This page is work-in-progress and frequently updated: please force a refresh in your browser to ensure you are seeing the latest version.

 

We highly recommend staying with the 20230630 version for now

 

Content:

Update:

  • 20240707: Added detailed analysis for the latest 20241002 version (including STARANISO plots and timings). See especially the twinning and timing analysis.
  • 20241002: We just seen that a new set of XDS binaries is on the XDS website that we have not tested yet (stay tuned). We recommend that users take advantage of the provided download for the old XDS version (Linux-only) on the same page for the time being if running in production mode. Unfortunately, MacOS users have only access to the newest version.

Introduction

This is principally a comparison of XDS versions "June 30, 2023" versus "June 30, 2024" done during late July to mid September 2024 (with about 4 weeks' interruption in between due to vacations).

  • See also XDS Release Notes
  • The XDS developers (Wolfgang Kabsch) kindly provided us with a XDS binary of the old version with an extended runtime for those tests.
  • Originally, we refered to the two versions used (on Linux) as 20230630 ("old") and 20240723 ("new") throughout the text and in plots.
  • Since a new version has been released on 2nd October 2024, we add this to our comparison and refer to it as 20241002.

Please note that we tried to stick to a "lowest common denominator" in terms of automation and expertise in this comparison: this is not intended to compare different packages that internally run XDS, so a lot of the advanced features (e.g. in our autoPROC package) are ignored. This means that one can reach better results when taking advantage of those features or maybe after manually fine-tuning data processing on a case-by-case basis as an expert. We wanted to keep it simple and show what happens when a user runs software in all-default mode (as this is also likely to happen on automatic data processing results triggered directly at a synchrotron beamline).


Examples

Starting from the PDB release as of 20240919, we can get various details about a particular PDB entry as present in the archive:

  • data collection date
  • initial deposition date
  • initial release date
  • detector name
  • beamline/synchrotron name
  • a raw diffraction data DOI (and does it point to a meaningful repository like proteindiffraction.org, SBGRID, Zenodo, MX-rdr, Fairdata.fi or XRDA (there are others with fewer entries or incorrect/unusable URL)

After sorting those by data collection date - and restricting ourselves to Pilatus/Eiger datasets only - we can pick the latest PDB entry for a given detector/beamline/synchrotron combo. This results in 176 examples:

    4FQN 4MTK 4QKI 4TVT 4TWN 4XGU 5AUI 5BXG 5CYZ 5E9I 5EO9 5FBO 5FD7
    5IUK 5K7M 5LP9 5LXW 5NHU 5NKT 5OD9 5ONZ 5RG0 5SP3 5SP6 5VZR 6BGA
    6BLH 6BLI 6CK7 6CW0 6DEX 6E80 6EXI 6IU9 6JGJ 6MOE 6NCR 6NKQ 6NQY
    6NW4 6O2H 6O6N 6OWX 6OWZ 6P8P 6P8S 6P8U 6R16 6SCX 6TPI 6UCA 6VZQ
    6VZW 6W4B 6WAY 6X3O 6YJP 6Z5F 6Z9G 6ZJ6 7ALO 7AO5 7AR4 7AV6 7BBP
    7BWH 7CLJ 7DF2 7DK1 7KCN 7KDS 7KMJ 7MJB 7N2P 7OTH 7OVP 7PD5 7PHO
    7POX 7PWM 7PWP 7PWU 7PWW 7QGF 7QOQ 7QT8 7R0K 7R3W 7R59 7S87 7SGW
    7SY9 7TAL 7TBO 7TCD 7THB 7TM9 7TPB 7UDI 7UV5 7WDA 7WEZ 7XLI 7XRC
    7YZX 7Z1V 7Z1Y 7Z41 7Z7L 7Z8Y 7Z8Z 8A19 8A9D 8AGQ 8AQ8 8B1N 8BFY
    8BXT 8CX4 8DA3 8DU7 8E5V 8E60 8EGN 8EPS 8EW7 8F8E 8FFE 8FG7 8FRF
    8FT8 8GCA 8GI4 8GM6 8HKR 8JYJ 8K4Q 8OIC 8PIE 8PQA 8PQC 8PQD 8Q90
    8QK1 8R2G 8R5Q 8R5R 8RCA 8RCB 8RCC 8S1R 8SDW 8SHR 8SIO 8SLU 8SO5
    8SXS 8TCA 8TYZ 8U09 8U0I 8U1E 8UFN 8UFO 8UM6 8V2T 8VEV 8VM2 8W6K
    8WT3 8YEQ 9B7F 9CPL 9CRW 9D8S 9EWK

For 89 of those we already had the data locally on disk and had run them at some point (over the last 10+ years) as part of our testing:

    4FQN 4QKI 5AUI 5E9I 5FBO 5LP9 5NHU 5OD9 5ONZ 5RG0 5SP3 5SP6 5VZR
    6BLI 6CK7 6CW0 6DEX 6NQY 6P8P 6P8U 6R16 6SCX 6TPI 6UCA 6VZQ 6VZW
    6ZJ6 7AO5 7AR4 7AV6 7BWH 7DK1 7KDS 7KMJ 7MJB 7PWP 7PWW 7QGF 7R0K
    7R59 7S87 7SY9 7TBO 7THB 7TM9 7UV5 7WDA 7XRC 7YZX 7Z1V 7Z1Y 7Z41
    7Z8Y 8AGQ 8AQ8 8B1N 8BXT 8CX4 8DA3 8E5V 8E60 8EGN 8EPS 8EW7 8F8E
    8FG7 8FT8 8GCA 8K4Q 8PQC 8PQD 8Q90 8R5Q 8R5R 8RCB 8RCC 8SDW 8SHR
    8SIO 8SLU 8SO5 8SXS 8TCA 8UFN 8UFO 8VEV 8WT3 9CPL 9D8S

Taking the first 60 of those (ordered by fastest runtime - to allow for as many tests as possible) gives us a final list of

    4FQN 4QKI 5AUI 5E9I 5FBO 5OD9 5ONZ 5RG0 5SP6 5VZR 6BLI 6CK7 6CW0
    6DEX 6NQY 6P8P 6P8U 6R16 6TPI 6UCA 6VZQ 6VZW 7AO5 7DK1 7KDS 7MJB
    7S87 7SY9 7TM9 7UV5 7WDA 7Z1V 7Z1Y 7Z41 8AGQ 8B1N 8BXT 8DA3 8E5V
    8E60 8EGN 8EPS 8EW7 8FG7 8FT8 8GCA 8K4Q 8PQC 8R5Q 8RCC 8SDW 8SHR
    8SIO 8SLU 8SO5 8TCA 8UFN 8VEV 8WT3 9CPL

This list contains entries originally (according to PDB entry) processed by XDS (32), Aimless (22), XSCALE (14), autoPROC (7), HKL-3000 (7), MOSFLM (5), DIALS (4), xia2 (3), SCALA (3), HKL-2000 (3), iMOSFLM (1), STARANISO (1) and SCALEPACK (1).


Computational details

      # set specific "xds_par" binary on command-line:
      process -nthreads NTHREADS xds=/where/ever/binary symm=SPGR cell="a b c al be ga" -I /where/ever/images
  • fast_dp version 1.7 via DIALS
      # specific "xds_par" binary is expected first in $PATH:
      fast_dp -j 1 -k NTHREADS -J 1 -s "SPGR" -c a,b,c,al,be,ga -l /where/ever/plugin /where/ever/image-file

      # NOTE: we are unable to make the -l switch work here (it
      # doesn't seem to have any effect and we don't get a LIB=
      # line in XDS.INP ... which means that our attempts with HDF5
      # datasets fail while the *.cbf.gz datasets process through
      # the slower "temporary-decompressed-file" path)
      generate_XDS.INP /where/ever/image-file
      # add/replace SPACE_GROUP_NUMBER= and UNIT_CELL_CONSTANTS=
      # adjust MAXIMUM_NUMBER_OF_JOBS= and MAXIMUM_NUMBER_OF_PROCESSORS=
      
      # run specific binary:
      xds_par
  
      # if indexing fails, remove the "0 0 0" spots from SPOT.XDS and re-run again

We are also looking at using XDSME (latest change in Jan 2023), Xia2 and any other available command-line tools for data processing using XDS as the central indexing/integration engine.

As a simple check on some basic data quality metrics, we are using MRFANA from autoPROC on the final scaled and unmerged data provided by each package/run. This isotropic analysis uses the criterion of CC(1/2) above 0.3 in the outer shell (using equal-number binning):

    mrfana -cutoff_rpim 99.9 -cutoff_isigi 0.0 -cutoff_cchalf 0.3 -nref -1000 
 

Smoke test

Recording for each of the 60 examples if a job (see above) went through to the end, we test this via:

  • autoPROC: existence of report.pdf
  • plain XDS: existence of XDS_ASCII.HKL
  • fast_DP: existence of XDS_ASCII.HKL
program success 20230630 success 20240723 success 20241002
autoPROC 100% 97% 97%
plain_XDS 95% 97% 97%
fast_dp 87% 87% NA

Please note that the non-autoPROC systems might be run differently when an expert configures them for a specific instrument/beamline ... so your mileage might vary.


Overall statistics

Let's take two metrics as simple proxies for data quality (after determining the high-resolution limit based on outer-shell CC(1/2) > 0.3):

  • the high resolution limit
  • the overall <I/sigI> value

Please note that this completely ignores any anisotropy in the data, so is definitely sub-optimal. But we wanted to stick with the lowest common denominator between the various runs and packages.

We can then do a comparison for those two re-processing jobs between resolution (if there is a change of at least 10% in reciprocal-space volume, that change is considered significant) and <I/sigI> (at least 10% difference in value seems significant) for the various XDS versions:

20230630 versus 20240723:

program higher resolution 20230630 higher resolution 20240723 higher <I/sigI> 20230630 higher <I/sigI> 20240723
autoPROC 15% 15% 23% 10%
plain_XDS 43% 18% 55% 12%
fast_dp 28% 32% 38% 15%

20230630 versus 20241002:

program higher resolution 20230630 higher resolution 20241002 higher <I/sigI> 20230630 higher <I/sigI> 20241002
autoPROC 0% 3% 3% 43%
plain_XDS 2% 12% 7% 7%
fast_dp NA NA NA NA

Notes:

  • percentages are computed over the 60 examples mentioned above
  • the 20241002 version behaves much more similar to the old 2023 one - with one exception: when used within autoPROC it seems to produce higher <I/sigI> values

Specific comparisons

The most interesting examples are those where one of the two investigated XDS versions gives significantly different statistics (resolution and/or <I/sigI>) in any of the packages tested. We only look at those examples where the compared packages ran successfully with both XDS versions.

Warning about twinning (POINTLESS)

Feeding the final scaled+unmerged results of each package into POINTLESS via one of the following commands

    pointless xdsin XDS_ASCII.HKL                 # autoPROC, plain XDS, fast_dp
    pointless xdsin aimless_alldata_unmerged.mtz  # autoPROC
    pointless xdsin fast_dp_unmerged.mtz          # fast_dp

and looking for a message in the standard output of POINTLESS

WARNING: the L-test suggests that the data may be twinned

gives us:

program reflection data (scaling) warnings 20230630 warnings 20240723 warnings 20241002
autoPROC XDS_ASCII.HKL (CORRECT) 2% 35% 17%
aimless_alldata_unmerged.mtz (AIMLESS) 2% 32% 12%
plain_XDS XDS_ASCII.HKL (CORRECT) 2% 55% 17%
fast_dp XDS_ASCII.HKL (CORRECT) 2% 52% NA
fast_dp_unmerged.mtz (CORRECT) 2% 52% NA

Notes:

  • percentages are computed over the 60 examples mentioned above
  • aimless_alldata_unmerged.mtz produced by autoPROC takes INTEGRATE.HKL and runs the actual scaling then with AIMLESS
  • fast_dp_unmerged.mtz takes XDS_ASCII.HKL (after scaling in CORRECT), converts this to MTZ in POINTLESS and then runs AIMLESS in analysis-only mode, i.e. no actual scaling in AIMLESS and only computation of data quality metrics (merging statistics)
    • this is often confused and subsequently reported (e.g. in PDB entries) as if AIMLESS was used for data scaling ... software usage statistics in the PDB archive are notoriously unreliable (the presence of a aimless.log file doesn't mean it was used for scaling: the content of it makes that very clear).

Please note that other programs (e.g. our own STARANISO) might come to slightly different conclusions about twinning. Again: we tried to keep it simple and to stick to widely used systems.

See also the discussion in e.g. Parkhurst et al, 2016.

STARANISO analysis

For consistent planes, the HKL plots are best suited since they only depend on the crystal symmetry. The PQR plots (planes defined by the principal axes of the fitted ellipsoid) will be affected by the underlying distribution of signal in reciprocal space and by the resulting cut-off surface (to which an ellipsoid is then fitted).

For autoPROC these are created automatically (after AIMLESS scaling of INTEGRATE.HKL), while for other packages we use the final, scaled+merged reflection data (using XDSCONV on the XDS_ASCII.HKL file to create an MTZ file if required).

PDB supposedly "better" autoPROC HKL plain_XDS HKL
5SP6 20240723 5SP6_autoPROC_staraniso_alldata.ismean-hkl.png 5SP6_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
6UCA 20230630 6UCA_autoPROC_staraniso_alldata.ismean-hkl.png 6UCA_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
7KDS 20240723 7KDS_autoPROC_staraniso_alldata.ismean-hkl.png 7KDS_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8EGN 20240723 8EGN_autoPROC_staraniso_alldata.ismean-hkl.png 8EGN_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8GCA 20230630 NA 8GCA_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8FG7 20240723 8FG7_autoPROC_staraniso_alldata.ismean-hkl.png 8FG7_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8K4Q 20230630 8K4Q_autoPROC_staraniso_alldata.ismean-hkl.png 8K4Q_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8TCA 20230630 NA 8TCA_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8UFN 20230630 8UFN_autoPROC_staraniso_alldata.ismean-hkl.png 8UFN_plain_XDS_staraniso_xds_alldata.ismean-hkl.png
8VEV 20240723 8VEV_autoPROC_staraniso_alldata.ismean-hkl.png NA
8WT3 20230630 8WT3_autoPROC_staraniso_alldata.ismean-hkl.png 8WT3_plain_XDS_staraniso_xds_alldata.ismean-hkl.png

Notes:

  • The latest 20241002 version behaves much more similar to the old 20230630 version.
  • The criterion for selecting supposedly "better" processing had to be relaxed for the 20240723 version compared to the 20230630 one (since otherwise there would have been no hits): we wanted a fairly equal number of examples showing differences in both directions.
    • This description has to be taken with a pinch of salt, since it is only based on the isotropic diffraction limit (judged by CC(1/2)>0.3 in the outer shell) and the overall <I/sigI> value.
    • However, this procedure has the advantage of being simple, consistent and using meaningful and well-established metrics.
  • fast_dp already applies a spherical resolution cut early on, so no scaled+merged data without such a cutoff is available here.
  • 8GCA and 8TCA can't easily be compared since the autoPROC runs using the NEW version automatically excluded some poor image ranges (and therefore comparison is not relevant or would not be fair).

Outlier classification in XDS

We can stay even closer to the raw, integrated intensities coming out of INTEGRATE by looking at the scaling and outlier analysis of the resulting INTEGRATE.HKL file (this is where the "new method for estimating the background in each image pixel" will have the biggest impact).

CORRECT will apply several correction factors ("MODULATION" across the detector surface, "DECAY" as a function of image number and resolution, and "ABSORPTION" as a function of image number and detector region). After these correction factors are applied, measurements are rejected if they deviate too much from their symmetry-equivalent reflections or if they don't follow expected Wilson distribution.

We can visualise those correction factors and outliers easily for the same set of examples (the latter is done automatically by autoPROC already):

Misfits

PDB autoPROC plain_XDS fast_dp
5SP6 5SP6_autoPROC_0_misfits.png 5SP6_plain_XDS_misfits.png 5SP6_fast_dp_misfits.png
6UCA 6UCA_autoPROC_0_misfits.png 6UCA_plain_XDS_misfits.png 6UCA_fast_dp_misfits.png
7KDS 7KDS_autoPROC_0_misfits.png 7KDS_plain_XDS_misfits.png 7KDS_fast_dp_misfits.png
8EGN 8EGN_autoPROC_0_misfits.png 8EGN_plain_XDS_misfits.png 8EGN_fast_dp_misfits.png
8GCA NA 8GCA_plain_XDS_misfits.png 8GCA_fast_dp_misfits.png
8FG7 8FG7_autoPROC_0_misfits.png 8FG7_plain_XDS_misfits.png 8FG7_fast_dp_misfits.png
8K4Q 8K4Q_autoPROC_0_misfits.png 8K4Q_plain_XDS_misfits.png NA
8TCA NA 8TCA_plain_XDS_misfits.png 8TCA_fast_dp_misfits.png
8UFN 8UFN_autoPROC_0_misfits.png 8UFN_plain_XDS_misfits.png 8UFN_fast_dp_misfits.png
8VEV 8VEV_autoPROC_0_misfits.png 8VEV_plain_XDS_misfits.png NA
8WT3 8WT3_autoPROC_0_misfits.png 8WT3_plain_XDS_misfits.png 8WT3_fast_dp_misfits.png

Assuming that nothing has changed in the way the outlier rejection is performed in CORRECT after applying those correction factors, there seem to be some kind of knock-on effects from differences in integration results with the 20240723 version. The pattern of outlier rejections in CORRECT was already rather puzzling in older versions (but without any impact on autoPROC, since it is using the raw integrated intensities from INTEGRATE.HKL directly in the scaling module aP_scale - partly for that reason), but that version gives rise to even more puzzling shapes.

The 20241002 version seems to have reverted back mostly to the old behaviour at this point.

Timing

Looking at the timings as reported by COLSPOT and INTEGRATE ("cpu time (sec)" in *.LP files) for all the above 60 examples (where the same set of images was finally used):

XDS v1 XDS v2 COLSPOT v1/v2 INTEGRATE v1/v2
20240723 20230630 0.84 0.89
20241002 20230630 0.80 1.84
20241002 20240723 0.97 2.17

Notes:

  • The COLSPOT step is about 20% faster in both 2024 versions compared to the 2023 one.
  • The better behaving integration in version 20241020 is nearly twice slower than the old 20230630 version in our hands (using either autoPROC or plain XDS).

User examples

This currently only compares the old 2023 version to the 2024 version(s) prior to the latest 20241002 release!

Some users were able to provide us with specific examples of recent data collections and processing results - thanks a lot for that!

Example A

Data was collected in multiple orientations using the Global Phasing workflow (and sometimes at multiple wavelengths). Here "OLD" refers to the 20230630 version and "NEW" to the 20240723 one.

Crystal Wavelength HKL PQR
01 1 PowerUser_GPhLWorkflowSession_xtal-01_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-01_wvl-1_staraniso_alldata.ismean-pqr.png
02 1 PowerUser_GPhLWorkflowSession_xtal-02_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-02_wvl-1_staraniso_alldata.ismean-pqr.png
03 1 PowerUser_GPhLWorkflowSession_xtal-03_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-03_wvl-1_staraniso_alldata.ismean-pqr.png
04 1 PowerUser_GPhLWorkflowSession_xtal-04_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-04_wvl-1_staraniso_alldata.ismean-pqr.png
05 1 PowerUser_GPhLWorkflowSession_xtal-05_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-05_wvl-1_staraniso_alldata.ismean-pqr.png
06 1 PowerUser_GPhLWorkflowSession_xtal-06_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-06_wvl-1_staraniso_alldata.ismean-pqr.png
2 PowerUser_GPhLWorkflowSession_xtal-06_wvl-2_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-06_wvl-2_staraniso_alldata.ismean-pqr.png
07 1 PowerUser_GPhLWorkflowSession_xtal-07_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-07_wvl-1_staraniso_alldata.ismean-pqr.png
08 1 PowerUser_GPhLWorkflowSession_xtal-08_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-08_wvl-1_staraniso_alldata.ismean-pqr.png
09 1 PowerUser_GPhLWorkflowSession_xtal-09_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-09_wvl-1_staraniso_alldata.ismean-pqr.png
10 1 PowerUser_GPhLWorkflowSession_xtal-10_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-10_wvl-1_staraniso_alldata.ismean-pqr.png
11 1 PowerUser_GPhLWorkflowSession_xtal-11_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-11_wvl-1_staraniso_alldata.ismean-pqr.png
12 1 PowerUser_GPhLWorkflowSession_xtal-12_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-12_wvl-1_staraniso_alldata.ismean-pqr.png
13 1 PowerUser_GPhLWorkflowSession_xtal-13_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-13_wvl-1_staraniso_alldata.ismean-pqr.png
14 1 PowerUser_GPhLWorkflowSession_xtal-14_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-14_wvl-1_staraniso_alldata.ismean-pqr.png
15 1 PowerUser_GPhLWorkflowSession_xtal-15_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-15_wvl-1_staraniso_alldata.ismean-pqr.png
16 1 PowerUser_GPhLWorkflowSession_xtal-16_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-16_wvl-1_staraniso_alldata.ismean-pqr.png
17 1 PowerUser_GPhLWorkflowSession_xtal-17_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-17_wvl-1_staraniso_alldata.ismean-pqr.png
18 1 PowerUser_GPhLWorkflowSession_xtal-18_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-18_wvl-1_staraniso_alldata.ismean-pqr.png
19 1 PowerUser_GPhLWorkflowSession_xtal-19_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-19_wvl-1_staraniso_alldata.ismean-pqr.png
20 1 PowerUser_GPhLWorkflowSession_xtal-20_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-20_wvl-1_staraniso_alldata.ismean-pqr.png
21 1 PowerUser_GPhLWorkflowSession_xtal-21_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-21_wvl-1_staraniso_alldata.ismean-pqr.png
22 1 PowerUser_GPhLWorkflowSession_xtal-22_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-22_wvl-1_staraniso_alldata.ismean-pqr.png
23 1 PowerUser_GPhLWorkflowSession_xtal-23_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-23_wvl-1_staraniso_alldata.ismean-pqr.png
24 1 PowerUser_GPhLWorkflowSession_xtal-24_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-24_wvl-1_staraniso_alldata.ismean-pqr.png
25 1 PowerUser_GPhLWorkflowSession_xtal-25_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-25_wvl-1_staraniso_alldata.ismean-pqr.png
2 PowerUser_GPhLWorkflowSession_xtal-25_wvl-2_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-25_wvl-2_staraniso_alldata.ismean-pqr.png
3 PowerUser_GPhLWorkflowSession_xtal-25_wvl-3_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-25_wvl-3_staraniso_alldata.ismean-pqr.png
27 1 PowerUser_GPhLWorkflowSession_xtal-27_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-27_wvl-1_staraniso_alldata.ismean-pqr.png
2 PowerUser_GPhLWorkflowSession_xtal-27_wvl-2_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-27_wvl-2_staraniso_alldata.ismean-pqr.png
3 PowerUser_GPhLWorkflowSession_xtal-27_wvl-3_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-27_wvl-3_staraniso_alldata.ismean-pqr.png
28 3 PowerUser_GPhLWorkflowSession_xtal-28_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-28_wvl-1_staraniso_alldata.ismean-pqr.png
29 1 PowerUser_GPhLWorkflowSession_xtal-29_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-29_wvl-1_staraniso_alldata.ismean-pqr.png
30 1 PowerUser_GPhLWorkflowSession_xtal-30_wvl-1_staraniso_alldata.ismean-hkl.png PowerUser_GPhLWorkflowSession_xtal-30_wvl-1_staraniso_alldata.ismean-pqr.png

Example B

A single-sweep datasets processed with an older (20220220) XDS version and the first new binary (20240712):

B_20220220-vs-20240712_staraniso_alldata.ismean-pqr.png