Bijvoet Pair analysis on Twinned crystals

[AmyS]

Hello again!

I've been using Olex2 to refine a twinned dataset of a chiral compound. I'm curious what exactly is going into the Bijvoet Pair analysis calculation that runs at the end of each refinement.

If I use "detwinned" data in an HKLF4 format file I get this output:

Checking absolute structure...
Bijvoet pair analysis using Gaussian distribution
Bijvoet pairs (all): 7356
Bijvoet pairs (used): 7339
Bijvoet pairs coverage: 0.89
G: 1.020(13)
P2(true): 1.000
P3(true): 1.000
P3(false): 0.000
P3(racemic twin): 0.000
Hooft y: -0.010(6)
Flack x: -0.001(11)
OK

If I use an HKLF5 format file for refinement, I get this:

Checking absolute structure...
Bijvoet pair analysis using Gaussian distribution
Bijvoet pairs (all): 7980
Bijvoet pairs (used): 7919
Bijvoet pairs coverage: 0.96
G: 0.55(2)
P2(true): 1.000
P3(true): 1.000
P3(false): 0.000
P3(racemic twin): 0.000
Hooft y: 0.225(10)
Flack x: -0.003(9)
OK

If I run the analysis through Platon on the HKLF5 refined data I get:

Bayesian Statistics
Type ....... Gaussian
Select Pairs 13648
P2(true).... 1.000
P3(true).... 1.000
P3(rac-twin) 0.0E+00
P3(false) .. 0.0E+00
G .......... 0.9328
G (su) ..... 0.0128
Hooft y .... 0.034
Hooft (su) . 0.006

In the second example, I'm not sure why the Hooft parameter is so much worse than the Flack parameter (I'm not sure I've ever seen it be so much worse), and also I'm not sure why in the third example the Platon results are so different (though I do note there is a different number of Pairs used).

I think I can at least say that I have correctly assigned the stereochemistry of my compound, but I'm at a loss for which set of statistics to report and why.

Thanks as always!

PS> Should I re-post this in some other thread? Part of my question though is what exactly Olex2 is doing that generates a difference from Platon.

[johnewarren]

I think the analysis comes from CCTBX (I am probably wrong) so it may fall outside of the Olex2 remit but since R is a CCTBX person now I'm sure if he reads this he will have an answer?

PS If you find your topic drifting down the posts and it is a tadd old with still no reply you may see a bump button at the bottom of your post you can re-bumped it back up to the top.

[rjgildea]

Currently the Bijvoet pairs analysis in Olex2 doesn't work with HKLF5 twins and the answer you are given is almost certainly incorrect (presumably it should just refuse to calculate anything when you have HKLF5 format).

I am not sure what the correct approach is here for HKLF5 twins, and how meaningful an answer would be (I don't know if there is any physical reason that each twin component should have the same twin fraction). Perhaps Ton Spek could comment further on how his implementation in PLATON treats HKLF5 twins.