 20 0.0732 3.7014014 2149 30 45.4 90.96.5 > 43.7 96.13.4 > 21 0.0769 3.6129456 4242 57 87.5 90.76.9 > 86.1 98.43.5 > 22 0.0807 3.5235722 4940 61 99.3 91.37.2 > 99.2 99.83.6 > > If I included these resolution range, the completeness is more than 90%. > However, when look at the Mn(I/sd), it does not follow Willson law at high > resolution. > Is there a compromise to just exclude partial of these region? > > Thanks! > N 1/d^2Dmid Rmrg Rfull Rcum Rmeas RpimNmeas AvI > RMSdevsd I/RMS Mn(I/sd) FrcBias Chi^2 Chi^2c $$ $$ > 1 0.0017 24.37 0.037 0.037 0.037 0.043 0.022 4236 1235 > 112 65 11.0 42.7 - 1.30 0.97 > 2 0.0050 14.07 0.041 0.041 0.038 0.047 0.024 8627 389 > 32 24 12.1 37.3 - 0.99 0.98 > 3 0.0084 10.90 0.040 0.040 0.039 0.046 0.02311346 438 > 37 26 11.7 37.3 - 0.97 0.96 > 4 0.0118 9.21 0.044 0.044 0.040 0.051 0.02613482 275 > 24 18 11.3 33.1 - 0.95 0.94 > 5 0.0151 8.12 0.051 0.051 0.041 0.061 0.03411303 148 > 15 12 10.1 23.1 - 0.93 0.88 > 6 0.0185 7.35 0.070 0.070 0.043 0.084 0.04514124 84 > 10 98.3 18.7 - 0.93 0.93 > 7 0.0219 6.76 0.086 0.086 0.045 0.102 0.05415700 65 > 9 87.1 15.9 - 0.98 0.97 > 8 0.0252 6.29 0.100 0.100 0.047 0.118 0.06217497 54 > 8 86.4 14.3 - 0.96 0.96 > 9 0.0286 5.91 0.106 0.106 0.050 0.124 0.06518961 54 > 9 86.1 14.0 - 0.95 0.94 > 10 0.0320 5.59 0.113 0.113 0.053 0.133 0.06820406 52 > 9 95.8 13.5 - 0.95 0.95 > 11 0.0353 5.32 0.119 0.119 0.056 0.138 0.07121836 53 > 10 95.5 13.2 - 0.98 0.98 > 12 0.0387 5.08 0.111 0.111 0.059 0.129 0.06722914 64 > 11 105.9 14.2 - 1.01 1.00 > 13 0.0421 4.87 0.117 0.117 0.062 0.137 0.07023980 63 > 11 105.6 13.6 - 1.02 1.01 > 14 0.0454 4.69 0.127 0.127 0.065 0.149 0.07624916 61 > 12 115.2 12.8 - 1.02 1.01 > 15 0.0488 4.53 0.162 0.162 0.069 0.189 0.09726030 47 > 11 114.1 10.5 - 1.01 1.01 > 16 0.0522 4.38 0.202 0.202 0.073 0.235 0.12026790 38 > 12 113.3 8.9 - 0.99 0.99 > 17 0.0555 4.24 0.261 0.261 0.078 0.306 0.15826547 30 > 12 112.6 7.1 - 1.01 1.01 > 18 0.0589 4.12 0.316 0.316 0.081 0.387 0.22021957 24 > 12 112.0 5.2 - 1.00 0.96 > 19 0.0623 4.01 0.498 0.498 0.086 0.602 0.33324279 16 > 13 121.3 3.7 - 1.01 0.98 > 20 0.0656 3.90 0.588 0.588 0.094 0.718 0.40425888 21 > 33 130.6 3.9 - 3.12 1.24 > 21 0.0690 3.81 0.822 0.822 0.101 0.974 0.51528128 12 > 15 140.8
[ccp4bb] ice rings
Hi, All CCP4 users, This might be a little off-topic, but I cannot find a mail list for XDS. During XDS processing, we can exclude the resolution range to "remove" ice ring. In my case, when I excluded these range, the completeness at this range will be much lower, ~45%. N 1/d^2DmidNmeas NrefNcent %poss C%poss Mlplct AnoCmp AnoFrc AnoMlt $$ $$ 16 0.0582 4.1525565 4013 54 95.2 98.96.4 94.6 99.23.2 17 0.0619 4.0226217 4167 54 95.6 98.66.3 94.7 99.03.2 18 0.0657 3.9011393 1973 23 44.4 94.25.8 41.2 92.53.1 19 0.0694 3.7931643 4556 60 99.0 94.66.9 98.7 99.73.5 20 0.0732 3.7014014 2149 30 45.4 90.96.5 43.7 96.13.4 21 0.0769 3.6129456 4242 57 87.5 90.76.9 86.1 98.43.5 22 0.0807 3.5235722 4940 61 99.3 91.37.2 99.2 99.83.6 If I included these resolution range, the completeness is more than 90%. However, when look at the Mn(I/sd), it does not follow Willson law at high resolution. Is there a compromise to just exclude partial of these region? Thanks! N 1/d^2Dmid Rmrg Rfull Rcum Rmeas RpimNmeas AvI RMSdevsd I/RMS Mn(I/sd) FrcBias Chi^2 Chi^2c $$ $$ 1 0.0017 24.37 0.037 0.037 0.037 0.043 0.022 4236 1235 112 65 11.0 42.7 - 1.30 0.97 2 0.0050 14.07 0.041 0.041 0.038 0.047 0.024 8627 389 32 24 12.1 37.3 - 0.99 0.98 3 0.0084 10.90 0.040 0.040 0.039 0.046 0.02311346 438 37 26 11.7 37.3 - 0.97 0.96 4 0.0118 9.21 0.044 0.044 0.040 0.051 0.02613482 275 24 18 11.3 33.1 - 0.95 0.94 5 0.0151 8.12 0.051 0.051 0.041 0.061 0.03411303 148 15 12 10.1 23.1 - 0.93 0.88 6 0.0185 7.35 0.070 0.070 0.043 0.084 0.04514124 84 10 98.3 18.7 - 0.93 0.93 7 0.0219 6.76 0.086 0.086 0.045 0.102 0.05415700 65 9 87.1 15.9 - 0.98 0.97 8 0.0252 6.29 0.100 0.100 0.047 0.118 0.06217497 54 8 86.4 14.3 - 0.96 0.96 9 0.0286 5.91 0.106 0.106 0.050 0.124 0.06518961 54 9 86.1 14.0 - 0.95 0.94 10 0.0320 5.59 0.113 0.113 0.053 0.133 0.06820406 52 9 95.8 13.5 - 0.95 0.95 11 0.0353 5.32 0.119 0.119 0.056 0.138 0.07121836 53 10 95.5 13.2 - 0.98 0.98 12 0.0387 5.08 0.111 0.111 0.059 0.129 0.06722914 64 11 105.9 14.2 - 1.01 1.00 13 0.0421 4.87 0.117 0.117 0.062 0.137 0.07023980 63 11 105.6 13.6 - 1.02 1.01 14 0.0454 4.69 0.127 0.127 0.065 0.149 0.07624916 61 12 115.2 12.8 - 1.02 1.01 15 0.0488 4.53 0.162 0.162 0.069 0.189 0.09726030 47 11 114.1 10.5 - 1.01 1.01 16 0.0522 4.38 0.202 0.202 0.073 0.235 0.12026790 38 12 113.3 8.9 - 0.99 0.99 17 0.0555 4.24 0.261 0.261 0.078 0.306 0.15826547 30 12 112.6 7.1 - 1.01 1.01 18 0.0589 4.12 0.316 0.316 0.081 0.387 0.22021957 24 12 112.0 5.2 - 1.00 0.96 19 0.0623 4.01 0.498 0.498 0.086 0.602 0.33324279 16 13 121.3 3.7 - 1.01 0.98 20 0.0656 3.90 0.588 0.588 0.094 0.718 0.40425888 21 33 130.6 3.9 - 3.12 1.24 21 0.0690 3.81 0.822 0.822 0.101 0.974 0.51528128 12 15 140.8 2.7 - 1.03 1.00 22 0.0724 3.72 1.140 1.140 0.109 1.347 0.709295839 18 150.5 2.0 - 1.25 1.01 23 0.0757 3.63 0.819 0.819 0.118 0.989 0.54330705 18 58 160.3 2.2 - 1.81 1.09 24 0.0791 3.56 1.643 1.643 0.128 1.926 0.995316267 18 160.4 1.5 - 1.03 1.00 25 0.0825 3.48 2.505 2.505 0.139 2.946 1.534323275 23 170.2 1.0 - 1.33 1.06 26 0.0858 3.41 1.601 1.601 0.151 1.925 1.05132349 10 52 180.2 1.4 - 2.72 1.17 -- *** Charles Chen Research Instructor University of Pittsburgh School of Medicine Department of Anesthesiology ** To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB=1
[ccp4bb] ice rings
Hi, All CCP4 users -- *** Charles Chen Research Instructor University of Pittsburgh School of Medicine Department of Anesthesiology ** To unsubscribe from the CCP4BB list, click the following link: https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=CCP4BB=1
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On 10/11/2011 12:33 PM, Garib N Murshudov wrote: We need better way of estimating unobserved reflections. Indeed we do! Because this appears to be the sum total of how the correctness of the structure is judged. It is easy to forget I think that from the point of view of the refinement program, all reflections flagged as belonging to the free set are, in effect, missing. So Rfree is really just a score for how well DFc agrees with Fobs? -James Holton MAD Scientist
Re: [ccp4bb] Ice rings...
Automated outlier rejection in scaling will handle a lot of things, including ice. Works better with high multiplicity. Unless, of course, your ice rings are even, then any integration error due to ice will be the same for all the symmetry mates and the scaling program will be none the wiser. That said, the integration programs these days tend to have pretty sensible defaults for rejecting spots that have weird backgrounds. Plenty of structures get solved from data that has horrible-looking ice rings using just the defaults. In fact, I am personally unconvinced that ice rings are a significant problem in and of themselves. More often, they are simply an indication that something else is wrong, like the crystal warmed up at some point. Nevertheless, if you suspect your ice rings are causing a problem, you can try to do something about them. The deice program already mentioned sounds cool, but if you just want to try something quick, excluding the resolution ranges of your ice rings can be done in sftools like this: select resol 3.89 select resol 3.93 absent col F SIGF DANO SIGDANO if col F 0 and repeat this for each resolution range you want to exclude. Best to get these ranges from your integration program's graphics display. In mosflm, you can put EXCLUDE ICE on either the AUTOINDEX or RESOLUTION keywords and have any spots on the canonical hexagonal ice spacings removed automatically. The problem with excluding resolution ranges, of course, is that your particular ice rings may not be where they are supposed to be. Either due to something physical, like the cooling rate, or something artificial, like an error in the camera parameters. It is also possible that what you think are ice rings are actually salt rings. Some salts will precipitate out upon cryo-cooling. Large ice/salt crystals can also produce a lot of non-Bragg scatter, which means that you can get sharp features far away from the resolution range you expect. On the other hand, if you have cubic ice instead of hexagonal ice (very common in MX samples), then there are no rings at 3.91A, 3.45A, 2.68A and throwing out these resolution ranges would be a waste. Another way to exclude ice is to crank up background-based rejection criteria. In denzo/HKL2K, you do this with the reject fraction keyword, and in mosflm, REJECT MINBG does pretty much the same thing. There are lots of rejection options in integration programs, and which one works in your particular case depends on what your ice rings look like. Noone has written a machine-vision type program that can recognize and handle all the cases. You will need to play with these options until the spots you don't like turn red in the display. Of course, the best way to deal with ice rings would be to inspect each and every one of the spots you have near ice rings and decide on its intensity manually. Then edit the hkl file. Which brings me to perhaps a more important point: What, exactly, is the problem you are having that makes you think the ice rings are to blame? Can't get an MR solution? Can't get MAD/SAD phases? Ice has a bad rep in MX, and an undeserved one IMHO. In fact, by controlling either cryoprotectant concentration or cooling rate carefully, you can achieve a mixture of amorphous and cubic ice, and this mixture has a specific volume (density) intermediate between the two. Many crystals diffract much better when you are able to match the specific volume of the stuff in the solvent channels to the specific volume protein lattice is trying to achieve on its own. A great deal of effort has gone into characterizing this phenomenon (authors: Juers, Weik, Warkentin, Thorne and many others), but I often meet frustrated cryo-screeners who seem to have never heard of any of it! In general, the automated outlier rejection protocols employed by modern software have taken care of most of the problems ice rings introduce. For example, difference Pattersons are VERY sensitive to outliers, and all it takes is one bad spot to give you huge ripples that swamp all you peaks, but every heavy-atom finding program I am aware of calculates Pattersons only after fist doing an outlier rejection step. You might also think that ice rings would mess up your preciously subtle anomalous differences, but again, outlier rejection to the rescue. Now, that said, depending on automated outlier rejection to save you is of course a questionable policy, but it is an equally bad idea to pretend that it doesn't exist either. It is funny how in MX we are all ready to grab our torch and pitchfork if we hear of someone manually editing their hkl files to get rid of reflections they don't like, but as long as the software does it, it is okay. Plausible deniability runs deep. -James Holton MAD Scientist On 10/11/2011 8:16 AM, Francis E Reyes wrote: All, So I have two intense ice rings where there appear to be
Re: [ccp4bb] Ice rings... [maps and missing reflections]
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 I am glad the structures that have been solved using the free-lunch-algorithm as implemented in shelxe did not know they were not allowed to be solved. Of course there is DM involved, as has been pointed out ;-) On 10/12/2011 10:12 PM, Edward A. Berry wrote: Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/11/2011 09:58 PM, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Some people call this the free-lunch-algorithm ;-) Tim Doesn't work- the Fourier transform is invertable. As someone already said in this thread, if the map was made with coefficients of zero for certain reflections (which is equivalent to omitting those reflections) The back-transform will give zero for those reflections. Unless you do some density modification first. So free-lunch is a good name- there aint no such thing! - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.10 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFOlqZiUxlJ7aRr7hoRAgYqAKD1vthQQ3WJmHXxklWZiroRYvdFHgCeO0MP FSF50BnydKjR7ajI3XshBqE= =F0JM -END PGP SIGNATURE-
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Here we are I presume only worried about strong reflections lost behind an ice ring. At least that is where the discussion began. Isnt the best approach t this problem to use integration software which attempts to give a measurement, albeit with a high error estimate? The discussion has strayed into what to do with incomplete data sets.. In these cases there might be something to learn from the Free Lunch ideas used in ACORN and SHELX and other programs - set the missing reflections to E=1, and normalise them properly to an appropriate amplitude. Eleanor On 10/11/2011 08:33 PM, Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. It may give better refinement behaviour and less biased map. Another one is integration over all errors (too many parameters for numerical integration, and there is no closed form formula) of model as well as experimental data. This would give less bia sed map with more pronounced signal. Regards Garib On 11 Oct 2011, at 20:15, Randy Read wrote: If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averagedFobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- After much deep and profound brain things inside my head, I have decided to thank you for bringing peace to our home. Julian, King of Lemurs -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] Ice rings... [maps and missing reflections]
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/11/2011 09:58 PM, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Some people call this the free-lunch-algorithm ;-) Tim Ethan [...] - -- - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.10 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iD8DBQFOlVi4UxlJ7aRr7hoRAlU+AKDo+c449pUQ/1cnQAl6SMRqzVkp6wCcDETj GHB8hFXt1McbxWHfpUAsHtE= =FOWk -END PGP SIGNATURE-
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/11/2011 09:58 PM, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Some people call this the free-lunch-algorithm ;-) Tim Doesn't work- the Fourier transform is invertable. As someone already said in this thread, if the map was made with coefficients of zero for certain reflections (which is equivalent to omitting those reflections) The back-transform will give zero for those reflections. Unless you do some density modification first. So free-lunch is a good name- there aint no such thing!
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Wednesday, October 12, 2011 01:12:11 pm Edward A. Berry wrote: Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/11/2011 09:58 PM, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Some people call this the free-lunch-algorithm ;-) Tim Doesn't work- the Fourier transform is invertable. As someone already said in this thread, if the map was made with coefficients of zero for certain reflections (which is equivalent to omitting those reflections) The back-transform will give zero for those reflections. Unless you do some density modification first. So free-lunch is a good name- there aint no such thing! Tim refers to the procedure described in Sheldrick, G. M. (2002). Z. Kristallogr. 217, 644–65 which was later incorporated into shelxe as the Free Lunch Algorithm. It does indeed involve a form of density modification. Tim is also correct that this procedure is the precedent I had in mind, although I had forgotten its clever name. cheers, Ethan -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Dear Ethan, Thankyou for the reference, but actually it's the wrong paper and anyway my only contribution to the 'free lunch algorithm' was to name it (in the title of the paper by Uson et al., Acta Cryst. (2007) D63, 1069-1074). By that time the method was already being used in ACORN and by the Bari group, who were the first to describe it in print (Caliandro et al., Acta Cryst. Acta Cryst. (2005) D61, 556-565). As you correctly say, it only makes sense in the context of density modification, but under favorable conditions, i.e. native data to 2A or better, inventing data to a resolution that you would have liked to collect but didn't can make a dramatic improvement to a map, as SHELXE has often demonstrated. Hence the name. And of course there is no such thing as a free lunch! Best regards, George On Wed, Oct 12, 2011 at 01:25:12PM -0700, Ethan Merritt wrote: On Wednesday, October 12, 2011 01:12:11 pm Edward A. Berry wrote: Tim Gruene wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 On 10/11/2011 09:58 PM, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Some people call this the free-lunch-algorithm ;-) Tim Doesn't work- the Fourier transform is invertable. As someone already said in this thread, if the map was made with coefficients of zero for certain reflections (which is equivalent to omitting those reflections) The back-transform will give zero for those reflections. Unless you do some density modification first. So free-lunch is a good name- there aint no such thing! Tim refers to the procedure described in Sheldrick, G. M. (2002). Z. Kristallogr. 217, 644–65 which was later incorporated into shelxe as the Free Lunch Algorithm. It does indeed involve a form of density modification. Tim is also correct that this procedure is the precedent I had in mind, although I had forgotten its clever name. cheers, Ethan -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742 -- Prof. George M. Sheldrick FRS Dept. Structural Chemistry, University of Goettingen, Tammannstr. 4, D37077 Goettingen, Germany Tel. +49-551-39-3021 or -3068 Fax. +49-551-39-22582
[ccp4bb] Ice rings...
All, So I have two intense ice rings where there appear to be lattice spots in between them. I understand that any reflections that lie directly on the ice ring are useless, however, how do software programs (HKL2000, d*Trek, mosflm, XDS) deal with these intermediate spots? It would seem to me that employing a 'resolution cut off' just before the ice ring (on the low resolution side) would be improper, as there are spots on the high resolution side of the ice. (see enclosed .tiff) In fact, how do these programs deal with spots lying on ice rings? Are they rejected by some algorithm by those programs during integration, or is it up to the scaling/merging (by SCALA for example) step to deal with them? Thanks! F inline: PastedGraphic-1.tiff - Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder
Re: [ccp4bb] Ice rings...
Dear Francis, the spots will be excluded individually based on the inhomogeneous background, so you don't need to apply a resolution cutoff. However, once you have determined and refined your structure it may be worth predicting the intensity of these spots and put them back for map calculation, this might avoid gaps in your map corresponding to inter-atom distances for which data are missing in the resolution range of the ice rings; as long as this is done only for a relatively small set of reflections there is not much risk of introducing a bias here. HTH, Bruno -Message d'origine- De : CCP4 bulletin board [mailto:CCP4BB@JISCMAIL.AC.UK] De la part de Francis E Reyes Envoyé : Tuesday, October 11, 2011 5:17 PM À : CCP4BB@JISCMAIL.AC.UK Objet : [ccp4bb] Ice rings... All, So I have two intense ice rings where there appear to be lattice spots in between them. I understand that any reflections that lie directly on the ice ring are useless, however, how do software programs (HKL2000, d*Trek, mosflm, XDS) deal with these intermediate spots? It would seem to me that employing a 'resolution cut off' just before the ice ring (on the low resolution side) would be improper, as there are spots on the high resolution side of the ice. (see enclosed .tiff) In fact, how do these programs deal with spots lying on ice rings? Are they rejected by some algorithm by those programs during integration, or is it up to the scaling/merging (by SCALA for example) step to deal with them? Thanks! F ### Dr. Bruno P. Klaholz Department of Integrated Structural Biology Institute of Genetics and of Molecular and Cellular Biology IGBMC - UMR 7104 - U 964 1, rue Laurent Fries BP 10142 67404 ILLKIRCH CEDEX FRANCE Tel. from abroad: 0033.388.65.57.55 Tel. inside France: 03.88.65.57.55 Fax from abroad: 0033.388.65.32.76 Fax inside France: 03.88.65.32.76 e-mail: klah...@igbmc.fr websites: http://www.igbmc.fr/ http://igbmc.fr/Klaholz
Re: [ccp4bb] Ice rings...
If the ice rings are really sharp, they trigger the bad background rejection in denzo/HKL2000. To reject more spots, increase the reject fraction 0.7 parameter to something greater than .7. This rejection is on a spot by spot basis, so spots with good background between the rings should not be affected. During integration, if you are monitoring the process with Xdisp, you will see the rejected spots turn red and/or disappear. To verify they are being rejected by background fraction, try again with reject fraction .3 and see if they stay green/yellow. If the ice ring is broad compared to the integrating box, it shows up as a high, slanting baseline and the normal baseline correction procedure is valid, but sigma will be higher than for a spot on a white background. Francis E Reyes wrote: All, So I have two intense ice rings where there appear to be lattice spots in between them. I understand that any reflections that lie directly on the ice ring are useless, however, how do software programs (HKL2000, d*Trek, mosflm, XDS) deal with these intermediate spots? It would seem to me that employing a 'resolution cut off' just before the ice ring (on the low resolution side) would be improper, as there are spots on the high resolution side of the ice. (see enclosed .tiff) In fact, how do these programs deal with spots lying on ice rings? Are they rejected by some algorithm by those programs during integration, or is it up to the scaling/merging (by SCALA for example) step to deal with them? Thanks! F - Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder
Re: [ccp4bb] Ice rings...
Francis, I would like to bring your attention to our paper in Acta Cryst D Volume 66 (6), 741-744 (2010) where we deal with spots under the ice-rings. We have been very successful in eliminating the ice-rings and recover the data underneath. If you are interested you can request the Python script from Michael Chapman at OHSU. De-icing: recovery of diffraction intensities in the presence of ice rings, Michael S. Chapman and^^Thayumanasamy Somasundaram If you need help please e-mail me outside the CCP4BB. ** On 10/11/2011 11:16 AM, Francis E Reyes wrote: All, So I have two intense ice rings where there appear to be lattice spots in between them. I understand that any reflections that lie directly on the ice ring are useless, however, how do software programs (HKL2000, d*Trek, mosflm, XDS) deal with these intermediate spots? It would seem to me that employing a 'resolution cut off' just before the ice ring (on the low resolution side) would be improper, as there are spots on the high resolution side of the ice. (see enclosed .tiff) In fact, how do these programs deal with spots lying on ice rings? Are they rejected by some algorithm by those programs during integration, or is it up to the scaling/merging (by SCALA for example) step to deal with them? Thanks! F - Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder -- Dr. Thayumanasamy Somasundaram [Soma] Director, X-Ray Crystallography Facility (XRF) Off. Ph: (850)644-6448| Lab Ph: (850)645-1333 Fax:(850)644-7244 | E-mail: tsomasunda...@fsu.edu URI: www.sb.fsu.edu/~soma | URI: www.sb.fsu.edu/~xray Postal Address-- 91, Chieftan Way | KLB 414 Institute of Molecular Biophysics Florida State University Tallahassee, FL 32306-4380, USA.
Re: [ccp4bb] Ice rings...
I've used a technique called annealing, which amounts to holding an index card between the cryo stream and the crystal for a few seconds then removing the card quickly. In my experience, about 70% of the time the diffraction is worse and about 30% of the time the ice rings will be gone with slightly improved diffraction, allowing recovery of a significant range of data. Most of the time, though, I find another crystal that had a better initial freeze, so annealing has never been a life saver--but it could be under dire circumstances. James On Oct 11, 2011, at 9:30 AM, Dr. Thayumanasamy Somasundaram wrote: Francis, I would like to bring your attention to our paper in Acta Cryst D Volume 66 (6), 741-744 (2010) where we deal with spots under the ice-rings. We have been very successful in eliminating the ice-rings and recover the data underneath. If you are interested you can request the Python script from Michael Chapman at OHSU. De-icing: recovery of diffraction intensities in the presence of ice rings, Michael S. Chapman and Thayumanasamy Somasundaram If you need help please e-mail me outside the CCP4BB. On 10/11/2011 11:16 AM, Francis E Reyes wrote: All, So I have two intense ice rings where there appear to be lattice spots in between them. I understand that any reflections that lie directly on the ice ring are useless, however, how do software programs (HKL2000, d*Trek, mosflm, XDS) deal with these intermediate spots? It would seem to me that employing a 'resolution cut off' just before the ice ring (on the low resolution side) would be improper, as there are spots on the high resolution side of the ice. (see enclosed .tiff) In fact, how do these programs deal with spots lying on ice rings? Are they rejected by some algorithm by those programs during integration, or is it up to the scaling/merging (by SCALA for example) step to deal with them? Thanks! F - Francis E. Reyes M.Sc. 215 UCB University of Colorado at Boulder -- Dr. Thayumanasamy Somasundaram [Soma] Director, X-Ray Crystallography Facility (XRF) Off. Ph: (850)644-6448 | Lab Ph: (850)645-1333 Fax:(850)644-7244 | E-mail: tsomasunda...@fsu.edu URI: www.sb.fsu.edu/~soma | URI: www.sb.fsu.edu/~xray Postal Address-- 91, Chieftan Way | KLB 414 Institute of Molecular Biophysics Florida State University Tallahassee, FL 32306-4380, USA.
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tue, 2011-10-11 at 15:24 +, Bruno KLAHOLZ wrote: However, once you have determined and refined your structure it may be worth predicting the intensity of these spots and put them back for map calculation, REFMAC does this by default, because expected value of unknown structure factors for missing reflections are better approximated using DFc than with 0 values. CNS defaults to excluding them. As for phenix, I am not entirely sure - it seems that phenix.refine does too (fill_missing_f_obs= False), but if you use the GUI then the fill in option is turned on. -- Oh, suddenly throwing a giraffe into a volcano to make water is crazy? Julian, King of Lemurs
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tue, Oct 11, 2011 at 10:34 AM, Ed Pozharski epozh...@umaryland.eduwrote: CNS defaults to excluding them. As for phenix, I am not entirely sure - it seems that phenix.refine does too (fill_missing_f_obs= False), but if you use the GUI then the fill in option is turned on. In practice, it will be turned on for command-line phenix.refine too if you don't supply your own custom map definitions - actually it produces both filled and unfilled maps, but the former is what most users will see in Coot. -Nat
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tue, Oct 11, 2011 at 10:34 AM, Ed Pozharski epozh...@umaryland.eduwrote: expected value of unknown structure factors for missing reflections are better approximated using DFc than with 0 values. better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. Clearly there are cases where filling in a few missing reflections significantly improves map interpretability without introducing any bias. As for phenix, I am not entirely sure - it seems that phenix.refine does too (fill_missing_f_obs= False), but if you use the GUI then the fill in option is turned on. phenix.refine always outputs two 2mFo-DFc maps: one is computed using the original set of Fobs, and the other one is computed using set of Fobs where missing reflections filled in with DFc calculated using well determined atoms only. By default, Coot will open the filled one. Pavel
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- After much deep and profound brain things inside my head, I have decided to thank you for bringing peace to our home. Julian, King of Lemurs
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Hi Ed, On Tue, Oct 11, 2011 at 11:47 AM, Ed Pozharski epozh...@umaryland.eduwrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Yes, that's all true about what is DFc. In terms of missing-Fobs-filling it's not too important (as map appearance concerned) which values you take, DFc, Fobs , etc. I spent a few days playing with this some years ago. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. Yep, that was the point - sometimes it is good to do, and sometimes it is not, and ... On a practical side, one can always compare the maps with and without missing reflections. ... this is why phenix.refine outputs both maps -:) All the best, Pavel
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tue, 2011-10-11 at 11:54 -0700, Pavel Afonine wrote: Yep, that was the point - sometimes it is good to do, and sometimes it is not, and Do you have a real life example of Fobs=0 being better? You make it sound as if it's 50/50 situation. -- Hurry up before we all come back to our senses! Julian, King of Lemurs
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Do you have a real life example of Fobs=0 being better? Hopefully, there will be a paper some time soon discussing all this - we work on this right now. You make it sound as if it's 50/50 situation. No (sorry if what I wrote sounded that misleading). Pavel
Re: [ccp4bb] Ice rings... [maps and missing reflections]
If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- After much deep and profound brain things inside my head, I have decided to thank you for bringing peace to our home. Julian, King of Lemurs -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk
Re: [ccp4bb] Ice rings... [maps and missing reflections]
In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. It may give better refinement behaviour and less biased map. Another one is integration over all errors (too many parameters for numerical integration, and there is no closed form formula) of model as well as experimental data. This would give less biased map with more pronounced signal. Regards Garib On 11 Oct 2011, at 20:15, Randy Read wrote: If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- After much deep and profound brain things inside my head, I have decided to thank you for bringing peace to our home. Julian, King of Lemurs -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Ethan It may give better refinement behaviour and less biased map. Another one is integration over all errors (too many parameters for numerical integration, and there is no closed form formula) of model as well as experimental data. This would give less biased map with more pronounced signal. Regards Garib On 11 Oct 2011, at 20:15, Randy Read wrote: If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On 10/11/11 12:58, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Ethan Unless you do some density modification you'll just get back zeros for the reflections you didn't enter. Dale It may give better refinement behaviour and less biased map. Another one is integration over all errors (too many parameters for numerical integration, and there is no closed form formula) of model as well as experimental data. This would give less biased map with more pronounced signal. Regards Garib On 11 Oct 2011, at 20:15, Randy Read wrote: If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] Ice rings... [maps and missing reflections]
Best way would be to generate from probability distributions derived after refinement, but it has a problem that you need to integrate over all errors. Another, simpler way would be generate using Wilson distribution multiple times and do refinement multiple times and average results. I have not done any tests but on paper it looks like a sensible procedure. regards Garib On 11 Oct 2011, at 20:58, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Ethan It may give better refinement behaviour and less biased map. Another one is integration over all errors (too many parameters for numerical integration, and there is no closed form formula) of model as well as experimental data. This would give less biased map with more pronounced signal. Regards Garib On 11 Oct 2011, at 20:15, Randy Read wrote: If the model is really bad and sigmaA is estimated properly, then sigmaA will be close to zero so that D (sigmaA times a scale factor) will be close to zero. So in the limit of a completely useless model, the two methods of map calculation converge. Regards, Randy Read On 11 Oct 2011, at 19:47, Ed Pozharski wrote: On Tue, 2011-10-11 at 10:47 -0700, Pavel Afonine wrote: better, but not always. What about say 80% or so complete dataset? Filling in 20% of Fcalc (or DFcalc or bin-averaged Fobs or else - it doesn't matter, since the phase will dominate anyway) will highly bias the map towards the model. DFc, if properly calculated, is the maximum likelihood estimate of the observed amplitude. I'd say that 0 is by far the worst possible estimate, as Fobs are really never exactly zero. Not sure what the situation would be when it's better to use Fo=0, perhaps if the model is grossly incorrect? But in that case the completeness may be the least of my worries. Indeed, phases drive most of the model bias, not amplitudes. If model is good and phases are good then the DFc will be a much better estimate than zero. If model is bad and phases are bad then filling in missing reflections will not increase bias too much. But replacing them with zeros will introduce extra noise. In particular, the ice rings may mess things up and cause ripples. On a practical side, one can always compare the maps with and without missing reflections. -- Randy J. Read Department of Haematology, University of Cambridge Cambridge Institute for Medical Research Tel: + 44 1223 336500 Wellcome Trust/MRC Building Fax: + 44 1223 336827 Hills RoadE-mail: rj...@cam.ac.uk Cambridge CB2 0XY, U.K. www-structmed.cimr.cam.ac.uk Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742 Garib N Murshudov Structural Studies Division MRC Laboratory of Molecular Biology Hills Road Cambridge CB2 0QH UK Email: ga...@mrc-lmb.cam.ac.uk Web http://www.mrc-lmb.cam.ac.uk
Re: [ccp4bb] Ice rings... [maps and missing reflections]
On 10/11/11 12:58, Ethan Merritt wrote: On Tuesday, October 11, 2011 12:33:09 pm Garib N Murshudov wrote: In the limit yes. however limit is when we do not have solution, i.e. when model errors are very large. In the limit map coefficients will be 0 even for 2mFo-DFc maps. In refinement we have some model. At the moment we have choice between 0 and DFc. 0 is not the best estimate as Ed rightly points out. We replace (I am sorry for self promotion, nevertheless: Murshudov et al, 1997) absent reflection with DFc, but it introduces bias. Bias becomes stronger as the number of absent reflections become larger. We need better way of estimating unobserved reflections. In statistics there are few appraoches. None of them is full proof, all of them are computationally expensive. One of the techniques is called multiple imputation. I don't quite follow how one would generate multiple imputations in this case. Would this be equivalent to generating a map from (Nobs - N) refls, then filling in F_estimate for those N refls by back-transforming the map? Sort of like phase extension, except generating new Fs rather than new phases? Ethan Dale Tronrud wrote Unless you do some density modification you'll just get back zeros for the reflections you didn't enter. Sure. And different DM procedures would give you different imputations, or at least that was my vague idea. Garib N Murshudov wrote Best way would be to generate from probability distributions derived after refinement, but it has a problem that you need to integrate over all errors. Another, simpler way would be generate using Wilson distribution multiple times and do refinement multiple times and average results. I have not done any tests but on paper it looks like a sensible procedure. OK. That makes sense. Ethan -- Ethan A Merritt Biomolecular Structure Center, K-428 Health Sciences Bldg University of Washington, Seattle 98195-7742