Saturday, 19 November 2011

shuttered vs continuous data collection

From: James Holton
Date: 28 October 2011 20:19

Changed the subject line to try and keep the original thread on-topic.

George is right that the acceleration/decelleration introduces error, but only in what I like to call "standing start" mode.  These days I think most facilities use "running starts" where the spindle is brought up to the degrees/second required by the the exposure just before opening the shutter at the desired "starting phi" position, and then closing the shutter at the "ending phi" with the sample still moving at full speed, allowing it to decelerate afterward. The "overhead" for achieving a stable velocity depends on the speed, of course, but this "prep time" is generally not more than a few hundred milliseconds.  This is NOT the same thing as the shutter jitter!  In fact, the "opening time" of the shutter is also not the jitter.  Even if the shutter takes 100 ms to "open", as long as that is reproducible, it won't hurt the data quality.  Shutter jitter is the rms deviation of all the "true opening times" from the average.  I have not measured this on any beamline other than my own, but here the shutter jitter is rms ~0.6 millisecond, which translates to 0.0006 degrees at 1 degree/s.  Assuming a mosaic spread of 0.5 degrees, this introduces an error as large as 0.3%.  I say "as large as" because only partials are affected by shutter jitter, not fulls.

All that said, however, if Rmerge is 3%, then sqrt(3%^2-0.3%^2)= 2.985% error is due to something other than "the shutter".

We thought ourselves quite clever here at ALS 10 years ago when we implemented the running-starts for exposures on these new-fangled (at the time) "air bearing" spindles.  Only to find that they didn't really make all that much difference when compared side-by-side with standing starts.  Unless, of course, we started doing REALLY short exposures, like 0.08s or less.  Then the noise power spectrum of the beam (flicker noise) starts to become important.  This is actually easier to measure than you might think: just put up a photodiode and record samples of the "beam intensity" as fast as you can.  The rms variation in the recorded intensity, divided by the square root of the sampling rate is the "flicker noise".  I typically get 0.15%/sqrt(Hz), which means that the average variation in integrated beam intensity over a 1s exposure is 0.15%, and that of a 0.01s exposure is 1.5%.  That is, the longer you average, the more the "flicker noise" of the beam is averaged out.  The trick with flicker noise, however, is that the relevant "exposure time" is the time it takes the relp to transit the Ewald sphere, and not the commanded "exposure time" of the image.  This means that low-mosaic crystal data collections are more sensitive to beam instability.

An important property of flicker noise is that sampling more finely and averaging afterward does not change the signal-to-noise ratio (by definition, actually).  You can try this by generating random numbers with a mean of "100" and rms error "10%" each.  If you average them in groups of 100, the result will be 100 +/- 1%, and if you take them in groups of 10 and then average 10 of those groups at a time, you still get the same answer no matter what: 1% error.  However, if you "attenuate the beam" 10 fold and collect 10x more data points (averaging 1000 values of 10 +/- 10%) you will get 10 +/- 0.3%,  reducing the flicker noise by a factor of ~3.  This may seem magical if you are used to thinking about photon-counting noise because photon-counting noise does not depend on how much time you take to collect the photons, but every other kind of noise does!

It is therefore important not to forget that counting detectors introduce a "new" kind of error: pile-up.  The Lambert omega function can be used to correct for pile-up (and that is was Pilatus does), but this correction relies on the assumption that the "true" intensity over a given acquisition time was constant (Poisson statistics).  This is definitely not the case with spots moving quickly through the Ewald sphere, and that is why Dectris recommends fine phi-slicing.  Fine slicing has advantages on any detector (Pflugrath, 1999), but it is an absolute requirement for getting good data from any counting device, including the Pilatus.  Of course, this does not mean you can't add up fine-sliced images after they are acquired to form a coarser-sliced data set, but that is a form of "lossy compression".

Formally, pile-up, beam flicker and shutter jitter fall into what I call the "% error" category, as does the noise induced by crystal vibration (usually buffeting in the cryostream) or simply a "noisy" velocity control on the spindle motor.  But since continuous-readout detectors are not immune to beam flicker, sample vibration nor velocity control instability, I decided not to mention them earlier.

-James Holton
MAD Scientist


On 10/27/2011 6:36 AM, George M. Sheldrick wrote:
There are two further complications. In non-continuous mode, the goniometer has
to accelerate at the start of a frame and decellerate at the end, then wait for
the frame to be read. So even if the shutter always functions perfectly, my
intuition tells me that it must be more accurate to rotate at constant speed
(however my intuition is often wrong). Secondly, in continuous mode, usually
not all pixels are read out at precisely the same time.

George

On Thu, Oct 27, 2011 at 11:05:01AM +0100, David Waterman wrote:
I agree with Colin here. Framing is simply a process of sampling an original
signal at some 'frequency' (related to the phi-width of each frame). At some
point, delta phi is small enough that the original signal is oversampled, and
can be reconstructed _within the bounds of noise_. Beyond that point I see no
advantage to sampling finer - and certainly not going to the limit of
representing your data in some unframed continuous readout form.

Perhaps I am missing something, and I realise this is another OT diversion from
this most fruitful of threads.

Cheers

-- David


On 27 October 2011 08:55, Martin M. Ripoll  wrote:

    Dear Colin,

    I think you understood perfectly what George was saying regarding the loss
    of information, but he will probably answer better than I.

    In any case, and for the ones that did not understand it, what George was
    telling is related to the fact that a data collection made with a
    continuous
    crystal rotation contains more information than when this information is
    transformed into frames... The loss of information that we are referring to
    has the same meaning as when we calculate electron density maps with
    different grid sizes. The finer the grid, the greater is the information on
    the map.

    But you are right saying that the shorter the interval between produced
    frames, the lower the loss of information. However, the procedure that you
    are suggesting should have some limits... otherwise the amount of
    information would grow dramatically.

    All the best,
    Martin
     



    Dear George, Martin

    I don't understand the point that one is throwing away information by
    storing in frames. If the frames have sufficiently fine intervals (given by
    some sampling theorem consideration) I can't see how one loses information.
    Can one of you explain?
    Thanks
    Colin



    --

    Dear George, dear all,

    I was just trying to summarize my point of view regarding this important
    issue when I got your e-mail, that reflects exactly my own opinion!

    Martin
    ________________________________________
    Dr. Martin Martinez-Ripoll
     



      

    This raises an important point. The new continuous readout detectors such
    as
    the
    Pilatus for beamlines or the Bruker Photon for in-house use enable the
    crystal to
    be rotated at constant velocity, eliminating the mechanical errors
    associated with
    'stop and go' data collection. Storing their data in 'frames' is an
    artifical
    construction that is currently required for the established data
    integration
    programs but is in fact throwing away information. Maybe in 10 years time
    'frames'
    will be as obsolete as punched cards!

    George

    On Wed, Oct 26, 2011 at 09:39:40AM +0100, Graeme Winter wrote:
    >  Hi James,
    >
    >  Just to pick up on your point about the Pilatus detectors. Yesterday
    >  in 2 hours of giving a beamline a workout (admittedly with Thaumatin)
    >  we acquired 400 + GB of data*. Now I appreciate that this is not
    >  really routine operation, but it does raise an interesting point - if
    >  you have loaded a sample and centred it, collected test shots and
    >  decided it's not that great, why not collect anyway as it may later
    >  prove to be useful?
    >
    >  Bzzt. 2 minutes or less later you have a full data set, and barely
    >  even time to go get a cup of tea.
    >
    >  This does to some extent move the goalposts, as you can acquire far
    >  more data than you need. You never know, you may learn something
    >  interesting from it - perhaps it has different symmetry or packing?
    >  What it does mean is if we can have a method of tagging this data
    >  there may be massively more opportunity to get also-ran data sets for
    >  methods development types. What it also means however is that the cost
    >  of curating this data is then an order of magnitude higher.
    >
    >  Also moving it around is also rather more painful.
    >
    >  Anyhow, I would try to avoid dismissing the effect that new continuous
    >  readout detectors will have on data rates, from experience it is
    >  pretty substantial.
    >
    >  Cheerio,
    >
    >  Graeme
    >
    >  *by "data" here what I mean is images, rather than information which
    >  is rather more time consuming to acquire. I would argue you get that
    >  from processing / analysing the data...
    >
    >  On 24 October 2011 22:56, James Holton  wrote:
    >  >  The Pilatus is fast, but or decades now we have had detectors that can
    read
    >  >  out in ~1s.  This means that you can collect a typical ~100 image
    dataset in
    >  >  a few minutes (if flux is not limiting).  Since there are ~150
    beamlines
    >  >  currently operating around the world and they are open about 200
    days/year,
    >  >  we should be collecting ~20,000,000 datasets each year.
    >  >
    >  >  We're not.
    >  >
    >  >  The PDB only gets about 8000 depositions per year, which means either
    we
    >  >  throw away 99.96% of our images, or we don't actually collect images
    >  >  anywhere near the ultimate capacity of the equipment we have.  In my
    >  >  estimation, both of these play about equal roles, with ~50-fold
    attrition
    >  >  between ultimate data collection capacity and actual collected data,
    and
    >  >  another ~50 fold attrition between collected data sets and published
    >  >  structures.
    >  >
    >  >  Personally, I think this means that the time it takes to collect the
    final
    >  >  dataset is not rate-limiting in a "typical" structural biology
    >  >  project/paper.  This does not mean that the dataset is of little value.
    >  >   Quite the opposite!  About 3000x more time and energy is expended
    preparing
    >  >  for the final dataset than is spent collecting it, and these efforts
    require
    >  >  experimental feedback.  The trick is figuring out how best to compress
    the
    >  >  "data used to solve a structure" for archival storage.  Do the
    "previous
    >  >  data sets" count?  Or should the compression be "lossy" about such
    >  >  historical details?  Does the stuff between the spots matter?  After
    all,
    >  >  h,k,l,F,sigF is really just a form of data compression.  In fact, there
    is
    >  >  no such thing as "raw" data.  Even "raw" diffraction images are a
    >  >  simplification of the signals that came out of the detector
    electronics.
    >  >   But we round-off and average over a lot of things to remove "noise".
    >  >   Largely because "noise" is difficult to compress.  The question of how
    much
    >  >  compression is too much compression depends on which information (aka
    noise)
    >  >  you think could be important in the future.
    >  >
    >  >  When it comes to fine-sliced data, such as that from Pilatus, the main
    >  >  reason why it doesn't compress very well is not because of the spots,
    but
    >  >  the background.  It occupies thousands of times more pixels than the
    spots.
    >  >   Yes, there is diffuse scattering information in the background pixels,
    but
    >  >  this kind of data is MUCH smoother than the spot data (by definition),
    and
    >  >  therefore is optimally stored in larger pixels.  Last year, I messed
    around
    >  >  a bit with applying different compression protocols to the spots and
    the
    >  >  background, and found that ~30 fold compression can be easily achieved
    if
    >  >  you apply h264 to the background and store the "spots" with lossless
    png
    >  >  compression:
    >  >
    >  >  http://bl831.als.lbl.gov/~jamesh/lossy_compression/
    >  >
    >  >  I think these results "speak" to the relative information content of
    the
    >  >  spots and the pixels between them.  Perhaps at least the "online
    version" of
    >  >  archived images could be in some sort of lossy-background format?  With
    the
    >  >  "real images" in some sort of slower storage (like a room full of tapes
    that
    >  >  are available upon request)?  Would 30-fold compression make the
    storage
    of
    >  >  image data tractable enough for some entity like the PDB to be able to
    >  >  afford it?
    >  >
    >  >
    >  >  I go to a lot of methods meetings, and it pains me to see the most
    brilliant
    >  >  minds in the field starved for "interesting" data sets.  The problem is
    that
    >  >  it is very easy to get people to send you data that is so bad that it
    can't
    >  >  be solved by any software imaginable (I've got piles of that!).  As a
    >  >  developer, what you really need is a "right answer" so you can come up
    with
    >  >  better metrics for how close you are to it.  Ironically, bad,
    unsolvable
    >  >  data that is connected to a right answer (aka a PDB ID) is very
    difficult to
    >  >  obtain.  The explanations usually involve protestations about being in
    the
    >  >  middle of writing up the paper, the student graduated and we don't
    >  >  understand how he/she labeled the tapes, or the RAID crashed and we
    lost
    it
    >  >  all, etc. etc.  Then again, just finding someone who has a data set
    with
    the
    >  >  kind of problem you are interested in is a lot of work!  So is figuring
    out
    >  >  which problem affects the most people, and is therefore "interesting".
    >  >
    >  >  Is this not exactly the kind of thing that publicly-accessible
    centralized
    >  >  scientific databases are created to address?
    >  >
    >  >  -James Holton
    >  >  MAD Scientist
    >  >
    >  >  On 10/16/2011 11:38 AM, Frank von Delft wrote:
    >  >>
    >  >>  On the deposition of raw data:
    >  >>
    >  >>  I recommend to the committee that before it convenes again, every
    member
    >  >>  should go collect some data on a beamline with a Pilatus detector
    [feel
    free
    >  >>  to join us at Diamond].  Because by the probable time any
    recommendations
    >  >>  actually emerge, most beamlines will have one of those (or similar),
    we'll
    >  >>  be generating more data than the LHC, and users will be happy just to
    have
    >  >>  it integrated, never mind worry about its fate.
    >  >>
    >  >>  That's not an endorsement, btw, just an observation/prediction.
    >  >>
    >  >>  phx.
    >  >>
    >  >>
    >  >>
    >  >>
    >  >>  On 14/10/2011 23:56, Thomas C. Terwilliger wrote:
    >  >>>
    >  >>>  For those who have strong opinions on what data should be
    deposited...
    >  >>>
    >  >>>  The IUCR is just starting a serious discussion of this subject. Two
    >  >>>  committees, the "Data Deposition Working Group", led by John
    Helliwell,
    >  >>>  and the Commission on Biological Macromolecules (chaired by Xiao-Dong
    Su)
    >  >>>  are working on this.
    >  >>>
    >  >>>  Two key issues are (1) feasibility and importance of deposition of
    raw
    >  >>>  images and (2) deposition of sufficient information to fully
    reproduce
    >  >>>  the
    >  >>>  crystallographic analysis.
    >  >>>
    >  >>>  I am on both committees and would be happy to hear your ideas
    (off-list).
    >  >>>  I am sure the other members of the committees would welcome your
    thoughts
    >  >>>  as well.
    >  >>>
    >  >>>  -Tom T
    >  >>>
    >  >>>  Tom Terwilliger
    >  >>> 
    >  >>>
    >  >>>
    >  >>>>>  This is a follow up (or a digression) to James comparing test set
    to
    >  >>>>>  missing reflections.  I also heard this issue mentioned before but
    was
    >  >>>>>  always too lazy to actually pursue it.
    >  >>>>>
    >  >>>>>  So.
    >  >>>>>
    >  >>>>>  The role of the test set is to prevent overfitting.  Let's say I
    have
    >  >>>>>  the final model and I monitored the Rfree every step of the way and
    can
    >  >>>>>  conclude that there is no overfitting.  Should I do the final
    >  >>>>>  refinement
    >  >>>>>  against complete dataset?
    >  >>>>>
    >  >>>>>  IMCO, I absolutely should.  The test set reflections contain
    >  >>>>>  information, and the "final" model is actually biased towards the
    >  >>>>>  working set.  Refining using all the data can only improve the
    accuracy
    >  >>>>>  of the model, if only slightly.
    >  >>>>>
    >  >>>>>  The second question is practical.  Let's say I want to deposit the
    >  >>>>>  results of the refinement against the full dataset as my final
    model.
    >  >>>>>  Should I not report the Rfree and instead insert a remark
    explaining
    >  >>>>>  the
    >  >>>>>  situation?  If I report the Rfree prior to the test set removal, it
    is
    >  >>>>>  certain that every validation tool will report a mismatch.  It does
    not
    >  >>>>>  seem that the PDB has a mechanism to deal with this.
    >  >>>>>
    >  >>>>>  Cheers,
    >  >>>>>
    >  >>>>>  Ed.
    >  >>>>>
    >  >>>>>
    >  >>>>>
    >  >>>>>  --
    >  >>>>>  Oh, suddenly throwing a giraffe into a volcano to make water is
    crazy?
    >  >>>>>                                                  Julian, King of
    Lemurs
    >  >>>>>
    >  >
    >

    --
    Prof. George M. Sheldrick FRS
    Dept. Structural Chemistry,
    University of Goettingen,
    Tammannstr. 4,
    D37077 Goettingen, Germany



No comments:

Post a Comment