ComparisonProcessing202504

This page is work in progress - please update from time to time.

(see also previous investigations 1, 2 and 3)

• 23rd April, 2025: first version
• 29th April, 2025: added Wilson plot comparisons
• 4th May, 2025: added section about latest binary (showing good behaviour of that 20250430 version)

Content:

• Introduction
• Quick initial overview
• Intensity bias
• Wilson plot comparisons
• XDS version 20250430

---

Introduction

23rd April 2025

A new XDS binary (20250409) is available: we are not sure when this was posted to replace the previous version. According to the release notes it contains a "Simplification of the background fitting routine of the INTEGRATE step."

2nd May 2025

A new XDS binary (20250430) is available: "Reduction of a small bias in the intensity estimate of weak reflections in the INTEGRATE step".

We currently recommend using the 20230630, 20250327 or 20250430 XDS binary.

---

Quick initial overview

• A quick initial check strongly suggests that this version (again) introduces intensity bias: we get significantly lower Wilson B values across the board in the tests run so far.

• For the same set of 60 test-cases, we get again 7 datasets with a warning about twinning (from POINTLESS) compared to only one for the 20250327 version. A comparison of different versions:

 Version  #twin      Comment
---------------------------------------------
20230630     1   <<< considered good
20240712    19
20240723    23
20241002     9
20250119     7
20250224     7
20250320    12
20250327     1   <<< considered good
20250409     7

• Unfortunately, neither the 20230630 nor the 20250327 binaries are available any longer: we suggest using those versions for the moment if at all possible!

---

Intensity bias

Following the same procedure as before, we can see the intensity bias rearing its ugly head again for this 20250409 version (last column):

Res   I(simu)  I(2023)  I(2024)  I(2025/01)  I(2025/02)  I(20250320)  I(20250327)  I(20250409)
-----------------------------------------------------------------------------------------------
2.93  20439.8  21535.1  21640.4   21577.4      21944.8     21965.7      21887.9      21875.0
2.56   5257.5   5532.1   5587.2    5545.6       5676.1      5690.6       5629.8       5632.2
2.33   2056.9   2157.9   2196.5    2188.2       2238.4      2248.9       2199.6       2224.1
2.16   1007.1   1056.5   1088.1    1082.7       1109.8      1119.3       1074.0       1103.9
2.03    467.1    481.3    507.3     504.7        519.8       528.3        488.2        519.7
1.93    230.6    230.9    250.6     249.1        255.7       263.6        231.2        258.4
1.85    109.5    107.2    118.8     117.0        119.4       127.2        106.0        124.4
1.77     57.9     56.3     62.7      60.0         61.3        69.2         55.6         67.2
1.71     28.7     27.5     30.7      26.9         26.9        34.8         27.0         33.9
1.66     17.4     17.0     19.2      14.2         14.1        22.5         16.9         21.9
1.61     10.6     10.3     11.7       6.0          5.6        14.4         10.1         14.1
1.57      6.6      6.5      7.7       1.6          0.9         9.8          6.4          9.8
1.53      4.3      4.2      5.2      -1.5         -2.3         6.9          4.1          7.3
1.50      2.8      2.6      3.6      -3.5         -4.4         5.0          2.6          5.6
1.47      1.7      1.7      3.0      -4.2         -5.1         4.4          1.7          4.7
1.44      1.2      1.0      1.9      -5.5         -6.6         3.3          0.9          3.9
1.41      0.8      0.7      1.6      -6.3         -7.3         3.1          0.6          3.7
1.38      0.6      0.6      1.7      -5.9         -7.0         3.5          0.5          3.7
1.36      0.5      1.0      1.5      -6.7         -8.2         3.4          0.9          4.6

---

Wilson plot comparisons

For the same set of examples, the Wilson plots (using different XDS versions) are given below (using the XDS/CORRECT -> XDSCONV -> CTRUNCATE path): look for the black line (20250409) compared to the blue (20230630) and red (20250327) one.

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
8EGN		8EPS
8EW7		8FG7
8FT8		8GCA
8K4Q		8PQC
8R5Q		8RCC
8SDW		8SHR
8SIO		8SO5
8TCA		8UFN
8VEV		8WT3
9CPL

• some Wilson plots are fairly similar for most XDS versions
• several Wilson plots show a significant differences between the 20250327 and 20250409 version: a curvature in the (expected to be linear) slope with high-resolution data going up (e.g. 5FBO, 5ONZ, 5SP6, 6DEX, 6NQY, 6P8P, 6P8U, 6R16, 6TPI, 6UCA, 7AO5, 7DK1, 7KDS, 7MJB, 7S87, 7SY9, 7TM9, 7UV5, 7WDA, 7Z1V, 7Z1Y, 8B1N, 8EW7, 8FT8, 8GCA, 8K4Q, 8PQC, 8R5Q, 8RCC, 8SHR, 8SIO, 8SO5, 8TCA, 8UFN, 8VEV, 9CPL)
• This should make it clear why we recommend the 20230630 (or the 20250327) XDS binary for data processing at the moment.
• You might want to check which XDS version is currently being used for automatic data processing at the synchrotron/beamline of your choice.
• You might also consider re-processing important datasets that might be suboptimal when processed with some other version - even if only to confirm that there is/was no obvious potential problem.

---

XDS version 20250430

Repeating the same kind of analysis for the latest XDS binary (20250430): comparing the black line (20250430) to the blue (20230630) and red (20250327) one shows a high level of similarity - while the grey line (previous 20250409 binary) does not.

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
8EGN		8EPS
8EW7		8FG7
8FT8		8GCA
8K4Q		8PQC
8R5Q		8RCC
8SDW		8SHR
8SIO		8SO5
8TCA		8UFN
8VEV		8WT3
9CPL

This binary removes the intensity bias seen in the 20240409 version and behaves again very similar to the 20230630 version (as did/does the 20250327 one): the Wilson plots show the expected, nearly linear decay at higher resolution - while other versions often (not in every case!) show un-physical behaviour with distinct (upwards) curvatures. In our eyes there is no use in looking at any other (internal) data quality metrics (like <I/sig(I)> or CC1/2) when the basics of the Wilson plot are already pointing to some visible shortcomings: one can convince oneself of anything if unconsciously looking desperately for a smaller number ;-)

At the moment, those are the three binaries we recommend - although it still seems as if the 20230630 binary behaves slightly better when it comes to handling reflections close to module gaps or shadowed areas compared to 20250327 and 20250430.

---

Page started 23rd April 2025, 11:00:26 UTC