Dear Tim, thank you very much for your reply! I agree that it will be very difficult to get reasonable values for both, occupancy and ADPs, at usual neutron diffraction resolutions of around 2 A. However, we are talking about neutron diffraction data to 1.4 A and X-ray diffraction data to 0.9 A. So, we have a lot of data. I should have mentioned this right away, sorry. Do you think that in such a rare case it might be possible to get reasonable values? Also, I like your idea to group H/D sites by expected exchange values. However, from the high-quality data we have, it is obvious that in a lot of cases the exchange rate is not really easy to predict by chemical intuition alone. There are sites where you clearly would expect at least a partial exchange but yet you observe a strong negative peak in the neutron scattering length density indicating the presence of an almost non-exchanged hydrogen . Wouldn't it also be possible to fix the ADP ratio between the H/D atoms and the atom to which they are attached to a chemically likely value between 1.0 and 1.5 and then refine only the occupancy? To me auch an approach seems less prone to errors introduced into the model by the crystallographer's expectations. Best wishes, Johannes Am 13.10.2016 um 17:25 schrieb Tim Gruene:

Dear Johannes,

I would recommend not to refine the occupancy of individual H/D pairs, unless you have high resolution and other evidence for the resulting values. Occupancy and ADP-values are very strongly correlated, and the correlation becomes higher the worse your resolution. And neutron data often have resolution of 2A or worse.

You could group atoms together where you expect similar exchange ratios (based on chemical intuition) and refine one ratio per group. You can find a more detailed explanation in https://doi.org/10.1107/S1600576713027659

That article also recommends to treat X-ray data as additional information by means of geometry restraints to avoid the complications you get with joint refinement. The main effect of joint refinement are prettier maps <flame> and you may need to decide whether you want to focus on science or on art </flame>. This way we refined the structure also has the side effect to constrain H and D to the same location, although, admittedly, the coordinate difference in the structure you quote is probably negligible.

Best, Tim

On Thursday, October 13, 2016 05:06:36 PM Johannes Schiebel wrote:

...

Hi everyone,

I am currently working on an X-ray/neutron (XN) joint refinement using phenix.refine. As it should be, H/D coordinates and ADPs at exchangable sites are refined to equal values when using Phenix version 1.10.1-2155 and neutron data only. This is also stated in the paper describing the development of the method (Afonine et al. (2010) Joint X-ray and neutron refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 66, 1153-1163) as the default behavior: "Currently, phenix.refine maintains the H and D atoms at coinciding positions and constrains their ADPs to be equal to each other". However, when switching to XN-refinement using the same Phenix version, H and D atoms refine to different coordinates and ADPs, which should not be the case as it leads to unrealistic artifacts as can be seen from the deposited PDB 3X2P:

ATOM 186 H AALA A 13 3.486 -18.200 -14.123 0.38 14.54 H ATOM 187 D BALA A 13 3.488 -18.195 -14.111 0.62 38.07 D

In this example, the D-occupancy is likely overestimated, while the H-occupancy is underestimated because the ADPs refine to very different values, which is chemically not reasonable.

Hence my question: How can I treat my data in a way that ADPs and coordinates refine to the same values at exchangeable H/D sites also for XN-refinement? Is there a specific keyword I am currently overlooking or do I have to use another Phenix version?

I would be really glad to receive your feedback. Thanks in advance!

Kind regards, Johannes

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

Dear Leif, thanks for your feedback. The neutron data is to 90% complete. Yes, you are absolutely right. This is why in the first place I have refined the structure against neutron data only until convergence. It's just that we are very much interested in the water structure and here the addition of the X-ray data helps to orient waters correctly. The results agree astonishingly well with chemical intuition which is sometimes not the case when using only neutron data even though the resolution is comparably high. Waters simply have too many degrees of freedom and here the knowledge of the precise oxygen position helps a lot during modelling. Ultimately, we would like to deposit both, the pure neutron and the XN structure. Regards, Johannes Am 14.10.2016 um 15:11 schrieb Leif Hanson:

Dear Johannes, What is your completeness in your neutron data? I think at some point we need to stop hiding behind a joint refinement and use only the neutron data, especially with this level of resolution. I like your idea of fixing the ADP ratio. The more scientifically interesting question in this case is the exchange and not the B-factors. If you are really bothered by your X+N H/D coordinate and occupancy results, and your completeness is near 90% or higher, the more conservative result is to refine with SHELX. Best wishes, Leif Hanson

On Fri, Oct 14, 2016 at 5:45 AM, Johannes Schiebel mailto:[email protected]> wrote:

Dear Tim,

thank you very much for your reply! I agree that it will be very difficult to get reasonable values for both, occupancy and ADPs, at usual neutron diffraction resolutions of around 2 A. However, we are talking about neutron diffraction data to 1.4 A and X-ray diffraction data to 0.9 A. So, we have a lot of data. I should have mentioned this right away, sorry. Do you think that in such a rare case it might be possible to get reasonable values?

Also, I like your idea to group H/D sites by expected exchange values. However, from the high-quality data we have, it is obvious that in a lot of cases the exchange rate is not really easy to predict by chemical intuition alone. There are sites where you clearly would expect at least a partial exchange but yet you observe a strong negative peak in the neutron scattering length density indicating the presence of an almost non-exchanged hydrogen . Wouldn't it also be possible to fix the ADP ratio between the H/D atoms and the atom to which they are attached to a chemically likely value between 1.0 and 1.5 and then refine only the occupancy? To me auch an approach seems less prone to errors introduced into the model by the crystallographer's expectations.

Best wishes, Johannes

Am 13.10.2016 um 17:25 schrieb Tim Gruene:

...

Dear Johannes,

I would recommend not to refine the occupancy of individual H/D pairs, unless you have high resolution and other evidence for the resulting values. Occupancy and ADP-values are very strongly correlated, and the correlation becomes higher the worse your resolution. And neutron data often have resolution of 2A or worse.

You could group atoms together where you expect similar exchange ratios (based on chemical intuition) and refine one ratio per group. You can find a more detailed explanation inhttps://doi.org/10.1107/S1600576713027659 https://doi.org/10.1107/S1600576713027659

That article also recommends to treat X-ray data as additional information by means of geometry restraints to avoid the complications you get with joint refinement. The main effect of joint refinement are prettier maps <flame> and you may need to decide whether you want to focus on science or on art </flame>. This way we refined the structure also has the side effect to constrain H and D to the same location, although, admittedly, the coordinate difference in the structure you quote is probably negligible.

Best, Tim

On Thursday, October 13, 2016 05:06:36 PM Johannes Schiebel wrote:

...

Hi everyone,

I am currently working on an X-ray/neutron (XN) joint refinement using phenix.refine. As it should be, H/D coordinates and ADPs at exchangable sites are refined to equal values when using Phenix version 1.10.1-2155 and neutron data only. This is also stated in the paper describing the development of the method (Afonine et al. (2010) Joint X-ray and neutron refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 66, 1153-1163) as the default behavior: "Currently, phenix.refine maintains the H and D atoms at coinciding positions and constrains their ADPs to be equal to each other". However, when switching to XN-refinement using the same Phenix version, H and D atoms refine to different coordinates and ADPs, which should not be the case as it leads to unrealistic artifacts as can be seen from the deposited PDB 3X2P:

ATOM 186 H AALA A 13 3.486 -18.200 -14.123 0.38 14.54 H ATOM 187 D BALA A 13 3.488 -18.195 -14.111 0.62 38.07 D

In this example, the D-occupancy is likely overestimated, while the H-occupancy is underestimated because the ADPs refine to very different values, which is chemically not reasonable.

Hence my question: How can I treat my data in a way that ADPs and coordinates refine to the same values at exchangeable H/D sites also for XN-refinement? Is there a specific keyword I am currently overlooking or do I have to use another Phenix version?

I would be really glad to receive your feedback. Thanks in advance!

Kind regards, Johannes

_______________________________________________ phenixbb mailing list [email protected] mailto:[email protected] http://phenix-online.org/mailman/listinfo/phenixbb http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe:[email protected] mailto:[email protected]

_______________________________________________ phenixbb mailing list [email protected] mailto:[email protected] http://phenix-online.org/mailman/listinfo/phenixbb http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected] mailto:[email protected]

Dear Pavel. Thank you very much for your answer and explanations. Actually, in the first place I already refined against neutron data (without X-ray) until convergence and the results are quite good. However, water orientations (in which we are particularly interested) are not easily discernible in the maps since the knowledge about water oxygen positions from the X-ray data are not explicitly implemented during refinement. Hence, we collected room-temperature X-ray data from a deuterated crystal grown in the same drop as the crystal used for neutron diffraction experiments. With the combined XN-data, it is astonishing how the R-free of both X-ray and neutron data drop significantly (by 2% for the neutron part) and how well the data agree. To account for the longer X-D/H bond lengths in nuclear density maps, we used the respective restraints for neutron data (Settings --> Advanced --> Model interpretation --> Use the nuclear distances for X-H/D). Overall, I would really like to use the XN-refinement approach to improve the model in particular with respect to its water structure. Ultimately, we will anyhow deposit the neutron, X-ray and joint XN structures. Is it not possible at all to achieve coinciding coordinates and identical ADPs when using X and N at the same time? The only alternative that I have not already used and would further improve the water structure in the model would be what you describe under 3. From what I read about it, however, it will be difficult to generate restraints also for water oxygens for which it would be particularly important in our case. Or is this possible in addition to the dihedral restraints for the protein? Thanks a lot! Best, Johannes Am 14.10.2016 um 16:09 schrieb Pavel Afonine:

Hi Johannes,

historically, in joint Xray+Neutron refinement first tried by Coppens et al (1981) in small molecule world and later (1982) by Wlodawer & Hen­drickson in bio-macromolecules, and way later implemented in refinement tools such as nCNS (2009) and Phenix (2010), one single model is refined against two data sets, Xray and neutron.

The rationale to use two data sets, Xray and neutron, is to alleviate the seemingly poor data-to-parameters ratio because H and D are used explicitly in refinement using neutron data. This is all discussed in great details in that 6 years old paper:

https://www.phenix-online.org/papers/dz5209_reprint.pdf

Turns out with a proper data / restraints weight the problem of data-to-parameters amount is not as bad as it may appear, since by varying the weight one can dose the amount of a priori knowledge (restraints) as desired. This makes it possible to refine one model against X-ray or neutron data individually without problems.

Also, we know that X-H distances for X-ray and neutron are different (X-ray are shorter a tiny bit). We also know that both data sets may be collected at different temperatures (neutron at room, and X-ray at cryo), which means B factors are going to be different, some rotamers may be different, water structure may be slightly different, and so on. All in all we really do need to have two structures to maximize the use of available information. This is the new refinement paradigm that's being implemented in phenix.refine and will become available at some point.

The best you can do right now is:

1) Get the best possible refined X-ray structure (refined against x-ray data set alone); 2) Use refined x-ray structure as a starting point for neutron refinement (add H,D as appropriate - trivially done using phenix.ready_set or using Phenix GUI); 3) Refine structure from #2 against neutron data set alone. This will make sure H and D have identical coordinates and B. If really needed you can supply X-ray structure as a reference model - also trivially done from the phenix.refine GUI.

Let us know should you have questions or need help.

Pavel

On 10/13/16 23:06, Johannes Schiebel wrote:

...

Hi everyone,

I am currently working on an X-ray/neutron (XN) joint refinement using phenix.refine. As it should be, H/D coordinates and ADPs at exchangable sites are refined to equal values when using Phenix version 1.10.1-2155 and neutron data only. This is also stated in the paper describing the development of the method (Afonine et al. (2010) Joint X-ray and neutron refinement with phenix.refine. Acta Crystallogr D Biol Crystallogr 66, 1153-1163) as the default behavior: "Currently, phenix.refine maintains the H and D atoms at coinciding positions and constrains their ADPs to be equal to each other". However, when switching to XN-refinement using the same Phenix version, H and D atoms refine to different coordinates and ADPs, which should not be the case as it leads to unrealistic artifacts as can be seen from the deposited PDB 3X2P:

ATOM 186 H AALA A 13 3.486 -18.200 -14.123 0.38 14.54 H ATOM 187 D BALA A 13 3.488 -18.195 -14.111 0.62 38.07 D

In this example, the D-occupancy is likely overestimated, while the H-occupancy is underestimated because the ADPs refine to very different values, which is chemically not reasonable.

Hence my question: How can I treat my data in a way that ADPs and coordinates refine to the same values at exchangeable H/D sites also for XN-refinement? Is there a specific keyword I am currently overlooking or do I have to use another Phenix version?

I would be really glad to receive your feedback. Thanks in advance!

Kind regards, Johannes

_______________________________________________ phenixbb mailing list [email protected] http://phenix-online.org/mailman/listinfo/phenixbb Unsubscribe: [email protected]

Hi Pavel,

in 1978, there was a version of shelxl for joint refinement of X-ray and 
neutron data (Orpen, Pippard, Sheldrick, Acta Cryst (1978) B34 2466-2472)
The paper now cited for the Hirshfeld test (Acta Cryst A32 (1976) pp239) 
discusses the combination of refinement against neutron and X-ray data, 
including many references to Phil Coppens, although I cannot tell if this 
refers to joint refinement.

I appreciate your summary, it matches the conclusions about refinement against 
neutron data I mentioned earlier, to use the X-ray data as information for 
restraints.

For macromolecular data, the restraints often contribute greatly to the data 
to parameter ratio, e.g. for a shelxl refinement of 1L2K, there are about 1300 
reflections, 40,000 restraints and 9,000 parameters. And since the starting 
structure has often already been refined against the X-ray data anyhow, why do 
it again and risk a mix-up with your points of interest?

Cheers,
Tim

On Friday, October 14, 2016 10:09:55 PM Pavel Afonine wrote:
> Hi Johannes,
> 
> historically, in joint Xray+Neutron refinement first tried by Coppens et
> al (1981) in small molecule world and later (1982) by Wlodawer &
> Hen­drickson in bio-macromolecules, and way later implemented in
> refinement tools such as nCNS (2009) and Phenix (2010), one single model
> is refined against two data sets, Xray and neutron.
> 
> The rationale to use two data sets, Xray and neutron, is to alleviate
> the seemingly poor data-to-parameters ratio because H and D are used
> explicitly in refinement using neutron data. This is all discussed in
> great details in that 6 years old paper:
> 
> https://www.phenix-online.org/papers/dz5209_reprint.pdf
> 
> Turns out with a proper data / restraints weight the problem of
> data-to-parameters amount is not as bad as it may appear, since by
> varying the weight one can dose the amount of a priori knowledge
> (restraints) as desired. This makes it possible to refine one model
> against X-ray or neutron data individually without problems.
> 
> Also, we know that X-H distances for X-ray and neutron are different
> (X-ray are shorter a tiny bit). We also know that both data sets may be
> collected at different temperatures (neutron at room, and X-ray at
> cryo), which means B factors are going to be different, some rotamers
> may be different, water structure may be slightly different, and so on.
> All in all we really do need to have two structures to maximize the use
> of available information. This is the new refinement paradigm that's
> being implemented in phenix.refine and will become available at some point.
> 
> The best you can do right now is:
> 
> 1) Get the best possible refined X-ray structure (refined against x-ray
> data set alone);
> 2) Use refined x-ray structure as a starting point for neutron
> refinement (add H,D as appropriate - trivially done using
> phenix.ready_set or using Phenix GUI);
> 3) Refine structure from #2 against neutron data set alone. This will
> make sure H and D have identical coordinates and B. If really needed you
> can supply X-ray structure as a reference model - also trivially done
> from the phenix.refine GUI.
> 
> Let us know should you have questions or need help.
> 
> Pavel
> 
> On 10/13/16 23:06, Johannes Schiebel wrote:
> > Hi everyone,
> > 
> > I am currently working on an X-ray/neutron (XN) joint refinement using
> > phenix.refine. As it should be, H/D coordinates and ADPs at
> > exchangable sites are refined to equal values when using Phenix
> > version 1.10.1-2155 and neutron data only. This is also stated in the
> > paper describing the development of the method (Afonine et al. (2010)
> > Joint X-ray and neutron refinement with phenix.refine. Acta
> > Crystallogr D Biol Crystallogr 66, 1153-1163) as the default behavior:
> > "Currently, phenix.refine maintains the H and D atoms at coinciding
> > positions and constrains their ADPs to be equal to each other".
> > However, when switching to XN-refinement using the same Phenix
> > version, H and D atoms refine to different coordinates and ADPs, which
> > should not be the case as it leads to unrealistic artifacts as can be
> > seen from the deposited PDB 3X2P:
> > 
> > ATOM    186  H  AALA A  13       3.486 -18.200 -14.123  0.38
> > 14.54           H
> > ATOM    187  D  BALA A  13       3.488 -18.195 -14.111  0.62
> > 38.07           D
> > 
> > In this example, the D-occupancy is likely overestimated, while the
> > H-occupancy is underestimated because the ADPs refine to very
> > different values, which is chemically not reasonable.
> > 
> > Hence my question: How can I treat my data in a way that ADPs and
> > coordinates refine to the same values at exchangeable H/D sites also
> > for XN-refinement? Is there a specific keyword I am currently
> > overlooking or do I have to use another Phenix version?
> > 
> > I would be really glad to receive your feedback. Thanks in advance!
> > 
> > Kind regards,
> > Johannes
> > 
> > _______________________________________________
> > phenixbb mailing list
> > [email protected]
> > http://phenix-online.org/mailman/listinfo/phenixbb
> > Unsubscribe: [email protected]
> 
> _______________________________________________
> phenixbb mailing list
> [email protected]
> http://phenix-online.org/mailman/listinfo/phenixbb
> Unsubscribe: [email protected]
-- 
--
Paul Scherrer Institut
Dr. Tim Gruene
- persoenlich -
Principal Investigator
Biology and Chemistry
OFLC/102
CH-5232 Villigen PSI

Phone: +41 (0)56 310 5297

GPG Key ID = A46BEE1A