From: Frank von Delft
Date: 16 October 2011 19:38
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:
----------
From: Frank von Delft
One other question (for both key issues described): what exactly is the problem the committees are aiming to address?
Because I can't help noticing that Tom's email did not spark an on-list discussion; do people actually feel either are issues? Isn't the more burning problem how best to use the 10,000s of structures we're churning out? In the grand scheme of things, they're pretty inaccurate anyway: static snapshots of crippled fragments of proteins far from their many interaction partners. So do we need 100,000s of structures instead? If so, we may soon (collectively) stop being able to care about the original dataset or how to reproduce analysis number 2238 from 2 years ago.
(No, I'm not convinced this question is relevant only to structural genomics.)
phx.
----------
From: Felix Frolow
Committees, wherever you are!
I guess that abstaining from storing the raw diffraction data in the form of frames is not very wise
whatever its size is. I regret that for some PDB entries I do not have diffraction data (needless to say that authors
do not submitted even structure factors).
I maintain a bit more than 1.2 T diffraction data starting from 2001 and all is nicely
resides on two small WD pocket disks (needless to say that I have several copies of the data).
Generally I have all data I ever collected going back to beginning of 80's, but I am to lazy to reform DAT tapes.
Sure, running Pilatus for an olympic record, we will go home with several T of data after 24 h (will we?).
But this is an abuse of the system. The final goal is the structure determination, and there are much less
good crystals everywhere in one year that one Pilatus could collect in one week.
But to decide fast if the crystal diffraction data from Pilatus is good for storage or even for measurement whatever the speed of data collection is, good data processing
software is needed. I personnaly think that there is only one, the one.
Anyhow, I think if the author wish to publish his structure, and it is important, and I am a reviewer, and it is going
to prestigious journal, I will reprocess his data and will check his way to the final crystal structure solution from the beginning.
It is as in mathematics. If someone claim that he solved a long-staing problem from the past, he will not go away from his envious colleagues, who
will drop everything and will sit and check, until they will find a mistake. What a pleasure!!!
And if there are no mistakes - chapeau !!!
FF
Dr Felix Frolow
Professor of Structural Biology and Biotechnology
Department of Molecular Microbiology
and Biotechnology
Tel Aviv University 69978, Israel
Acta Crystallographica F, co-editor
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
Hi Fellows,
I was attending the inaugural meeting of the Data Deposition Working Group
in Madrid. They are aware of the various points raised, and a
document/recommendation has been prepared that I assume will be soon made
public (John?). Amount of data seems not an insurmountable technical
problem. The most important point is that the data are getting consistently
better and contain more information than we currently make use of. Just
think of diffuse solvent contributions, commensurate and incommensurate
superstructures, split reflections, and similar stuff that presently just
gets indexed away. Better data processing software will certainly be
developed and will provide a better data model, ultimately allowing better
structure models. At the same time, almost all forgery issues disappear
automatically as an added (but imho minor) bonus.
Cheers, BR
PS: on a cynical note, what makes one believe that data processing is
carried out at any higher level of competency than say the refinement? The
only comfort here is that when it is done immediately by the beamline
fellows, it is probably done well. In any case, maybe REPROCESS_PDB is the
thing of the future.
----------
From: Bosch, Juergen
----------
From: Guenter Fritz
On the technical feasibility of storage of original data: Yes, we do already. I just checked the number of images from PILATUS 6M
we have collected so far this year : ca. 1.7 millions. End of the year
it will be more than 2 mio.
I compress everything and store it on RAID systems. Disk space is
meanwhile so cheap and storage on disk is so easy compared to tapes. So
why bothering with the large number of images?
In the end for the determination of the structure only a few datasets
will be necessary. This means maybe 0.5 Tb of compressed data to
deposit. I don't think this is too much. The special properties of PILATUS forces you to think more about your
data collection AND (!) it gives you the time to think at the beamline
about data collection. Fine phi slicing and high redundancy gives much
better data (important if your crystals are not that good). The people
from Dectris will maybe add a note here.
Best,
Guenter
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
----------
From: Felix Frolow
----------
From: Bosch, Juergen
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
----------
From: Craig A. Bingman
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
----------
From: Craig A. Bingman
----------
From: Frank von Delft
On 17/10/2011 01:52, Wladek Minor wrote:
Yes I know and agree partially.
That said: I reckon we vastly overestimate the value of individual structures; it's the ensemble that is informative. A decade from now, depositing a single structure of a protein will be seen to be as just as silly as it is currently not to deposit structure factors.
Including those "very important biomedical structures". Those things tend to become suddenly "important" (in the grand scheme) only after the ligand's biological/clinical effects could be demonstrated. And even then the structure itself is only important if it helps a chemist.
If this sounds extreme: consider how much other data it now takes to get structures into nature/science/cell. Or what happened (or rather didn't) to all those patents on structures that were such a big deal a decade ago.
phx.
----------
From: Artem Evdokimov
----------
From: John R Helliwell
Dear Colleagues,
Following on from my posting to the CCP4bb of yesterday an IUCr Forum
has been set up for Public input on the diffraction data deposition
future. Thus this Forum will record an organised set of inputs for
future reference. Instructions on how to register at this Forum can be
found there:-
http://forums.iucr.org/index.php?sid=4e83bcc36ec972f5bed1508d5bb7c05a
and/or via Brian McMahon (bm@iucr.org) in case of difficulty.
As further background information you will find at the Forum the IUCr
Diffraction Data Deposition Working Group Terms of Reference, the
Minutes of the IUCr Madrid Congress inaugural meeting and a paper from
the Commission on Biological Macromolecules (lead author Tom
Terwilliger) .
We look forward to your inputs to the Forum.
Best wishes,
John
Prof John R Helliwell DSc
Chairman of the IUCr Diffraction Data Deposition Working Group (IUCr DDDWG)
Professor John R Helliwell DSc
----------
From: Chris Morris
Some crystals are hard to make, so storing all the data the best way to get reproducibility. On the other hand, no one needs more images of lysozyme. So using the same standard for every deposition doesn't sound right.
The discussion should be held on the basis of overall cost to the research budget - not on the assumption that some costs can be externalised. It is too easy to say "you should store the images, in case I want to reprocess them sometime". IT isn't free, nor is it always cheaper than the associated experimental work. The key comparison is:
Cost of growing new crystals + cost of beam line time
With:
Cost of storing images * probability of processing them again
At present, detectors are improving more quickly than processing software. Sample preparation methods are also improving. These forces both press downward the probability that a particular image will ever be reprocessed.
regards,
Chris
____________________________________________
Chris Morris
Daresbury Lab, Daresbury, Warrington, UK, WA4 4AD
----------
From: Peter Keller
Perhaps this kind of discussion that should be continued on the IUCr
forums, once people have had a chance to register. However, to answer
these particular points:
Yes, good point, but...
... this analysis assumes that the only value of the images are for an
improved structure determination of that particular sample in order to
get new scientific insights about it. Software and methods development
have different requirements. The kinds of images that are of interest
may include:
(1) Hard-to-process images of various kinds
(2) Images collected using non-obvious data collection protocols,
especially if someone has put a lot of time and thought into designing
them.
From that point of view, it doesn't matter if a high-quality structure
of the same protein has since been refined and deposited - the original
images can still be useful.
Regards,
Peter.
--
Peter Keller
Global Phasing Ltd.,
Sheraton House,
Castle Park,
Cambridge CB3 0AX
United Kingdom
----------
From: Felix Frolow
I could not agree less. There is constant development of the software for refinement that allow to do things that were not
possible or were not necessary in the past such as intelligent refinement of occupancies of mutually exclusive sites, entities and conformations.
I frequently remeasure lysozyme crystals. I use them as a test system for the beam lines, new detectors, novel software developments, refinement improvement etc. Sometimes I am collecting data in quite different wavelength than of existing structures. And what about diffraction data from a chemically modified lysozyme molecule?
They are good data that show evolution of the beam line stations if they are keeper in historical order.
To store them all, or not to store at all…
Storage of the diffraction data is not a drinking club with muscle-bound selectors outside :-)
Felix Frolow
----------
From: Boaz Shaanan
Hi Felix,
Excuse my question, but what have you discovered about lysozyme that we haven't already known before which justifies all these efforts?
After all, we're mostly after finding solutions to biological problems, aren't we?
Boaz
Boaz Shaanan, Ph.D.
Dept. of Life Sciences
Ben-Gurion University of the Negev
Beer-Sheva 84105
Israel
----------
From: Enrico Stura
With improving techniques, we should always be making progress!
If we are trying to answer a biological question that is really important, we would be better off
improving the purification, the crystallization, the cryo-conditions instead of having to rely on
processing old images with new software.
I have 10 years worth of images. I have reprocessed very few of them and never made any
sensational progress using the new software. Poor diffraction is poor diffraction.
Money can be better spent buying a wine cellar, storage works for wine.
Enrico.
Enrico A. Stura D.Phil. (Oxon) ,
----------
From: Richard Gillilan
For the record, the amount of disk storage space per unit cost has doubled every 14 months for the last 30 years. It's an exponential relationship:
www.mkomo.com/cost-per-gigabyte
So data generated at a very high rate today, will be trivial to store in the near future. That's not to say it is cost free, of course ... but exponentially approaching free.
I worked at a Supercomputing facility for 7 years. At that time whole rooms were filled with state-of-the-art tape archive robots that could hold an unimaginable amount of data: a whole terabyte. Today, of course, that same volume costs under 100 USD with much faster I/O ... and I have personal copies of everything I generated (even digitized, uncompressed analog video).
To keep data backed up and online, of course costs something, but distributed/cloud computing is also changing that picture dramatically.
I am curious to know: those who have Pilatus 6M, for example. How much data do you generate in a year?
I suspect this is limited by beam intensity ... at the moment.
Richard
----------
From: Peter Keller
Dear Enrico,
Please don't get me wrong: what you are saying is not incorrect, but it
is only half the story.
Yes, of course!
You have left X-ray crystallography out of this list. It is a technique
like the others, and can also be improved :-)
It may be true that the number of crystallographers that are working on
improving instrumental methodology and software is small compared to the
number working on improving wet-lab techniques, but that number is not
zero, and the contribution is significant. The rest of you benefit from
that work!
Maybe so, but certain types of datasets are useful for methods and
software development, even if no new biological insights could be gained
by reprocessing them. These datasets are often hard to get hold of in
practice, especially when they are in someone's lab on a tape that
no-one has a reader for any more.
Obtaining protein, growing crystals and collecting new data in such a
way that the interesting features of those datasets are reproduced can
be much much harder than curating the images would be. This is
especially true for software-oriented people like us who don't have
regular access to wet-lab facilities.
Images have already been lost that ought to have been kept. The
questions are: how to select the datasets that are potentially of value,
and how to make sure that they don't disappear.
----------
From: 苏晓东 <gswsxd
Sorry if this message reads a bit out of date, it has been sent two days ago without success, I am trying to sent it one more time.
XDS
发件人: gswsxd@pku.edu.cn
收件人: Felix Frolow <mbfrolow@POST.TAU.AC.IL>, CCP4BB@JISCMAIL.AC.UK, xdsu@pku.edu.cn
已发送邮件: Tue, 18 Oct 2011 10:36:51 +0800 (CST)
主题: Re: [ccp4bb] IUCr committees, depositing images
Dear All,
Since John has responded to the issues raised by Tom and several responders, and he has just described the overall goal of IUCr DDD WG (Diffraction Data Deposition Working Group), I will just second on this important issue particularly for the future biological macromolecular crystallography.
In my opinion, the reasons for raw data deposition can be summarized as following (overlapping with others)
* Keep permanent experimental records; this may not be done well for all crystallographers.
* Make it possible for rigorous thorough check by reviewers as FF suggested in this mail; in fact this has been a problem for some structures published prestigious journals, particularly for low-resolution results, and these low-resolution structures tend to become more and more.
* Minimize falsification and other types of mistakes and misconducts in structural biology.
* Make it easier for validations/collaborations, I bet many structures in PDB are wrong, but there is no way to find out by the current data deposition.
* Good for methods developers;
I also agree that there are essentially no much technical problem for the raw image(data) depostion, this is especially true if we leave the deposition at or near the synchrotron sites. I hope we will join in John's forum on this important issue at:
http://forums.iucr.org/
XDS
Xiao-Dong Su, Professor
Chairman of the Commission on Biological Macromolecules (CBM),
International Union of Crystallography (IUCr);
School of Life Sciences, Peking University
100871 Beijing, China
Phone: +86-10-62759743
FAX: +86-10-62765669
E-mail: xdsu@pku.edu.cn
----------
From: Enrico Stura
Dear Peter,
How many crystallographers does it take to transform bad data into good data?
None, you need a modeller. Only a modeller can give you a structure with perfect
geometry. Data just introduces experimental errors into what would otherwise be a perfect
structure.
If you have good data do you need crystallographers?
...
Of course there all the cases in between. That ... you are right, is the other half of the story.
From a biological point of view, only borderline cases make "cents" ($+€) to store.
The experimenter in consultation with a beamline scientist at an SR facility is the best
small commitee suitable to evaluate what is worth keeping. I am sure that the images
that are worth storing for a long long time would fit on a few Tb at a reasonable cost.
Storing everything would make it harder to find something worth improving in the future.
Enrico.
----------
From: Mark J van Raaij
given that:
- storage is becoming cheaper exponentially,
- computer power is increasing exponentially,
I think there is no reason to not store all images used for solving a structure - linked to the pdb entry and properly annotated with beam centre, lambda, pixel order and all other necessary processing parameters.
Possible uses:
- programmers of data processing (and experimental phasing) software can test their programs routinely on many sets of images instead of just a few, to better judge the result of changes they have made to the code.
- the PDB (or someone else), can periodically reprocess all the data with the then state-of-the-art software, remodel where necessary and re-refine all the structures, and thus periodically improve all the structures in the pdb. This does not mean the original authors did a bad job, just that technical improvements inevitably will allow doing a better one in the future.
Sure, this does not lead directly to new biomedical insights, but indirectly it will.
The improved data processing (and experimental phasing) programs will allow solving at least a few, new, structures that otherwise would not be solvable.
The improved old structures in the pdb may allow a few, new, structures to be solved by MR that otherwise could not have been solved.
The dataset of better structures will lead to better performance of structure prediction programs (better energy calculations etc), allowing better modelling. (I was, like Enrico, very sceptical of modelling, until the first modelled structure was successfully used in MR to solve a new structure. I now believe modelling will play an increasing role).
I agree with Enrico that storing and annotating "failed" data collections will be difficult to enforce - I for one would rather concentrate on the data collections that did work in that trip (if any), or spend my time drowning sorrows, resting and then trying to get better crystals for the next trip, rather than filling in meta-data for a data storage facility, even if it is only a two-minute job.
Another thing are the in-between cases, datasets collected, processed and for which structure solution has been tried but abandoned due to the ratio of difficulty / interest being to high (difficulties being irreproducibility of crystals, some pathology in the data, etc...). These data I would gladly submit for someone else to have a go. Currently, probably no-one would, but in the future, due to better software, understanding and possible a new structure in the pdb that could be used in MR, the difficulty may go down, and some-one might do it if the potential structure is still of interest. This some-one could be myself, and the images being stored safely and centrally would make it easier also for me to recover them.
Mark J van Raaij
Laboratorio M-4
Dpto de Estructura de Macromoleculas
Centro Nacional de Biotecnologia - CSIC
c/Darwin 3
E-28049 Madrid, Spain
http://www.cnb.csic.es/content/research/macromolecular/mvraaij
----------
From: Gerard Bricogne
Dear Enrico, Frank and colleagues,
I am glad to have suggested that everyone's views on this issue should
be aired out on this BB rather than sent off-list to an IUCr committee
member: this is much more interactive and thought-provoking.
There would seem to be clear biases in some of the positions - for
instance, the statement that we overvalue individual structures and that
there is value only in their ensemble has to be seen to be coming from
someone in a structural genomics centre ;-) . However, as Wladek pointed
out, when an investigator's project is crucially dependent on a result
embodied in a deposited structure, it would be of the greatest value to that
investigator to be able to double-check how reliable some features of that
structure (especially its ligands) actually are.
On the other hand Enrico, as a specialist of crystallisation and
modelling, sees value only in improving those contributors to the task of
structure determination. This is forgetting (1) an essential capability of
crystallography: that, through experimental phasing, it can show you what a
protein looks like even if you have never seen nor modelled one before,
through the wondrous process of producing model-free electron-density maps;
and (2) an essential aspect of the task of structure determination: that it
doesn't aim at producing a model with perfect geometry, but one that best
explains the measured data and neither under- nor over-interprets them (I
realise, though, that Enrico's statement "Data just introduces experimental
errors into what would otherwise be a perfect structure" is likely to be
tongue-in-cheek ...).
When it comes to making explicit the advantages of archiving at least
the raw images that yielded the data against which a deposited PDB entry was
refined, many good reasons have been given, but I feel that
(1) there is an over-emphasis on the preservation of diffuse scattering
that has a tendency to give this archiving a nuance of "blue-skies" research
and thus to detract from its practical urgency; time will come for diffuse
scattering to be fully appreciated, but at the moment its mention acts as a
bit of a distraction, if not a turn-off in this context for people who not
not love it already;
(2) as far as I see it, the highest future benefit of having archived
raw images will result from being able to reprocess datasets from samples
containing multiple lattices ("non-merohedral twinning"). Numerous
structures are determined and refined against data obtained by integrating
only the spots from the major lattice, without rejecting those that are
corrupted by overlap by a spot from a minor lattice. This leads to
systematic errors in these data that may only be incompletely taken out by
outlier rejection at the merging stage, and will create noise or confusing
residual features in difference maps, if not false features in the main map
and therefore its interpretation by the model. In my opinion it will be the
development of methods for dealing with overlapped lattices and for the
proper treatment of such data in scaling and refinement (as is already
possible with small molecules) that will bring about the major possibility
of substantially improving deposited results by reprocessing the raw images
co-deposited with them;
(3) there is also the more immediate possibility of better removing ice
rings, or ligand powder rings, from images, than by having to throw away
certain thin shells of merged data in the structure factor file.
I see the case for raw image deposition as absolutely compelling,
especially in view of the auto-catalytic process through which their
availability will speed up the development of precisely the new methods and
software to extract better data from them and better refine models against
them. The impact of structure factor deposition on the development of better
refinement programs is there to prove that this paradigm of a chain reaction
makes total sense.
Various arguments tend to be fired off as decoys - "get better
crystals", why not "get a better post-doc"? - but they are unhelpful in the
way they prolong procrastination when what we need is to bite the bullet.
The IUCr Forum that John Helliwell pointed at already contains draft plans
for a pilot run of a reasonable scheme.
With best wishes,
Gerard.
--
----------
From: Phoebe Rice
One more consideration:
Since organization is not one of my greatest talents, I would be absolutely delighted if a databank took over the burden of archiving my raw data for me.
Phoebe
Date: 16 October 2011 19:38
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
----------
From: Frank von Delft
One other question (for both key issues described): what exactly is the problem the committees are aiming to address?
Because I can't help noticing that Tom's email did not spark an on-list discussion; do people actually feel either are issues? Isn't the more burning problem how best to use the 10,000s of structures we're churning out? In the grand scheme of things, they're pretty inaccurate anyway: static snapshots of crippled fragments of proteins far from their many interaction partners. So do we need 100,000s of structures instead? If so, we may soon (collectively) stop being able to care about the original dataset or how to reproduce analysis number 2238 from 2 years ago.
(No, I'm not convinced this question is relevant only to structural genomics.)
phx.
----------
From: Felix Frolow
Committees, wherever you are!
I guess that abstaining from storing the raw diffraction data in the form of frames is not very wise
whatever its size is. I regret that for some PDB entries I do not have diffraction data (needless to say that authors
do not submitted even structure factors).
I maintain a bit more than 1.2 T diffraction data starting from 2001 and all is nicely
resides on two small WD pocket disks (needless to say that I have several copies of the data).
Generally I have all data I ever collected going back to beginning of 80's, but I am to lazy to reform DAT tapes.
Sure, running Pilatus for an olympic record, we will go home with several T of data after 24 h (will we?).
But this is an abuse of the system. The final goal is the structure determination, and there are much less
good crystals everywhere in one year that one Pilatus could collect in one week.
But to decide fast if the crystal diffraction data from Pilatus is good for storage or even for measurement whatever the speed of data collection is, good data processing
software is needed. I personnaly think that there is only one, the one.
Anyhow, I think if the author wish to publish his structure, and it is important, and I am a reviewer, and it is going
to prestigious journal, I will reprocess his data and will check his way to the final crystal structure solution from the beginning.
It is as in mathematics. If someone claim that he solved a long-staing problem from the past, he will not go away from his envious colleagues, who
will drop everything and will sit and check, until they will find a mistake. What a pleasure!!!
And if there are no mistakes - chapeau !!!
FF
Dr Felix Frolow
Professor of Structural Biology and Biotechnology
Department of Molecular Microbiology
and Biotechnology
Tel Aviv University 69978, Israel
Acta Crystallographica F, co-editor
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
Hi Fellows,
I was attending the inaugural meeting of the Data Deposition Working Group
in Madrid. They are aware of the various points raised, and a
document/recommendation has been prepared that I assume will be soon made
public (John?). Amount of data seems not an insurmountable technical
problem. The most important point is that the data are getting consistently
better and contain more information than we currently make use of. Just
think of diffuse solvent contributions, commensurate and incommensurate
superstructures, split reflections, and similar stuff that presently just
gets indexed away. Better data processing software will certainly be
developed and will provide a better data model, ultimately allowing better
structure models. At the same time, almost all forgery issues disappear
automatically as an added (but imho minor) bonus.
Cheers, BR
PS: on a cynical note, what makes one believe that data processing is
carried out at any higher level of competency than say the refinement? The
only comfort here is that when it is done immediately by the beamline
fellows, it is probably done well. In any case, maybe REPROCESS_PDB is the
thing of the future.
----------
From: Bosch, Juergen
Wasn't that already implemented in Phenix ?
Jürgen
On Oct 16, 2011, at 4:20 PM, Bernhard Rupp (Hofkristallrat a.D.) wrote:
REPROCESS_PDB
......................
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
----------
From: Guenter Fritz
On the technical feasibility of storage of original data: Yes, we do already. I just checked the number of images from PILATUS 6M
we have collected so far this year : ca. 1.7 millions. End of the year
it will be more than 2 mio.
I compress everything and store it on RAID systems. Disk space is
meanwhile so cheap and storage on disk is so easy compared to tapes. So
why bothering with the large number of images?
In the end for the determination of the structure only a few datasets
will be necessary. This means maybe 0.5 Tb of compressed data to
deposit. I don't think this is too much. The special properties of PILATUS forces you to think more about your
data collection AND (!) it gives you the time to think at the beamline
about data collection. Fine phi slicing and high redundancy gives much
better data (important if your crystals are not that good). The people
from Dectris will maybe add a note here.
Best,
Guenter
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
As in reprocess completely from images, I meant.
----------
From: Felix Frolow
Do you mean to reprocess determination of I and sig(I) from the diffraction images automatically???
Or just to get an access to the raw data?
FF
Dr Felix Frolow
----------
From: Bosch, Juergen
I think those 0.5 TB will be not essential, the more interesting datasets are those which failed to produce a structure. That's where our processing guru's and the SHARPest thoughts should be I guess :-)
Seriously, the bad data is the data developers need (if somebody ships me a disk I can provide you with tons of bad data) , so that we can even get a structure out of something you wouldn't have even mounted to begin with. The rest is comparable to a collection of stamps, although with the benefit as BR mentioned of adding an additional hurdle/layer to falsifying structures.
Jürgen
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
Ø Do you mean to reprocess determination of I and sig(I) from the diffraction images automatically??? Or just to get an access to the raw data?
Reprocessing the images with the to-be-developed new software that will process the information in the data in full, and then using the new-and-improved refinement software to automatically refine a physically more realistic model. Sounds far-fetched now but will happen soon, in the cloud of bored petaflop iphones 8, each with 32 TB of memory….
Cheers, BR
----------
From: Craig A. Bingman
Not if you are interested in scattering that falls between reciprocal lattice maxima, or if you want to preserve the possibility of applying future data reduction packages.
----------
From: Bernhard Rupp (Hofkristallrat a.D.)
Ø Not if you are interested in scattering that falls between reciprocal lattice maxima, or if you want to preserve the possibility of applying future data reduction packages.
Yep, this is exactly what I expressed in my original statement:
"diffuse solvent contributions, commensurate and incommensurate superstructures, split reflections, and similar stuff that presently just gets indexed away"
BR
----------
From: Craig A. Bingman
I'm simultaneously embarrassed to have not read the thread to completion before replying, and also totally pumped that I would answer a question the same way as Bernhard Rupp!
----------
From: Frank von Delft
On 17/10/2011 01:52, Wladek Minor wrote:
Frank,
This is serious problem for biologists. There is a structure with ligand. The same data were re-interpreted and people did not find the ligand. This re-interpretation is not really valid until we will look into diffraction data. Biologist lost tremendous amount of time and effort looking into interfratation and ...
NOw this is very important biomedical structure.
Yes I know and agree partially.
That said: I reckon we vastly overestimate the value of individual structures; it's the ensemble that is informative. A decade from now, depositing a single structure of a protein will be seen to be as just as silly as it is currently not to deposit structure factors.
Including those "very important biomedical structures". Those things tend to become suddenly "important" (in the grand scheme) only after the ligand's biological/clinical effects could be demonstrated. And even then the structure itself is only important if it helps a chemist.
If this sounds extreme: consider how much other data it now takes to get structures into nature/science/cell. Or what happened (or rather didn't) to all those patents on structures that were such a big deal a decade ago.
phx.
Wladek
Dr. Wladek Minor
Professor of Molecular Physiology and Biological Physics
http://krzys.med.virginia.edu/CrystUVa/wladek.htm
US-mail address:
Department of Molecular Physiology and Biological Physics
University of Virginia
PO Box 800736, Charlottesville, VA 22908-0736
Fed-Ex address:
Department of Molecular Physiology and Biological Physics
1340 Jefferson Park Avenue
University of Virginia
Charlottesville, VA 22908
----------
From: Artem Evdokimov
We overestimate the value of individual structures because we're human :)
If a problem is important enough that one structure makes or breaks the case, a sensible thing to do would be to get more structures and strive to obtain some other flavor of pertinent information by methods that are unlikely to suffer from the same bias as structures.
Objectivity of the experimenter is key.
I personally would love to see the development of computationally objective (i.e. human-free) methods for integrating various kinds of scientific data. If we could escape from the brain shackles imposed on us by our crunchy-primate ancestors that'd be very nice, since we rarely need to worry about quickly counting members of our primate pod (was Cousin Mo eaten by a tiger last night, and is the tiger now full), or to make split-second decisions regarding striped creepers swinging down from branches (is it a hidden leopard, a deadly Striped Death viper, or a harmless vine?) -- but these instinctive modes of thought seriously mess up our collective ability to perform complex science.
Artem
----------
From: John R Helliwell
Dear Colleagues,
Following on from my posting to the CCP4bb of yesterday an IUCr Forum
has been set up for Public input on the diffraction data deposition
future. Thus this Forum will record an organised set of inputs for
future reference. Instructions on how to register at this Forum can be
found there:-
http://forums.iucr.org/index.php?sid=4e83bcc36ec972f5bed1508d5bb7c05a
and/or via Brian McMahon (bm@iucr.org) in case of difficulty.
As further background information you will find at the Forum the IUCr
Diffraction Data Deposition Working Group Terms of Reference, the
Minutes of the IUCr Madrid Congress inaugural meeting and a paper from
the Commission on Biological Macromolecules (lead author Tom
Terwilliger) .
We look forward to your inputs to the Forum.
Best wishes,
John
Prof John R Helliwell DSc
Chairman of the IUCr Diffraction Data Deposition Working Group (IUCr DDDWG)
--
Professor John R Helliwell DSc
----------
From: Chris Morris
Some crystals are hard to make, so storing all the data the best way to get reproducibility. On the other hand, no one needs more images of lysozyme. So using the same standard for every deposition doesn't sound right.
The discussion should be held on the basis of overall cost to the research budget - not on the assumption that some costs can be externalised. It is too easy to say "you should store the images, in case I want to reprocess them sometime". IT isn't free, nor is it always cheaper than the associated experimental work. The key comparison is:
Cost of growing new crystals + cost of beam line time
With:
Cost of storing images * probability of processing them again
At present, detectors are improving more quickly than processing software. Sample preparation methods are also improving. These forces both press downward the probability that a particular image will ever be reprocessed.
regards,
Chris
____________________________________________
Chris Morris
Daresbury Lab, Daresbury, Warrington, UK, WA4 4AD
----------
From: Peter Keller
Perhaps this kind of discussion that should be continued on the IUCr
forums, once people have had a chance to register. However, to answer
these particular points:
Yes, good point, but...
... this analysis assumes that the only value of the images are for an
improved structure determination of that particular sample in order to
get new scientific insights about it. Software and methods development
have different requirements. The kinds of images that are of interest
may include:
(1) Hard-to-process images of various kinds
(2) Images collected using non-obvious data collection protocols,
especially if someone has put a lot of time and thought into designing
them.
From that point of view, it doesn't matter if a high-quality structure
of the same protein has since been refined and deposited - the original
images can still be useful.
Regards,
Peter.
--
Peter Keller
Global Phasing Ltd.,
Sheraton House,
Castle Park,
Cambridge CB3 0AX
United Kingdom
----------
From: Felix Frolow
I could not agree less. There is constant development of the software for refinement that allow to do things that were not
possible or were not necessary in the past such as intelligent refinement of occupancies of mutually exclusive sites, entities and conformations.
I frequently remeasure lysozyme crystals. I use them as a test system for the beam lines, new detectors, novel software developments, refinement improvement etc. Sometimes I am collecting data in quite different wavelength than of existing structures. And what about diffraction data from a chemically modified lysozyme molecule?
They are good data that show evolution of the beam line stations if they are keeper in historical order.
To store them all, or not to store at all…
Storage of the diffraction data is not a drinking club with muscle-bound selectors outside :-)
Felix Frolow
----------
From: Boaz Shaanan
Hi Felix,
Excuse my question, but what have you discovered about lysozyme that we haven't already known before which justifies all these efforts?
After all, we're mostly after finding solutions to biological problems, aren't we?
Boaz
Boaz Shaanan, Ph.D.
Dept. of Life Sciences
Ben-Gurion University of the Negev
Beer-Sheva 84105
Israel
----------
From: Enrico Stura
With improving techniques, we should always be making progress!
If we are trying to answer a biological question that is really important, we would be better off
improving the purification, the crystallization, the cryo-conditions instead of having to rely on
processing old images with new software.
I have 10 years worth of images. I have reprocessed very few of them and never made any
sensational progress using the new software. Poor diffraction is poor diffraction.
Money can be better spent buying a wine cellar, storage works for wine.
Enrico.
--
Enrico A. Stura D.Phil. (Oxon) ,
----------
From: Richard Gillilan
For the record, the amount of disk storage space per unit cost has doubled every 14 months for the last 30 years. It's an exponential relationship:
www.mkomo.com/cost-per-gigabyte
So data generated at a very high rate today, will be trivial to store in the near future. That's not to say it is cost free, of course ... but exponentially approaching free.
I worked at a Supercomputing facility for 7 years. At that time whole rooms were filled with state-of-the-art tape archive robots that could hold an unimaginable amount of data: a whole terabyte. Today, of course, that same volume costs under 100 USD with much faster I/O ... and I have personal copies of everything I generated (even digitized, uncompressed analog video).
To keep data backed up and online, of course costs something, but distributed/cloud computing is also changing that picture dramatically.
I am curious to know: those who have Pilatus 6M, for example. How much data do you generate in a year?
I suspect this is limited by beam intensity ... at the moment.
Richard
----------
From: Peter Keller
Dear Enrico,
Please don't get me wrong: what you are saying is not incorrect, but it
is only half the story.
Yes, of course!
You have left X-ray crystallography out of this list. It is a technique
like the others, and can also be improved :-)
It may be true that the number of crystallographers that are working on
improving instrumental methodology and software is small compared to the
number working on improving wet-lab techniques, but that number is not
zero, and the contribution is significant. The rest of you benefit from
that work!
Maybe so, but certain types of datasets are useful for methods and
software development, even if no new biological insights could be gained
by reprocessing them. These datasets are often hard to get hold of in
practice, especially when they are in someone's lab on a tape that
no-one has a reader for any more.
Obtaining protein, growing crystals and collecting new data in such a
way that the interesting features of those datasets are reproduced can
be much much harder than curating the images would be. This is
especially true for software-oriented people like us who don't have
regular access to wet-lab facilities.
Images have already been lost that ought to have been kept. The
questions are: how to select the datasets that are potentially of value,
and how to make sure that they don't disappear.
----------
From: 苏晓东 <gswsxd
Sorry if this message reads a bit out of date, it has been sent two days ago without success, I am trying to sent it one more time.
XDS
发件人: gswsxd@pku.edu.cn
收件人: Felix Frolow <mbfrolow@POST.TAU.AC.IL>, CCP4BB@JISCMAIL.AC.UK, xdsu@pku.edu.cn
已发送邮件: Tue, 18 Oct 2011 10:36:51 +0800 (CST)
主题: Re: [ccp4bb] IUCr committees, depositing images
Dear All,
Since John has responded to the issues raised by Tom and several responders, and he has just described the overall goal of IUCr DDD WG (Diffraction Data Deposition Working Group), I will just second on this important issue particularly for the future biological macromolecular crystallography.
In my opinion, the reasons for raw data deposition can be summarized as following (overlapping with others)
* Keep permanent experimental records; this may not be done well for all crystallographers.
* Make it possible for rigorous thorough check by reviewers as FF suggested in this mail; in fact this has been a problem for some structures published prestigious journals, particularly for low-resolution results, and these low-resolution structures tend to become more and more.
* Minimize falsification and other types of mistakes and misconducts in structural biology.
* Make it easier for validations/collaborations, I bet many structures in PDB are wrong, but there is no way to find out by the current data deposition.
* Good for methods developers;
I also agree that there are essentially no much technical problem for the raw image(data) depostion, this is especially true if we leave the deposition at or near the synchrotron sites. I hope we will join in John's forum on this important issue at:
http://forums.iucr.org/
XDS
Xiao-Dong Su, Professor
Chairman of the Commission on Biological Macromolecules (CBM),
International Union of Crystallography (IUCr);
School of Life Sciences, Peking University
100871 Beijing, China
Phone: +86-10-62759743
FAX: +86-10-62765669
E-mail: xdsu@pku.edu.cn
----------
From: Enrico Stura
Dear Peter,
How many crystallographers does it take to transform bad data into good data?
None, you need a modeller. Only a modeller can give you a structure with perfect
geometry. Data just introduces experimental errors into what would otherwise be a perfect
structure.
If you have good data do you need crystallographers?
...
Of course there all the cases in between. That ... you are right, is the other half of the story.
From a biological point of view, only borderline cases make "cents" ($+€) to store.
The experimenter in consultation with a beamline scientist at an SR facility is the best
small commitee suitable to evaluate what is worth keeping. I am sure that the images
that are worth storing for a long long time would fit on a few Tb at a reasonable cost.
Storing everything would make it harder to find something worth improving in the future.
Enrico.
----------
From: Mark J van Raaij
given that:
- storage is becoming cheaper exponentially,
- computer power is increasing exponentially,
I think there is no reason to not store all images used for solving a structure - linked to the pdb entry and properly annotated with beam centre, lambda, pixel order and all other necessary processing parameters.
Possible uses:
- programmers of data processing (and experimental phasing) software can test their programs routinely on many sets of images instead of just a few, to better judge the result of changes they have made to the code.
- the PDB (or someone else), can periodically reprocess all the data with the then state-of-the-art software, remodel where necessary and re-refine all the structures, and thus periodically improve all the structures in the pdb. This does not mean the original authors did a bad job, just that technical improvements inevitably will allow doing a better one in the future.
Sure, this does not lead directly to new biomedical insights, but indirectly it will.
The improved data processing (and experimental phasing) programs will allow solving at least a few, new, structures that otherwise would not be solvable.
The improved old structures in the pdb may allow a few, new, structures to be solved by MR that otherwise could not have been solved.
The dataset of better structures will lead to better performance of structure prediction programs (better energy calculations etc), allowing better modelling. (I was, like Enrico, very sceptical of modelling, until the first modelled structure was successfully used in MR to solve a new structure. I now believe modelling will play an increasing role).
I agree with Enrico that storing and annotating "failed" data collections will be difficult to enforce - I for one would rather concentrate on the data collections that did work in that trip (if any), or spend my time drowning sorrows, resting and then trying to get better crystals for the next trip, rather than filling in meta-data for a data storage facility, even if it is only a two-minute job.
Another thing are the in-between cases, datasets collected, processed and for which structure solution has been tried but abandoned due to the ratio of difficulty / interest being to high (difficulties being irreproducibility of crystals, some pathology in the data, etc...). These data I would gladly submit for someone else to have a go. Currently, probably no-one would, but in the future, due to better software, understanding and possible a new structure in the pdb that could be used in MR, the difficulty may go down, and some-one might do it if the potential structure is still of interest. This some-one could be myself, and the images being stored safely and centrally would make it easier also for me to recover them.
Mark J van Raaij
Laboratorio M-4
Dpto de Estructura de Macromoleculas
Centro Nacional de Biotecnologia - CSIC
c/Darwin 3
E-28049 Madrid, Spain
http://www.cnb.csic.es/content/research/macromolecular/mvraaij
----------
From: Gerard Bricogne
Dear Enrico, Frank and colleagues,
I am glad to have suggested that everyone's views on this issue should
be aired out on this BB rather than sent off-list to an IUCr committee
member: this is much more interactive and thought-provoking.
There would seem to be clear biases in some of the positions - for
instance, the statement that we overvalue individual structures and that
there is value only in their ensemble has to be seen to be coming from
someone in a structural genomics centre ;-) . However, as Wladek pointed
out, when an investigator's project is crucially dependent on a result
embodied in a deposited structure, it would be of the greatest value to that
investigator to be able to double-check how reliable some features of that
structure (especially its ligands) actually are.
On the other hand Enrico, as a specialist of crystallisation and
modelling, sees value only in improving those contributors to the task of
structure determination. This is forgetting (1) an essential capability of
crystallography: that, through experimental phasing, it can show you what a
protein looks like even if you have never seen nor modelled one before,
through the wondrous process of producing model-free electron-density maps;
and (2) an essential aspect of the task of structure determination: that it
doesn't aim at producing a model with perfect geometry, but one that best
explains the measured data and neither under- nor over-interprets them (I
realise, though, that Enrico's statement "Data just introduces experimental
errors into what would otherwise be a perfect structure" is likely to be
tongue-in-cheek ...).
When it comes to making explicit the advantages of archiving at least
the raw images that yielded the data against which a deposited PDB entry was
refined, many good reasons have been given, but I feel that
(1) there is an over-emphasis on the preservation of diffuse scattering
that has a tendency to give this archiving a nuance of "blue-skies" research
and thus to detract from its practical urgency; time will come for diffuse
scattering to be fully appreciated, but at the moment its mention acts as a
bit of a distraction, if not a turn-off in this context for people who not
not love it already;
(2) as far as I see it, the highest future benefit of having archived
raw images will result from being able to reprocess datasets from samples
containing multiple lattices ("non-merohedral twinning"). Numerous
structures are determined and refined against data obtained by integrating
only the spots from the major lattice, without rejecting those that are
corrupted by overlap by a spot from a minor lattice. This leads to
systematic errors in these data that may only be incompletely taken out by
outlier rejection at the merging stage, and will create noise or confusing
residual features in difference maps, if not false features in the main map
and therefore its interpretation by the model. In my opinion it will be the
development of methods for dealing with overlapped lattices and for the
proper treatment of such data in scaling and refinement (as is already
possible with small molecules) that will bring about the major possibility
of substantially improving deposited results by reprocessing the raw images
co-deposited with them;
(3) there is also the more immediate possibility of better removing ice
rings, or ligand powder rings, from images, than by having to throw away
certain thin shells of merged data in the structure factor file.
I see the case for raw image deposition as absolutely compelling,
especially in view of the auto-catalytic process through which their
availability will speed up the development of precisely the new methods and
software to extract better data from them and better refine models against
them. The impact of structure factor deposition on the development of better
refinement programs is there to prove that this paradigm of a chain reaction
makes total sense.
Various arguments tend to be fired off as decoys - "get better
crystals", why not "get a better post-doc"? - but they are unhelpful in the
way they prolong procrastination when what we need is to bite the bullet.
The IUCr Forum that John Helliwell pointed at already contains draft plans
for a pilot run of a reasonable scheme.
With best wishes,
Gerard.
--
--
----------
From: Phoebe Rice
One more consideration:
Since organization is not one of my greatest talents, I would be absolutely delighted if a databank took over the burden of archiving my raw data for me.
Phoebe
----------
From: Ed Pozharski
Certainly, one could argue here that the current PDB policy that
requires the deposition of the processed data already provides that
option. I must add though that I find it disturbingly easy to
manipulate the Fo's to produce any ligand anywhere and generate a fake
dataset that is essentially impossible to detect as such. It is only a
matter of time until someone under pressure does this, in all
likelihood, this might have already happened. True, diffraction images
can also be ultimately manipulated, but that requires much higher level
of sophistication, and the hope is that individuals capable of such feat
have better things to do with their skills.
It seems to me that the cost of storage is so cheap these days that even
if reducing the chances of another retraction disaster is the only
benefit, it is worth it. There are many other reasons why the benefits
of image deposition outweigh the costs, which reminds me of an old joke:
- Colonel, why your cannons did not fire?
- General, there were five reasons. First, we had no gunpowder,
second...
- This one is enough!
----------
From: <mjvdwoerd
Phoebe,
Just automate the archiving and come up with a reasonable scheme how to. Ours is that data sets are called:
userid_yearmonth_projectid_#
Userid is derived from the login into CrystalClear (oops, free advertizing), projectid is set by the PI (so she can remember 10 years from now what in the world these data are all about) and the users are asked (threatened) to call their data sets "projectid_#" (and not the ubiquitous "test"). We have a script that automatically archives everything away from our data collection computer into an archive - activated by an icon on the desktop - and it adds the userid and date to the filename. This has the nice added advantage that the data collection disk stays clean. This only breaks when we collect synchrotron data (which is all the time) because our synchrotron remote scientist who collects the data cannot (should not) be threatened. :-) I then rename all data sets for archiving so the naming is consistent and you can actually make (say in pdf) an index of all the data you have, organized by user, date, or project.
Our policy is that the PI decides if data should be maintained or if it really can go (no diffraction, really a test crystal to see that the crystal is in the beam etc). In practice this doesn't happen so someone else makes the decision. We tend to err on the side of caution. We tend to think that all results should be saved, unless it is blatantly obvious that there is no point. Storage is cheap (and cheaper every time you think of it).
After you automate in the previously agreed upon scheme, it is somewhat easier to find things back because if you can remember who collected it, or approximately when it was done, or what the project was, you can find it. The pain was up front: to come up with a scheme, to enable a rigorous naming convention and to implement it (data collection computer and archive are not physically on the same computer etc).
Maybe the Committee is also thinking about that issue - how are you going to keep all the data manageable and searchable. Presumably by something like a PDB id (this seems to make sense for published/deposited structures) but for "things that did not make it to PDB" one would have to come up with another plan.
Mark
Just automate the archiving and come up with a reasonable scheme how to. Ours is that data sets are called:
userid_yearmonth_projectid_#
Userid is derived from the login into CrystalClear (oops, free advertizing), projectid is set by the PI (so she can remember 10 years from now what in the world these data are all about) and the users are asked (threatened) to call their data sets "projectid_#" (and not the ubiquitous "test"). We have a script that automatically archives everything away from our data collection computer into an archive - activated by an icon on the desktop - and it adds the userid and date to the filename. This has the nice added advantage that the data collection disk stays clean. This only breaks when we collect synchrotron data (which is all the time) because our synchrotron remote scientist who collects the data cannot (should not) be threatened. :-) I then rename all data sets for archiving so the naming is consistent and you can actually make (say in pdf) an index of all the data you have, organized by user, date, or project.
Our policy is that the PI decides if data should be maintained or if it really can go (no diffraction, really a test crystal to see that the crystal is in the beam etc). In practice this doesn't happen so someone else makes the decision. We tend to err on the side of caution. We tend to think that all results should be saved, unless it is blatantly obvious that there is no point. Storage is cheap (and cheaper every time you think of it).
After you automate in the previously agreed upon scheme, it is somewhat easier to find things back because if you can remember who collected it, or approximately when it was done, or what the project was, you can find it. The pain was up front: to come up with a scheme, to enable a rigorous naming convention and to implement it (data collection computer and archive are not physically on the same computer etc).
Maybe the Committee is also thinking about that issue - how are you going to keep all the data manageable and searchable. Presumably by something like a PDB id (this seems to make sense for published/deposited structures) but for "things that did not make it to PDB" one would have to come up with another plan.
Mark
----------
From: Tom Peat
If we are talking schemes, here is another one that we use that might be considered:
Date/person/project/barcode/well#/crystal#
At the Australian synchrotron, a directory is automatically made with the date, so that is our starting point.
We sometimes skip the person, but project-barcode-well are always there, as then it can correspond to our crystal database.
I imagine that most high throughput centres use barcodes, so barcodes and well numbers would be good things to have in the path.
Cheers, tom
What do you consider good ? r.m.s.d of 2.5 Å ? fatal for drug design.
No just the right programs. Which brings us back to the hypothetical question of why do we care ? I mean more precisely the question each of us is trying to answer in their own research program.
Crystallography is a high-end tool, not quite as simple as a western blot or ELISA but in the end it's just a tool to answer some critical questions.
----------
From: Felix Frolow
Sure they will
There is no irony in what I say
FF
----------
From: Alun Ashton
Sorry for my boring response………
'Short' bit:
Has anyone here considered DOI's onto data? Facility sites within Europe and planning to make this available, I hope to do a proof of principle this year on data from Diamond (volunteers?). But as an example the ISIS neutron site on the same campus as us have started to do this, as a random example you can go to http://doi.org and put in the DOI reference 10.5286/ISIS.E.24079772 (catchy), but this takes you to a landing page where you can see some details of the data and an actual citable (I think) reference to the data for a publication. There is a link to the data but the data has not yet been made public by the author or facility, but at least its (should be) there and will eventually be public. The responsibility is now on the facility for looking after and making the data available.
This wouldn't suit everyone, and also there is the issue of home sources, but tools are under development to make this easy. I could easily imagine that within the UK STFC would probably host something like this for non facility data (it is actually them who host Diamond data for us)…. Maybe at a nominal cost of course….
Long bit:
Something similar at Diamond, /dls/$Beamline_name/data/$Year/$proposal-$visit and permissions are set accordingly so only the people on the visit or the PI's of the proposal can see the data therein. What happens within that directory is still pretty much the users choice at the moment. Though once the data is collected its read only and its all recorded in ISPyB (beamline database with web pages developed at ESRF and Diamond). You can also record details of the sample and link the data collections to it.
There is an EU funded initiative that I have make the IUCr DDDwg aware of in Europe called PanData (http://www.pan-data.eu/) which includes most of Europe's X-ray and neutron sites. Under this initiative the facilities are attempting to standardise on authorisation, data formats, some software, access policies (making data public) data retention and cataloguing.
Here we've been a bit lucky to get ahead on this and we have been able to keep a copy of all our data off all beamlines, raw and processed on tape (that's just under 200Tb and 53 million catalogued files so far, lots of data including processed data its not yet catalogued but is on tape). We are currently beta testing a web page to the data that is catalogued, so anyone who has collected data at diamond should be able to get it from https://icat.diamond.ac.uk. The data will probably be coming off tape so can take a while, also it's a little bit clumsy as an interface but it will get better. This is the same technology as is being proposed for PanData facilities, but the backend of the actual data archive is the choice of each facility, ours is hosted in a tape robot by STFC at the moment.
This is by no means the only solution out there but DOI's could help unify the solutions?
Alun
>
----------
From: Eleanor Dodson
Has anyone raided the point that while archiving is good, it will only be generally useful if the image HEADERS are informative and use a comprehensible format - and the data base is documented...
Eleanor
----------
From: Graeme Winter
Hi Eleanor,
So far I have managed to "lurk" on this one - keeping an eye on things
but not getting involved. However this has prompted me to respond!
There are a number of issues here:
- whether to publish data
- how to publish data
- how to make the published data useful
- whether to centrally archive that data
- whether to standardise the data, and if so how
- who should pay
- moving the data around
etc. The image headers issue is clearly one of these, however properly
resolving this has thus far proved to be if not intractable, at least
challenging. There is at least one "standard" comprehensible format
which is out there and has been for a while, but is thus far lacking
widespread adoption. However, and this is a huge however, we really
need to ask how much effort it is worth putting into (handling) each
data set. The more effort is needed to make the data available, the
less likely it is that it will ever become available. I know from
experience that even when people want you to have data the activation
energy is substantial. Making the process more complex will decrease
the likelihood of it occurring. If on the other hand we lower this
barrier (i.e. you make available the data you have on the hard disk
exactly how it is) you lower this barrier some way. We can make this
"useful" by also including the processing logs - i.e. your mosflm /
denzo / XDS log file - so that anyone who really wants to know can
look and figure it out.
This biases the effort in the right direction. Even if every data set
was perfectly published, it is pretty unlikely that any given data set
would be re-analysed - unless it is really interesting. If it is
really interesting, it is then worth the effort to figure out the
parameters, so make this possible if inconvenient. As I see it, a main
benefit of this is to allow that occasional questionable structure to
be looked at really hard - and simply the need to upload the original
data and processing results would help in reducing the possibility of
depositing anything "fake".
Another factor I am painfully aware of is that disks die - certainly
mine do. This is all well and good - however the time it takes to move
4TB of data from one drive to another is surprisingly long, as even
with things like eSATA we have oceans of data connected by hosepipes.
Moving all of your raw data from a visit to the synchrotron (which
could easily be TB's) home is a challenge enough - subsequently moving
it to some central archive could be a real killer. Equally making the
data public from your own lab is also difficult for many. At least
facility sites are equipped to meet some of these challenges.
So - as I can see it we have much bigger fish to fry than getting the
headers for historical data standardised! Current and future data,
well that's a different pot of worms. And that's from someone who
really does care about image headers ;o)
Cheerio,
Graeme
----------
From: Thomas C. Terwilliger
John Helliwell points out to me that it might be useful to know what MX
crystallographic data researchers in different countries are already
expected to deposit or save. He notes that research funding agencies in
the UK expect researchers to preserve their raw experimental data for at
least 5 years.
Can people comment on what data they are already expected to save in their
countries, and what mechanisms they already have for facilitating this
(for example the Australian Research Council TARDIS initiative which helps
store raw diffraction images)?
If you want to see or post comments on this thread on the IUCR Forum you
can do that at:
http://forums.iucr.org/viewforum.php?f=21
Also of course posts here on this mailing list are fine.
-Tom T
----------
From: James Holton
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
----------
From: Jrh
Dear James,
This is technically ingenious stuff.
Perhaps it could be applied to help the 'full archive challenge' ie containing many data sets that will never lead to publication/ database deposition?
However for the latter,publication/deposition, subset you would surely not 'tamper' with the raw measurements?
The 'grey area' between the two clearcut cases ie where eventually publication/deposition MAY result then becomes the challenge as to whether to compress or not? (I would still prefer no tampering.)
Greetings,
----------
From: James Stroud
Just to put this into dollars. If each dataset is about 17 GB in size, then that's about 14 TB of storage that needs to come online every year to store the raw data for every structure. A two second search reveals that Newegg has a 3GB hitachi for $200. So that's about $1000 / year of storage for the raw data behind PDB deposits.
James
----------
From: Herbert J. Bernstein
To be fair to those concerned about cost, a more conservative estimate
from the NSF RDLM workshop last summer in Princeton is $1,000 to $3,000
per terabyte per year for long term storage allowing for overhead in
moderate-sized institutions such as the PDB. Larger entities, such
as Google are able to do it for much lower annual costs in the range of
$100 to $300 per terabyte per year. Indeed, if this becomes a serious
effort, one might wish to consider involving the large storage farm
businesses such as Google and Amazon. They might be willing to help
support science partially in exchange for eyeballs going to their sites.
Regards,
H. J. Bernstein
--
=====================================================
Herbert J. Bernstein, Professor of Computer Science
Dowling College, Brookhaven Campus, B111B
1300 William Floyd Parkway, Shirley, NY, 11967
=====================================================
----------
From: Pete Meyer
This is probably an idea that has already been tried (or discarded as unsuitable for reasons that don't occur to me at the moment) - but why not start with good crystals (such as lysozyme) and deliberately make them worse? Exactly how would depend on what kind of methods you were trying to develop - but I'd imaging "titrating in" organic solvents/detergents would be able to turn a well-diffracting crystal into a poor one (with a known, or at least knowable, answer). Deliberately causing radiation damage, or using known poor cryo-conditions would also work - probably the type of "badness" in the data would be different.
I don't think you'd be able to tune solvent content or number of anomalous scatterers by damaging good crystals. This would also require a decent number of crystals (but lysozyme is reasonably inexpensive). But making good crystals from bad ones is difficult - making bad ones from good ones shouldn't be.
Any ideas why this wouldn't work (or citations where it did)?
Pete
----------
From: Nat Echols
It probably depends on what kind of methods you are trying to develop, but there are any number of reasons why it wouldn't be a complete substitute for bad data found in the wild (you've already listed a couple), all of them essentially coming down to the same problem: you're limiting what kind of imperfections and pathologies you'll see. How do you get a crystal that is just slightly split, for instance? Or one that diffracts to high resolution but has a mosaicity of 2.5 degrees? I can think of plenty of other special cases, and the data processing people probably have dozens of ideas. We could spend a month abusing lysozyme crystals and still not come up with anywhere near the diversity that the community can generate with their membrane proteins, glycoproteins, RNA, etc. This is true for everything from data processing to refinement. (Fortunately, for refinement we already have most of what we need in the PDB.)
That doesn't mean it's not a good idea anyway; I've considered doing something similar to generate pairs of realistic high-resolution and low-resolution data. (Probably not using lysozyme - it's much too small to be representative of the average 4A structure.) It's a good summer project for a capable undergraduate student, if you have synchrotron time to burn.
-Nat
----------
From: Frank von Delft
Since when has the cost of any project been limited by the cost of hardware? Someone has to implement this -- and make a career out of it; thunderingly absent from this thread has been the chorus of volunteers who will write the grant.
phx
phx
----------
From: Graeme Winter
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...
----------
From: Matthew BOWLER
The archiving of all raw data and subsequently making it public is something that the large facilities are currently debating whether to do. Here at the ESRF we store user data for only 6 months (and I believe that it is available longer on tape) and we already have trouble with capacity. My personal view is that facilities should take the lead on this - for MX we already have a very good archiving system - ISPyB - also running at Diamond. ISPyB stores lots of meta data and jpgs of the raw images but not the images themselves but a link to the location of the data with an option to download if still available. My preferred option would be to store all academically funded data and then make it publicly available after say 2-5 years (this will no doubt spark another debate on time limits, special dispensation etc). What needs to be thought about is how to order the data and how to make sure that the correct meta data are stored with each data set - this will rely heavily on user input at the time of the experiment rather than gathering together data sets for depositions much later. As already mentioned, this type of resource could be extremely useful for developers and also as a general scientific resource. Smells like an EU grant to me. Cheers, Matt.
-- Matthew Bowler Structural Biology Group European Synchrotron Radiation Facility B.P. 220, 6 rue Jules Horowitz F-38043 GRENOBLE CEDEX FRANCE ===================================================
----------
From: George M. Sheldrick
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
--
Prof. George M. Sheldrick FRS
Dept. Structural Chemistry,
University of Goettingen,
Tammannstr. 4,
D37077 Goettingen, Germany
----------
From: Loes Kroon-Batenburg
Dear James,
Good analysis! You bring up important points. Your estimation says: we collect 1/50 * 20,000,000 = 400,000 data sets of which only 8,000 get deposited.
An average Pilatus data set (0.1 degree scan) takes about 4 Gb (compressed, without loosing information. EVAL can read those!). Storing the 8,000 data sets, as James Stroud mentions, can not be the problem.
It is the 392,000 other data sets that we have to find a home for. That would be 1568 Tb and would cost 49,000 $/year. This may be a slight overestimation, but it shows us the problems we face if we want to store ALL raw data.
Even if we would find a way to store all these data, how would we set up a useful data base? If we store all data by name, date and beamline, we will in the end inevitable be drowning is a sea of information. It is very unlikely that the very interesting data sets will ever be found and used.
It would be much more useful if every data sets would be annotated by the user or beam line scientist. Like: "impossible to index", "bad data from integration step", "overlap", "diffuse streaks" etc. Such information could be part of the meta data. This however, takes time and may not fit the eagerness to get results from one of the other data sets recorded at the same synchrotron trip.
I am afraid that just throwing data sets in a big pool, will not be very useful.
Loes.
--
__________________________________________
Dr. Loes Kroon-Batenburg
Dept. of Crystal and Structural Chemistry
Bijvoet Center for Biomolecular Research
Utrecht University
Padualaan 8, 3584 CH Utrecht
The Netherlands
____________________
----------
From: John R Helliwell
Dear Frank,
re 'who will write the grant?'.
This is not as easy as it sounds, would that it were!
There are two possible business plans:-
Option 1. Specifically for MX is the PDB as the first and foremost
candidate to seek such additional funds for full diffraction data
deposition for each future PDB deposiition entry. This business plan
possibility is best answered by PDB/EBI (eg Gerard Kleywegt has
answered this in the negative thus far at the CCP4 January 2010).
Option 2 The Journals that host the publications could add the cost to
the subscriber and/or the author according to their funding model. As
an example and as a start a draft business plan has been written by
one of us [JRH] for IUCr Acta Cryst E; this seemed attractive because
of its simpler 'author pays' financing. This proposed business plan is
now with IUCr Journals to digest and hopefully refine. Initial
indications are that Acta Cryst C would be perceived by IUCr Journals
as a better place to start considering this in detail, as it involves
fewer crystal structures than Acta E and would thus be more
manageable. The overall advantage of the responsibility being with
Journals as we see it is that it encourages such 'archiving of data
with literature' across all crystallography related techniques (single
crystal, SAXS, SANS, Electron crystallography etc) and fields
(Biology, Chemistry, Materials, Condensed Matter Physics etc) ie not
just one technique and field, although obviously biology is dear to
our hearts here in the CCP4bb.
Yours sincerely,
John and Tom
John Helliwell and Tom Terwilliger
----------
From: Gerard Bricogne
Dear John and colleagues,
There seem to be a set a centrifugal forces at play within this thread
that are distracting us from a sensible path of concrete action by throwing
decoys in every conceivable direction, e.g.
* "Pilatus detectors spew out such a volume of data that we can't
possibly archive it all" - does that mean that because the 5th generation of
Dectris detectors will be able to write one billion images a second and
catch every scattered photon individually, we should not try and archive
more information than is given by the current merged structure factor data?
That seems a complete failure of reasoning to me: there must be a sensible
form of raw data archiving that would stand between those two extremes and
would retain much more information that the current merged data but would
step back from the enormous degree of oversampling of the raw diffraction
pattern that the Pilatus and its successors are capable of.
* "It is all going to cost an awful lot of money, therefore we need a
team of grant writers to raise its hand and volunteer to apply for resources
from one or more funding agencies" - there again there is an avoidance of
the feasible by invocation of the impossible. The IUCr Forum already has an
outline of a feasibility study that would cost only a small amount of
joined-up thinking and book-keeping around already stored information, so
let us not use the inaccessibility of federal or EC funding as a scarecrow
to justify not even trying what is proposed there. And the idea that someone
needs to decide to stake his/her career on this undertaking seems totally
overblown.
Several people have already pointed out that the sets of images that
would need to be archived would be a very small subset of the bulk of
datasets that are being held on the storage systems of synchrotron sources.
What needs to be done, as already described, is to be able to refer to those
few datasets that gave rise to the integrated data against which deposited
structures were refined (or, in some cases, solved by experimental phasing),
to give them special status in terms of making them visible and accessible
on-line at the same time as the pdb entry itself (rather than after the
statutory 2-5 years that would apply to all the rest, probably in a more
off-line form), and to maintain that accessibility "for ever", with a link
from the pdb entry and perhaps from the associated publication. It seems
unlikely that this would involve the mobilisation of such large resources as
to require either a human sacrifice (of the poor person whose life would be
staked on this gamble) or writing a grant application, with the indefinite
postponement of action and the loss of motivation this would imply.
Coming back to the more technical issue of bloated datasets, it is a
scientific problem that must be amenable to rational analysis to decide on a
sensible form of compression of overly-verbose sets of thin-sliced, perhaps
low-exposure images that would already retain a large fraction, if not all,
of the extra information on which we would wish future improved versions of
processing programs to cut their teeth, for a long time to come. This
approach would seem preferable to stoking up irrational fears of not being
able to cope with the most exaggerated predictions of the volumes of data to
archive, and thus doing nothing at all.
I very much hope that the "can do" spirit that marked the final
discussions of the DDDWG (Diffraction Data Deposition Working Group) in
Madrid will emerge on top of all the counter-arguments that consist in
moving the goal posts to prove that the initial goal is unreachable.
--
===============================================================
* *
* Gerard Bricogne
----------
From: Patrick Shaw Stewart
Could you perhaps use the principle of "capture storage" that is used by wild-life photographers with high-speed cameras?
The principle is that the movie is written to the same area of memory, jumping back to the beginning when it is full (this part is not essential, but it makes the principle clear). Then, when the photographer takes his finger off the trigger, the last x seconds is permanently stored. So you keep your wits about you, and press the metaphorical "store" button just after you have got the movie in the can so to speak
Just a thought
Just a thought
Patrick
--
----------
From: Gloria Borgstahl
I just want to jump in to state that I am ALL FOR the notion of
depositing the images that go with the structure factors and the
refined structure.
Through the years, I have been interviewing folks about the strange
satellite diffraction they saw, but ignored,
used the mains that they could integrate and deposited that structure,
does not help me to
justify the existance of modulated protein crystals to reviewers.
But if I could go and retrieve those images, and reanalyze with new methods.
Dream come true. Reviewers convinced.
----------
From: Jacob Keller
Is anyone seriously questioning whether we should archive the images
used for published structures? That amount of space is trivial, could
be implemented just as another link in the PDB website, and would be
really helpful in some cases. One person could set it up in a day! You
could just make it a policy: no images, no PDB submission, no
publishing!
Jacob
--
*******************************************
Jacob Pearson Keller
Northwestern University
Medical Scientist Training Program
----------
From: Frank von Delft
Cool - we've found our volunteer!!
----------
From: Jacob Keller
Touche! But alas, I have no access to the PDB's server, so...
JPK
On Wed, Oct 26, 2011 at 11:54 AM, Frank von Delft
----------
From: Herbert J. Bernstein
Dear Colleagues,
Gerard strikes a very useful note in pleading for a "can-do"
approach. Part of going from "can-do" to "actually-done"
is to make realistic estimates of the costs of "doing" and
then to adjust plans appropriately to do what can be afforded
now and to work towards doing as much of what remains undone
as has sufficient benefit to justify the costs.
We appear to be in a fortunate situation in which some
portion of the raw data behind a signficant portion of the
studies released in the PDB could probably be retained for some
significant period of time and be made available for further
analysis. It would seem wise to explore these possibilities
and try to optimize the approaches used -- e.g. to consider
moves towards well documented formats, and retention of critical
metadata with such data to help in future analysis.
Please do not let the perfect be the enemy of the good.
Regards,
Herbert Dowling College, Kramer Science Center, KSC 121
Idle Hour Blvd, Oakdale, NY, 11769
=====================================================
----------
From: James Stroud
This idea seems equivalent to only storing permanently those datasets that actually yield structures worthy of deposition.
James
----------
From: Colin Nave
I have been nominated by the IUCr synchrotron commission (thanks colleagues!) to represent them for this issue. However, at the moment, this is a personal view.
1. For archiving "raw" diffraction image data for structures in the PDB, it should be the responsibility of the worldwide PDB. They are by far the best place to do it and as Jacob says the space requirements are trivial. Gerard K's negative statement at CCP4-2010 sounds rather ex cathedra (in increasing order of influence/power do we have the Pope, US president, the Bond Market and finally Gerard K?). Did he make the statement in a formal presentation or in the bar? More seriously, I am sure he had good reasons (e.g. PDB priorities) if he did make this statement. It would be nice if Gerard could provide some explanation.
2. I agree with the "can do" attitude at Madrid as supported by Gerard B. Setting up something as best one can with existing enthusiasts will get the ball rolling, provide some immediate benefit and allow subsequent improvements.
3. Ideally the data to be deposited should include all stages e.g. raw images, "corrected" images, MIR/SAD/MAD images, unmerged integrated intensities, scaled, merged etc. Plus the metadata, software & versions used for the various stages. Worrying too much about all of this should not of course prevent a start being made. (An aside. I put the "corrected" in quotes because the raw images have fewer errors. The subsequent processing for detector distortions etc. depend on an imperfect model for the detector. I don't like the phrase data correction).
4. Doing this for PDB depositions would then provide a basis for other data which did not result in PDB depositions. There seems to be a view that the archiving of this should be the responsibility of the synchrotrons which generated the data. This should be possible for some synchrotrons (e.g. Diamond) where there is pressure in any case from their funders to archive all data generated at the facility. However not all synchrotrons will be able to do this. There is also the issue of data collected at home sources. Presumably it will require a few willing synchrotrons to pioneer this in a coordinated way. Hopefully others will then follow. I don't think we can expect the PDB to archive the 99.96% of the data which did not result in structures.
5. My view is that for data in the PDB the same release rules should apply for the images as for the other data. For other data, the funders of the research might want to define release rules. However, we can make suggestions!
6. Looking to the future, there is FEL data coming along, both single molecule and nano-crystals (assuming the FEL delivers for these areas).
7. I agree with Gerard B - "as far as I see it, the highest future benefit of having archived raw images will result from being able to reprocess datasets from samples containing multiple lattices"
My view is that all crystals are, to a greater or lesser extent, subject to this. We just might not see it easily as the detector resolution or beam divergence is inadequate. Just think we could have several structures (one from each lattice) each with less disorder rather than just one average structure. Not sure whether Gloria's modulated structures would be as ubiquitous but her argument is along the same lines.
Regards
Colin
----------
From: Martin M. Ripoll
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
Research Professor
Department of Crystallography & Structural Biology
Consejo Superior de Investigaciones Científicas
Spanish National Research Council
www.csic.es
----------
From: Gerard Bricogne
Dear Colin,
Thank you for accepting the heavy burden of responsibility your
colleagues have thrown onto your shoulders ;-) . It is great that you are
entering this discussion, and I am grateful for the support you are bringing
to the notion of starting something at ground level and learning from it,
rather that staying in the realm of conjecture and axiomatics, or entering
the virility contest as to whose beamline will make raw data archiving most
impossible.
One small point, however, about your statement regarding multiple
lattices, that
"... all crystals are, to a greater or lesser extent, subject to this.
I am not sure that what you describe in your last sentence is a realistic
prospect, nor that it would in any case constitute the main advantage of
better dealing with multiple lattices. The most important consequence of
their multiplicity is that their spots overlap and corrupt each other's
intensities, so that the main benefit of improved processing would be to
mitigate that mutual corruption, first by correctly flagging overlaps, then
by partially trying to resolve those overlaps internally as much as scaling
procedures will allow (one could call that "non-merohedral detwinning" - it
is done e.g. by small-molecule softeware), and finally by adapting
refinement protocols to recognise that they may have to refine against
measurements that are a mixture of several intensities, to a degree and
according to a pattern that varies from one observation to another (unlike
regular twinning).
Currently, if a "main" lattice can be identified and indexed, one tends
to integrate the spots it successfully indexes, and to abstain from worrying
about the accidental corruption of the resulting intensities by accidental
overlaps with spots of the other lattices (whose existence is promptly
forgotten). It is the undoing of that corruption that would bring the main
benefit, not the fact that one could see several variants of the structure
by fitting the data attached to the various lattices: that would be possible
only if overlaps were negligible. The prospects for improving electron
density maps by reprocessing raw images in the future are therefore
considerable for mainstream structures, not just as a way of perhaps teasing
interestingly different structures from each lattice in infrequent cases.
I apologise if I have laboured this point, but I am concerned that
every slight slip of the pen that makes the benefits of future reprocessing
look as if they will just contribute to splitting hairs does a disservice to
this crucial discussion (and hence, potentially, to the community) by
belittling the importance and urgency of the task.
With best wishes,
Gerard (B.)
--
----------
From: Colin Nave
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
----------
From: Colin Nave
Dear Gerard
Yes, perhaps I was getting a bit carried away with the possibilities. Although I believe that, with high resolution detectors and low divergence beams, one should be able to separate out the various lattices it is not really relevant to the main issue - getting the best from existing data. The point I made about "correcting" data probably comes in a similar category - taking the opportunity to air a favourite subject.
Regards
Colin
PS. While here though I realise one of my points was a bit unclear. Point 5 should be
"5. My view is that for data in the PDB the same release rules should apply for the images as for the other data. For data not (yet) in the PDB, the funders of the research might want to define release rules. However, we can make suggestions!"
The original had "For other data" rather than "For data not (yet) in the PDB"
----------
From: James Holton
In the spirit of supporting a "can do" attitude, I have decided to try and frame the binary "images or no images" question as a gradual scale. Below is a list of ways to represent crystallographic data, with increasing amounts of "information" as you move down. That is, making validation more robust and allowing more and more yet-to-be-developed technologies to be applied, but also requiring higher costs, such as validation effort.
a) depositing coordinates only
b) coordinates and structure factors
c) coordinates, structure factors, and their sigmas (there was a time when we didn't do this!)
d) scaled and merged intensities (before "truncate" or sqrt?)
e) scaled and "unmerged" intensities with combined partials (future absorption corrections)
f) scaled spot intensities with partials separate (correct the shutter jitter someday?)
g) unscaled individual spot intensities (with geometry for calculating Lorentz, polarization, etc corrections)
h) spot intensities with a separate column for the "local background level" (this would be an efficient way to "compress" the images)
i) spots, local background, and background levels partway between spots (for reconstructing diffuse scatter)
j) all pixels from spot areas (~1000x compression over normal "corrected" images)
k) spot pixels plus lossy compression of background (1000-30x over corrected images)
l) losslessly compressed images (~2x over corrected images)
m) "corrected" images: all pixels from spatially-corrected dark-and-flood applied images
n) uncorrected images with dark, flood and bad-pixel map for relevant detector
o) raw output from detector's ADC or counter (no idea how to capture this...)
p) cryo-preserved data crystal (for verification of cell, or just chemical forensics, such as verifying the identity of the protein)
q) cryo-preserved duplicate crystal, to be held in a vault until you are accused of fraud.
r) sample of pure protein with crystallization conditions
s) protein sequence, let the PDB "verify" the experiment by repeating it, knowing that it "can be solved".
Now, I think it is clear that both the "benefit to the community" as well as the burden on PDB resources increase as we move down this list. In such situations one looks for "inflection points" where the next, small increase in benefit requires a disproportionately large cost. Historically, going from a to b was such an inflection point. This was back when the compact disc was a new thing, and the whole PDB could fit on one! In fact, if you had a "multi-session" drive you could back up your hard drive onto the remaining space. Those days are gone.
Recently, I heard the PDB is seriously considering jumping to somewhere between "e" and "g" (unmerged data). Doing so, I think, will be an interesting exercise in file format "standardization" since every major data processing package treats partials and postrefinement differently. And this is perhaps the main reason why "the PDB" (aka Gerard K) are wary of the idea of going all the way to "m" (corrected images). Do we expect PDB staff to re-process our dataset as part of the "validation" procedure?
Of course, if we are willing to relax the requirement of validation and curation, this could be a whole lot easier. In fact, there is already an image deposition infrastructure in place! It is called TARDIS:
http://tardis.edu.au/
Perhaps the best way forward would be for "the PDB" to introduce a new field for one or more TARDIS ids in a PDB deposition? It would be optional at the first, but no doubt required in the future.
-James Holton
MAD Scientist
3. Ideally the data to be deposited should include all stages e.g. raw images, "corrected" images, MIR/SAD/MAD images, unmerged integrated intensities, scaled, merged etc. Plus the metadata, software& versions used for the various stages. Worrying too much about all of this should not of course prevent a start being made. (An aside. I put the "corrected" in quotes because the raw images have fewer errors. The subsequent processing for detector distortions etc. depend on an imperfect model for the detector. I don't like the phrase data correction).
----------
From: Colin Nave
James
Thanks - your list is far more comprehensive than mine. I presume MAD/SAD/MIR data is implicit.
One doesn't need to progress through the list in this order (I am sure you are not suggesting this). One could usefully go to m without some of the intermediate stages.
Your comments about problems the PDB might have are useful. I have been in meetings where the PDB has been criticised for not following modern database standards and here we are trying to create more anarchy. In principle multiple formats would work so long as they are well defined but I can see why the PDB might object to this.
I should go through the documents from the Madrid meeting and see what is actually being proposed to get going. If a separate exercise independent of, but linked to, the PDB is set up for storing image data stage m) then that would be a good start.
Regards
Colin
----------
From: Alun Ashton
Sorry Matt, some large facilities do already keep all their raw and processed data. And I think the EU grant you mention to coordinate this is http://www.pan-data.eu soon to be odi, includes the ESRF :), don't your computing people tell you anything ?! :)
PanData have a meeting in early November and they (ahem we) are already in touch with the working group and will formalise that soon after the meeting.
----------
From: Ethan Merritt
As I understand it, TARDIS is just an indexing system.
You're still on your own to actually store the images.
The TARDIS setup will cough up the information that your images
are stored on a machine named pony.lbl.gov, or at least they were
at the time you registered them for indexing, where they supposedly
can be retrieved using tag #XYZ.
But so far as I know you would still be at the mercy of pony.lbl.gov
going up in flames, or being renamed twinkle.lbl.gov, or being
decommissioned when the next budget crunch hits.
For that matter, I don't know what the provision or expectation
is that anyone outside your institution could see or access the
machine holding the set of files that TARDIS told them were there.
If I've got this wrong, perhaps Ashley Buckle can chime in with
an update on TARDIS.
Ethan
----------
From: Martin M. Ripoll
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,
----------
From: Alun Ashton <Alun.Ashton@diamond.ac.uk>
Date: 27 October 2011 09:17
To: CCP4BB@jiscmail.ac.uk
Hi James,
1) thanks for sending and email in this thread longer than mine, I was "worried" I had killed it... ;)
2) you say:
And Ethan has just beat me to this point, from the Tardis web site, first paragraph last sentence:
"Storage was and remains federated, meaning the public index, TARDIS.edu.au contains no data itself and merely points to data stored in external labs and institutions."
So is the raw data even public? In that sense I think what were supposed to adopt in European facilities is ODI's which link accordingly to the facilities own suppository, and how that is developed and where you put it is down to regional preferences. It can obviously be large, small, publicly accessible, shared between friends or private. To be effective it will need to be around for a long time.
But to build on the success of model for the 'PDB' I would agree that someone should pay to have someone host this, and at least in the first instance make sure PDB structures have their raw data available and accessible to all. Tardis is in the right direction as a catalogue, ICAT in the EU is a simpler DB, ISPyB would need a bit of work to get data sharable through the interfaces, but I think it is a richer data structure.
Standardisation of the data would also be good, its amazing what a simple word checker can do to your emails these days....
BTW we pay the equivalent to about a weeks beamtime on one beamline for 200TB a year data storage and access to it for the whole facility, and someone else to take the pain of hosting it....
----------
From: David Waterman
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
-- David
----------
From: Martin M. Ripoll
Dear David,
As you probably have read in my previous message to Colin, I also agreed with him.
I just wanted to clarify (and obviously not to him) what George said. It was just a comment and not a lesson! Sorry if my message was misunderstood!
----------
From: Francis E Reyes
This discussion of image deposition and archival has certainly been illuminating. While there have been clear directions on how the process would actually work, I am becoming increasingly curious on why it should be done (outside of threatening non publication, social acceptance, funding, etc). I think a huge challenge is to convince the users that filling out the metadata with as much detail as possible is a worthwhile endeavor.
Specifically, are there documented cases where reinterpretation of an MX dataset (the raw images) has resulted in new biological insight into the system being modeled?
If possible I'd like to compile a list (off list) for my own education (though I can certainly report my results to the BB if the interest is there).
(Ok I have a selfish reason as well as I will be considering the reanalysis of some really old datasets that proved troublesome for a colleague and am looking for small hints of wisdom)
Some that come to mind for me:
Meyer et al. Structure of the 12-subunit RNA polymerase II refined with the aid of anomalous diffraction data. J Biol Chem (2009) vol. 284 (19) pp. 12933-9
Wang. Inclusion of weak high-resolution X-ray data for improvement of a group II intron structure. Acta Crystallogr D Biol Crystallogr (2010) vol. 66 (Pt 9) pp. 988-1000
F
---------------------------------------------
Francis E. Reyes M.Sc.
215 UCB
University of Colorado at Boulder
----------
From: George M. Sheldrick
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
> Consejo Superior de Investigaciones Cient ficas
> Consejo Superior de Investigaciones Cient ficas
> Enviado el: mi rcoles, 26 de octubre de 2011 11:52
----------
From: Ed Pozharski
Someone should be able to confirm this, but I was under impression that
at synchrotrons the acceleration/decelaration occurs outside
phi_start-phi_end range to allow for constant speed when the shutter
opens.
--
"I'd jump in myself, if I weren't so good at whistling."
Julian, King of Lemurs
----------
From: Ashley Buckle
A few responses regarding TARDIS:
While it's true that TARDIS.edu.au is just an index to files hosted elsewhere, a new solution, MyTARDIS has been developed as a repository that holds information on diffraction (and other kinds of) data. The idea being that facilities, labs or institutions take responsibility by hosting their own instance of MyTARDIS on their own storage with the eventual view to have entries released publicly and indexed by TARDIS.edu.au.
An example of this is the MyTARDIS experiment released here:
Also accessible in the tardis.edu.au entry here:
Both point to the exact same file location.
We also have a mechanism in which MyTARDIS has been placed at the Australian Synchrotron and linked to beamlines there to automatically store diffraction data locally. This data is then routed to MyTARDIS at a user's local university, where applicable. From here the workflow to publication takes similar form as that shown above.
It's open source licensed and deployable today:
With this as the installation documentation:
And here is how new instrument data can reach MyTARDIS:
This function has seen deployments at the Australian neutron facility ANSTO and in Microscopy/Microanalysis labs, too.
In response to James Holton's suggestion of getting PDB to accept TARDIS id's, I spoke with Kim Henrick 4 years ago if it would be possible simply to put a TARDIS id/url/handle in a REMARK, so the coordinates could be associated with the raw data that produced them. Kim was very positive that this should happen, and told me that it would be discussed in a wwPDB advisory board meeting, but I never heard anything back. In the meantime we have managed to get TARDIS URL's in several papers, usually in supplementary information. The editors/publishers didn't really object but didnt really take much notice I think either!
TARDIS development is an active area of my group and now that we are contributing to the IUCr efforts (thanks to John Helliwell), I'm optimistic that things are moving forward. We discussed some of the points raised here in the Acta Cryst TARDIS paper a couple of years ago (http://scripts.iucr.org/cgi-bin/paper?S0907444908015540) particularly the usefulness of raw data availability for methods development, so I'm really happy to see continued active discussion in this area!
Cheers
Ashley Buckle
Steve Androulakis
Associate Professor Ashley M Buckle
NHMRC Senior Research Fellow
The Department of Biochemistry and Molecular Biology,
Faculty of Medicine
Building 77
Monash University, Clayton, Vic 3800
Australia
No comments:
Post a Comment