Thanks George! I didn't feel remotely qualified to describe how SHELX
works, and that is why I didn't.
On 1/7/2011 3:11 PM, George M. Sheldrick wrote:
> Dear James,
>
> I did once write a refinement program so I should try to answer your
> question. All hydrogen atoms attached to C and N in standard proteins and
> almost all in DNSs and RNAs can be calculated geometrically with no
> ambiguities. This has been standard practice in SHELX for about 35 years
> and I believe that Refmac and Phenix_refine do so too. The bond lengths to
> hydrogen are set to values about 0.1 shorter than the neutron diffraction
> values to best fit the electron density. Methyl groups attached to sp3
> atoms are staggered. For other methyls SHELX has a facility to calculate
> the electron density around a circle and does 3-fold averaging to find
> the best torsion angle. H on O can either be found in this way or the
> torsion angle chosen that makes the best H-bond. For macromolecular
> refinements I recommend including hydrogens at any resolution but leaving
> out H on O unless (at very high resolution) they can be seen in the
> difference map. Adding the electrons to the C, N or O atoms is not a good
> idea.
>
> Best wishes, George
>
> Prof. George M. Sheldrick FRS
> Dept. Structural Chemistry,
> University of Goettingen,
> Tammannstr. 4,
> D37077 Goettingen, Germany
> Tel. +49-551-39-3021 or -3068
> Fax. +49-551-39-22582
>
>
> On Fri, 7 Jan 2011, James Holton wrote:
>
>> Obviously, this depends a bit on which refinement program you use, and I am
>> not familiar with all of them. However, I have had some conversations with
>> Garib "refmac" Murshudov and Pavel "phenix.refine" Afonine about this, and the
>> shocking answer appears to be "no". Carbon has six electrons, even if it is a
>> methyl carbon with three unmodeled hydrogens stuck to it. Yes, this does
>> introduce a 33% error in the number of electrons at every CH3! I am also
>> suspicious that free ions like "Cl" are often modeled as free elemental Cl,
>> and not the Cl- ion, which has 1 more electron. But this is only a 3% error.
>>
>> Personally, I think that "something should be done" about disordered aliphatic
>> hydrogens (particularly since I don't have to code it). I suggest that simply
>> adding "3" to the "constant" coefficient of the "C" form factor to make a "CH3
>> form factor" might just do the trick. Not only are the electrons brought in
>> by the H atoms generally delocalized into the carbon's electron cloud anyway,
>> but at most MX resolutions, the whole CH3 group is just one big fat lump of 9
>> electrons. Mathematically, this is equivalent to saying that the B-factor
>> Gaussian (1.5 A FWHM when B=30) dominates the 5 much narrower Gaussians of the
>> atomic form factor.
>>
>> That said, I should also add that our illustrious refinement program authors
>> have good reasons for doing things the way they do. Mostly because users can
>> say some very unkind things if the new version makes their R factors go up.
>> It is true that riding hydrogens "count" as one electron each, but it is not
>> always clear where to put them, and in situations like methyl group hydrogens
>> they could be anywhere in the "doughnut of uncertainty" traced out by rotating
>> the X-CH3 about its X-C bond. It is tempting to say that this doughnut should
>> be modeled in somehow, but an important caveat to remember is that putting an
>> atom in the wrong place is twice as bad as leaving it out. That is, you get
>> one difference feature if the atom is missing, but two (one positive, one
>> negative) if it is wrong. So, a "doughnut atom" is expected to at best cancel
>> itself out. One could call this a corollary of of the general principles of
>> model building: "When it doubt, don't".
>>
>> But what about MolProbity? and all those neat inter-digitating hydrogens?
>> Should it not be possible for the refinement program to be "smarter" about
>> where it puts riding hydrogens? Well, that is always possible, but I don't
>> think MolProbity is exactly built into refmac. You can, however, run
>> MolProbity before you put your model into refinement!
>>
>> The problem, of course, is if you have a 4.5 A structure with all the
>> hydrogens built in you will get hateful comments containing the word
>> "parameters" from undereducated reviewers. I say "undereducated" because
>> getting the electron count right is actually MORE important for low-angle
>> structure factors than it is for high-angle ones (the extreme case of this
>> being F000 itself). Yes, explicit hydrogen atoms do slow down the refinement,
>> and clutter the graphics, but they do not really add much by way of "free
>> parameters". Not if the geometric restraints are sensible. In fact, the
>> central lesson of MolProbity is that hydrogens do introduce (at least
>> potentially) better geometric restraints. Sort of like what an elastic
>> network model can do (ahem).
>>
>> -James Holton
>> MAD Scientist
>>
>> On 1/6/2011 8:15 PM, Kenneth A. Satyshur wrote:
>>> Persons of interest:
>>>
>>> I understood that refinements in refmac of low res structures without
>>> hydrogens includes a component of the Hydrogen atom attached to the
>>> carbon in the carbon scattering factors. Or is this just if the Hydrogen
>>> atoms
>>> are in the riding position. If so, why bother to add hydrogens in the
>>> riding
>>> position.
>>>
>>> Thanks
>>> kas
>>>
>>>
>>> --
>>> Kenneth A. Satyshur, M.S.,Ph.D.
>>> Associate Scientist
>>> University of Wisconsin
>>> Madison, Wisconsin 53706
>>> 608-215-5207
>>
|