 |
 |
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Hi Pavel, maybe I should have explained in better detail to avoid confusion. The importance of the contribution to the X-ray term from hydrogens has been well-known (and used in refinement programs) for - I guess - more than 40 years, but I am sure you know that. I meant to say that small variations in the hydrogen position don't show up for X-ray data, which is why they can be calculated even at 0.8A resolution. Maybe you could repeat the study you point at and move the hydrogen atoms, e.g. 0.3A out of their calculated position - my guess is such a variation would no show up in the map or the R-values at resolutions ranges significant for macromolecules. But these differences may matter when it comes to quantum chemical calculations, and Jeffrey's post seems to support this. I am puzzled that you observe the same with neutron data, i.e. you see no difference between the riding hydrogen model and restraining them. Does phenix, by any chance, refine the B-values rather than constrain them as implied by 'riding atom model'? This would be very unusual and might explain why the differences are not detected with phenix. In our work using shelxl, which I mentioned before, the quality differences where really striking despite the low data completeness. Interesting that phenix seems to wipe out the most interesting part of (MX) neutron data. Cheers, Tim On 06/16/2014 08:09 PM, Pavel Afonine wrote: > Hi Tim, > > just to spice your words up with some numbers.... > > You may also want to note that constrained hydrogen positions are > a >> crude approximation and only work with X-ray data where hydrogen >> atoms have little impact on the data. > > > This contribution can be as large as 1.5% difference in R-factor > (with vs without H), as shown in Figure 2 (page 19; "On the > contribution of hydrogen atoms to X-ray scattering"): > http://phenix-online.org/newsletter/CCN_2012_01.pdf > > >> Our comparison between hydrogen restraints and constraints >> (http://dx.doi.org/10.1107/S1600576713027659) report the greater >> quality of restraints vs. constraints when it comes to neutron >> data, where hydrogen atoms do matter. > > > I just re-refined (phenix.refine) all neutron structures available > in PDB (for which I could extract diffraction data without manual > labor; 55 in total) with two ways of handling H (D and H/D) atoms: > a) refine H individually, and b) using riding model for H > (rotatable H are adjusted to fit the map). In terms of Rfree and > Rwork I don't see a huge difference. However, using riding model > results in less overfitting: > http://cci.lbl.gov/~afonine/tmp/r_stats.pdf > > This is not surprising given typical quality of neutron data: > average (all neutron entries in PDB) completeness of neutron data > sets is 76%, while average completeness of comparable X-ray data > sets is 94% (page 21): > http://phenix-online.org/presentations/latest/2012_afonine_ecm27-final.pdf > > All the best, Pavel > - -- Dr Tim Gruene Institut fuer anorganische Chemie Tammannstr. 4 D-37077 Goettingen GPG Key ID = A46BEE1A -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Icedove - http://www.enigmail.net/ iD8DBQFToEdAUxlJ7aRr7hoRAk/2AJ9WEomb6EVbAj7vkhHq3cv6l1YivQCcCGgc 4ydPJJRvRwer6SnDnvhvG+k= =rG3w -----END PGP SIGNATURE-----