Based on this, a vision for the future:



A warehouse filled with sealed-tube, top-hat-profiled sources, super-accurate goniostats, and Dectris detectors, a robot running back and forth from a central dewar to place the crystals, all images 1-bit, intensities measured as probabilities; a day when crystal-frying will finally come to an end, or will be used routinely (as RIP) as the sure-fire way to solve crystal structures.



JPK







-----Original Message-----

From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of James Holton

Sent: Thursday, July 20, 2017 2:07 PM

To: [log in to unmask]

Subject: Re: [ccp4bb] Fine Phi Slicing



An important aspect of fine phi slicing that has not been mentioned yet (and took me a long time to figure out) is the impact of read-out time.  

Traditionally, read-out time is simply a delay that makes the overall data collection take longer, but with so-called "shutterless" data collection the read-out time can have a surprising impact on data quality.  It's 2 ms on my Pilatus3 S 6M.  This doesn't sound like much, and indeed 2 ms is also the timing jitter of my x-ray shutter, which had not been a problem with CCD detectors for 15 years.  The difference is that with so-called "shutterless" data collection not only can appreciable intensity fall into this 2 ms hole, but none of the data processing programs have a way to "correct" for it.  What you end up with is Rmerge/Rmeas values of 15-30% in the lowest-angle bin, and correspondingly low overall I/sigma.  At first,  I couldn't even solve lysozyme by S-SAD!  This had been an easy task with the Q315r I had just replaced.  The difference turned out to be "noise" coming from this read-out gap.



The 2 ms gap between images is only important if it is comparable to the time it takes a relp to transit the Ewald sphere.  At 1 deg/s and mosaic spread of 0.5 deg this is 0.002 deg of missing data, or about 1% error in integrated intensity.  This is fine for most applications.  But if you are turning at 25 deg/s with a room-temperature crystal of mosaicity

0.05 deg, then you could loose the spot entirely in a 2 ms read-out gap (100% error).  This is one of several arguments for fine phi slicing, where you make sure that every spot is not just observed, but split over

2-3 images.  This also helps the pile-up correction Gerd already mentioned.  What is often overlooked, however, is that the error due to read-out gap is only relevant to partials.  Fulls don't experience it at all, so wide phi slicing is practically immune to it.  But with fine phi slicing everything is a partial, and 100% of the spots are going to take on read-out-gap error. So, what is the solution?  Slow down.



The problem with slowing down the spindle, of course, is radiation damage.  If you've got a flux of 1e12 photons/s into a 100x100 micron beam spot and 1 A wavelength you are dosing metal-free protein crystals at about 50 kGy/s.  Most room-temperature crystals can endure no more than 200 kGy, so they will live for about 4 seconds in this beam.  A detector framing at 25 Hz will only get 100 images, no matter what the spindle speed. The decision then: is it better to get 100 deg at 1 

deg/image?  or 2.5 deg with fine phi slicing?   That is, if the mosaic 

spread is 0.05 deg, you can do no more than 0.025 deg/image and still barely qualify as "fine phi slicing". The 25 Hz framing rate dictates no less than 40 ms exposures, and that means turning the spindle at (0.025 deg/ 0.04 s) = 0.625 deg/s. Thus, we cover 2.5 deg in the 4 seconds before the crystal dies.  That's just algebra.  The pragmatic consequence is the difference between getting a complete dataset from one crystal and needing to merge 40 crystals.



Of course, you can attenuate 40x and get 100 fine-sliced degrees, but that will take 40x more beam time.  The images will also be 40x weaker.  

In my experience you need at least an average of 1 photon/pixel/image before even the best data processing algorithms start to fall over.  You can actually calculate photons/pixel/image beforehand if you know your flux and how thick your sample is:



photons/pixel = 1.2e-5*flux*exposure*thickness/pixels



where flux is in photons/s, exposure in seconds, thickness in microns and 1.2e-5 comes from the NIST elastic scattering cross section of light atoms (C, N, O are all ~ 0.2 cm^2/g), the rough density of protein crystals (1.2 g/cm^3), and the fact that about half of all scattered photons land on a flat detector at typical distance from the sample, or: 

0.2*1.2*(1e-4 cm/um)/2 = 1.155e-5



So, if your flux is 1e12 photons/s and your sample is 100 um thick you will get ~1 photon/pixel on a 6M in about 5 ms.  That corresponds to a framing rate of 200 Hz.  If the detector can't go that fast, you need to attenuate.  Note this is the total sample thickness, including the stuff around the crystal.  Air scatter counts as 1 micron of sample thickness for every mm of air between the beamstop and collimator.  So, in a way, the beam properties and detector properties can be matched.  What is counter-intuitive is that a 25 Hz detector is sub-optimal for a 100 micron beam with flux 1e12 photons/s.  Certain fluxaholics would already call that a "weak" beam, so why does getting a faster detector mean you should attenuate it?



It would seem that fine slicing requires throwing away a lot of beam, unless you only want a few degrees from every crystal. Maybe there is a better way?



I have experimented with 1-deg images with no attenuation and room temperature crystals and the results are surprisingly good. Median real-world Rmeas value are 5%.  Not beautiful, but more than adequate for molecular replacement.  I expect what is going on is the pile-up issues from wide slicing are being compensated by the "instant re-trigger" feature of Pilatus3.  This is basically a pile-up correction that does not depend on even illumination over an entire image.  As for the accumulation of "extra background" in the wide slices, in the weak-image limit the background level is already close to zero, so it doesn't add up very fast over 1 deg.  That is, you don't get more than 0 or 1 photon/pixel of background anyway, even with 1 deg images.



So, it would seem that the realities of crystal lifetimes in modern beams can make fine slicing at full beam difficult to the point of being inadvisable.  Everyone's goals and situation are different, but I do advise everyone to bear in mind the significance of the read-out gap noise when deciding on an exposure time and spindle rotation rate.  You should either fine-slice properly, or not at all.  And you should certainly point this out when writing a grant to get an Eiger, which has a read-out time of 3 microseconds.



-James Holton

MAD Scientist



On 7/13/2017 3:02 PM, Gerd Rosenbaum wrote:

> Hi Fred,

>

> fine slicing does not alleviate the count RATE limitation of photon 

> counting detectors because fine slicing does not reduce the 

> instantaneous photon flux on the detector when you cross the 

> diffraction maximum. Fine slicing does help if you push the maximum 

> counts per pixel per frame to the counter register limit. On 

> integrating detectors, like CCDs, there is practically no count RATE 

> limit. They do have a charge (~photons) per pixel per frame limit, as 

> well, which is mostly much lower than for the photon counting 

> detectors - about 1/10 even after taking into account the diffraction 

> spot covers many more pixels.

>

> Different from integrating detectors where you only have to watch the 

> overflow, for counting detectors you have to watch for exceeding 

> either the count rate limit or the total count limit. The former is 

> not an easy task because you will see only the count per exposure.

> Divide that by the exposure time you get the AVERAGE rate, not the 

> peak rate.

>

> The count rate limit, very short exposure time (using high flux) and 

> the 1-pixel point spread function work against each other. Exposure 

> time = 0.01 s, count rate limit 1e6 /sec (Pilatus2), 1-2 pixel per 

> spot => 10-20k counts per spot maximum for the strongest peak.

>

> Dectris has come up with an ingenious hardware and software in the

> Pilatus3 pushing the rate for reasonable dead time correction to over

> 1e7 counts/sec so even with 10 ms exposures weak reflections can be 

> well recorded besides strong reflections.

>

> Gerd

>

> On 13.07.2017 15:34, Dyda wrote:

>> I could be wrong here, but isn't the case that fine slicing is an 

>> option with a CCD and a necessity on a PAD b/c of dead time and/or 

>> counter dynamic range issues?

>>

>> (no current and/or former financial ties to any manufacturer)

>>

>> Fred

>> ****************************************************************

>> ***************

>>

>> Fred Dyda, Ph.D.                       Phone:301-402-4496

>> Laboratory of Molecular Biology        Fax: 301-496-0201

>> DHHS/NIH/NIDDK                         e-mail:[log in to unmask]

>> Bldg. 5. Room 303

>> Bethesda, MD 20892-0560      URGENT message e-mail: 

>> [log in to unmask]

>> Google maps coords: 39.000597, -77.102102 

>> http://www2.niddk.nih.gov/NIDDKLabs/IntramuralFaculty/DydaFred

>> *********************************************************************

>> **********

>>