Re: [ccp4bb] Why MAD didn't work but SAD works well
My default MAD strategy is to do single-image inverse beam with round robin wavelength changes. That is: energy phi peak 0 peak 180 remote 0 remote 180 peak1 peak 181 remote1 etc with one image taken for each line above. I do this until a full sphere is collected for each wavelength (720 images) with an exposure time short enough so that the final dose is less than 5 MGy. On ALS 8.3.1 that's about 1 second/image. The 5 MGy comes from the half-dose of the fastest-decaying SeMet site I have ever seen (Holton, 2007). Once the initial 5 MGy pass is done, then I quadruple the exposure time and move the detector a little closer to the sample for another sphere. Moving the detector is to try and put the spots on fresh pixels and average over the systematic error associated with using exactly the same part of the detector over and over again. This becomes important for Bijvoet ratios less than ~2%. It is also a good idea to always do a full sphere for the same reason: never use the same pixel twice. The quadrupling of the exposure time is mainly for expediency. Given that any rad dam reaction will be essentially exponential, but with an unknown half-dose, the best way to sample the curve is with a geometric series of exposure times. Doubling the exposure time increases signal/noise by not more than 40%, which seems hardly worth it. Quadrupling the exposure doubles the S/N for counting statistics. So: 1s, 4s, 16s, and then the crystal is usually pretty dead (at ~0.5 MGy/min). This then gives the user the opportunity to do RIP using the long exposures as the native. Or, if there is little damage, they can just merge everything together and get the best signal. The influence of read-out noise (if any) also gets effectively washed out in the longer exposures. Now, what I call peak is actually a compromise between the usual peak and the inflection. What I do is split the difference between these two for an inf-eak or pea-lection wavelength. From a signal-vs-damage point of view this seems to be optimal in my hands. Two wavelengths are about twice as good as one, even if the f and f' to the remote are only 80% of what they would be at their maxima. Three wavelengths are better than two, but only ~20% better. I judge this by looking at map correlations and the number of sites I can leave unmodeled in a 3-wavelength dataset and still get the same map quality as a 2-wavelength dataset. I call such a 2-wavelength dataset DAD, or sometimes Bijvoet Anomalous and Dispersive Anomalous Scattering (BADAS). The only time using the same pixel twice could actually be an advantage is if you could somehow put the same spot on the same pixel at two different wavelengths. You can sort of do this by moving the detector by a distance proportional to the change in wavelength. Doesn't work exactly because the Ewald sphere is curved and the detector isn't, but you can get some spots close. This might be why Gonzalez et al. (2007) noticed that using inflection-and-remote tended to perform better than using just the peak. I haven't done an experiment of my own to show this is due to pixel calibration, however. Of course, for most test crystals it doesn't really matter how you collect the data because the anomalous signal is so strong relative to pixel calibration, or almost any other source of error for that matter. The problem with differentiating the efficacy of one strategy over another is that the transition between solvable and unsolvable is very very sharp. Basically, phase improvement methods either make your phases better or they don't, and then you iterate. But, in a rad dam-limited world (such as a very very small test crystal), the best strategy will prevail. The minimum crystal size you should need if you do everything right is what is reported by this web page: http://bl831.als.lbl.gov/xtalsize.html As for the terms, inverse beam I think came from Stout and Jensen in their description of absorption corrections. It is supposed to be a variation on normal beam (which is where the x-ray beam is perpendicular to the spindle). But like most things, the widespread use of the term arises because a popular piece of software (BLU-ICE) chose to put those words next to a button on the GUI. The term round robin I take from a simple load-distribution technique in computer science where each CPU, network card, etc takes turns getting the next job. This way each of the things being switched up gets the same amount of exposure with minimal granularity. Apparently, this name is derived from competitive sporting events where the athletes do pretty much the same thing. One final word to the wise: My strategy of single-image-round-robin is not appropriate to all beamlines! Some shutters are better than others (even the electronic shutter used for shutterless data collections can experience some jitter),
Re: [ccp4bb] Why MAD didn't work but SAD works well
Yafang, I'm afraid that just because you still have spots at the end of your dataset does not mean radiation damage was not a problem. The reactions that disorder your heavy atom sites go to completion at doses that can be as little as 1/30th of the dose required to noticeably fade your spots. There are a number of nice reviews written about this: http://dx.doi.org/10.1107/S0909049509004361 http://dx.doi.org/10.1107/S0909049512050418 http://dx.doi.org/10.1107/S0909049506048898 http://dx.doi.org/10.1107/S0907444907019580 Also, If your datasets were collected one wavelength at a time, such as a complete dataset at the peak, then another complete dataset at the inflection, and then, after all that, you collect the reference dataset at the remote, then what you have is not a MAD dataset. This is a series of SAD datasets (M-SAD). Of these three SAD datasets only the peak is at the optimum energy for anomalous, and also has the least radiation damage, so that one will work better than the other two. I use the term M-SAD instead of MAD because you are effectively using a different crystal for each wavelength, and that means the inter-wavelength differences are dominated by non-isomorphism. Non-isomorphism can easily bury an anomalous signal, and radiation damage is a pretty efficient way to make a crystal non-isomorphous with its former self. By looking at examples in the literature, (such as Banumathi et al. 2004) one can guestimate that the degree of non-isomorphism induced by radiation damage is about 1% per MGy of dose. You can look up the nominal dose rate of the beamline you collected these data at here: http://bl831.als.lbl.gov/damage_rates.pdf I try to keep the numbers in this document up to date, but most beamlines are attenuated to the point where they deliver about 1 MGy per minute of shutter-open time. That's for a crystal with ~20 mM heavy atoms, and unattenuated beam. So, if the dispersive signal you are looking for is 3%, then once your crystal has endured more than ~3 minutes of shutter-open time, the non-isomorphism will start to overwhelm that signal, and then trying to use dispersive (inter-wavelength) differences becomes counterproductive. This is because the software is trying to reconcile all the observed differences in terms of heavy-atom positions, and when half the differences are coming from non-isomorphism, the equations all fall apart. This is probably why treating your M-SAD dataset as a MAD experiment fails. Anomalous (Bijvoet) differences, however, tend to come up fairly close together in phi because once a spot passes through the Ewald sphere its Friedel mate will generally pop up on the opposite side of the beamstop a few degrees later. Basically, if you're measuring a difference, it is best to measure the two numbers you are going to subtract as close together in time as possible. This is why inverse beam with round robin wavelength changes is the approach that is most robust to damage effects. Yes, you still get damage, but at least the differences you are subtracting are close together, and therefore comparing apples to apples. I suppose it was the advent of saggital-focusing monochromators that made wavelength changes more difficult and more recently the advent of so-called shutterless data collection has led to more and more M-SAD data collections than MAD. This is a pity, really, because as George has already said, MAD gives you significantly better phases than SAD. It just requires a little more patience to collect it properly. -James Holton MAD Scientist On Tue, Aug 20, 2013 at 2:05 PM, Yafang Chen yafangche...@gmail.com mailto:yafangche...@gmail.com wrote: Hi All, I have three datasets of SeMet-incorporated protein at peak, infl and high wavelength respectively. SAD with peak dataset works well to solve the phase problem. However, MAD with all three datasets didn't work at all. The completeness of all three datasets are more than 99%. So I think radiation damage should not be a problem. Does anyone have any idea about the possible reasons that MAD didn't work in this case? Thank you so much for any of your help! Best, Yafang -- Yafang Chen Graduate Research Assistant Mesecar Lab Department of Biological Sciences Purdue University Hockmeyer Hall of Structural Biology 240 S. Martin Jischke Drive West Lafayette, IN 47907
Re: [ccp4bb] Why MAD didn't work but SAD works well
Dear Yafang, If radiation damage is not a major problem, MAD should give you more phase information than SAD, i.e. better maps, especially at low resolution. If SAD works but MAD doesn't, there are several possible explanations: 1. (most likely) your datasets are inconsistently indexed. This can hapen in various ways depending on the Laue group and the unit-cell. If you are using hkl2map the plots of the anomalous CC between the different datasets are a good quick check. Some programs (e.g. the current shelxc) will try to detect this and correct it automatically. 2. You have mixed up the wavelengths or labels of the datasets and so the dispersive difference comes out with the wrong sign. 3. You have significant radiation damage and the RIP and dispersive differences have opposite signs. Always measure the inflection dataset last so that they reinforce each other rather than canceling. 4. You have severe radiation damage and only the first (peak) dataset is usable. Best wishes, George On 08/20/2013 11:05 PM, Yafang Chen wrote: Hi All, I have three datasets of SeMet-incorporated protein at peak, infl and high wavelength respectively. SAD with peak dataset works well to solve the phase problem. However, MAD with all three datasets didn't work at all. The completeness of all three datasets are more than 99%. So I think radiation damage should not be a problem. Does anyone have any idea about the possible reasons that MAD didn't work in this case? Thank you so much for any of your help! Best, Yafang -- Yafang Chen Graduate Research Assistant Mesecar Lab Department of Biological Sciences Purdue University Hockmeyer Hall of Structural Biology 240 S. Martin Jischke Drive West Lafayette, IN 47907 -- Prof. George M. Sheldrick FRS Dept. Structural Chemistry, University of Goettingen, Tammannstr. 4, D37077 Goettingen, Germany Tel. +49-551-39-33021 or -33068 Fax. +49-551-39-22582
Re: [ccp4bb] Why MAD didn't work but SAD works well
Dear George, if #4 is correct, shouldn't he be able to get good SIRAS using the peak dataset as HA and the last collected dataset as native Jürgen .. Jürgen Bosch Johns Hopkins University Bloomberg School of Public Health Department of Biochemistry Molecular Biology Johns Hopkins Malaria Research Institute 615 North Wolfe Street, W8708 Baltimore, MD 21205 Office: +1-410-614-4742 Lab: +1-410-614-4894 Fax: +1-410-955-2926 http://lupo.jhsph.edu On Aug 20, 2013, at 5:42 PM, George Sheldrick wrote: Dear Yafang, If radiation damage is not a major problem, MAD should give you more phase information than SAD, i.e. better maps, especially at low resolution. If SAD works but MAD doesn't, there are several possible explanations: 1. (most likely) your datasets are inconsistently indexed. This can hapen in various ways depending on the Laue group and the unit-cell. If you are using hkl2map the plots of the anomalous CC between the different datasets are a good quick check. Some programs (e.g. the current shelxc) will try to detect this and correct it automatically. 2. You have mixed up the wavelengths or labels of the datasets and so the dispersive difference comes out with the wrong sign. 3. You have significant radiation damage and the RIP and dispersive differences have opposite signs. Always measure the inflection dataset last so that they reinforce each other rather than canceling. 4. You have severe radiation damage and only the first (peak) dataset is usable. Best wishes, George On 08/20/2013 11:05 PM, Yafang Chen wrote: Hi All, I have three datasets of SeMet-incorporated protein at peak, infl and high wavelength respectively. SAD with peak dataset works well to solve the phase problem. However, MAD with all three datasets didn't work at all. The completeness of all three datasets are more than 99%. So I think radiation damage should not be a problem. Does anyone have any idea about the possible reasons that MAD didn't work in this case? Thank you so much for any of your help! Best, Yafang -- Yafang Chen Graduate Research Assistant Mesecar Lab Department of Biological Sciences Purdue University Hockmeyer Hall of Structural Biology 240 S. Martin Jischke Drive West Lafayette, IN 47907 -- Prof. George M. Sheldrick FRS Dept. Structural Chemistry, University of Goettingen, Tammannstr. 4, D37077 Goettingen, Germany Tel. +49-551-39-33021 or -33068 Fax. +49-551-39-22582
