[ccp4bb]: Riding Hydrogens Summary

[Jason Yano <jkyano@scripps.edu>]

***  For details on how to be removed from this list visit the  ***
***          CCP4 home page http://www.ccp4.ac.uk         ***

Hi All,

Here is the summary from the question posted regaurding riding hydrogens:


Ethan Merrit

This question is a bit more subtle than you may think.
Since hydrogen is a weak scatterer, it only contributes significantly at
low resolution (yes - low not high).  So if you can place all those H
electrons at their correct position in the cell then your R and Rfree
should improve even at quite low resolution.  The difficulty is in that
"correct position" bit.  Usually it is only when your structure is well
refined to high resolution (better than 2A let's say) that the heavy atom
positions are accurate enough that you have a good shot at placing
the hydrogens correctly.

So if you take a model refined at atomic resolution and calculate
R factors against 3A data with and without hydrogens, it is likely
you will see an improvement from including the hydrogen scattering.
But if you only had 3A data to begin with for that same structure,
it is unlikely that you would have been able to generate a model
accurate enough to benefit from adding hydrogens.  Not because
the hydrogen scattering is insignificant, but because you cannot
place the hydrogens correctly.

The bottom line IMHO is that the lower R-free you observe is in itself the
justification you are asking for.  You have added what - 12 parameters?
(I'm not sure exactly how many parameters are in the refmac riding-H
model but it must be on that order).  If adding 12 parameters lowers
R-free by 1% that is good justification for including them in your model.
The fact that these particular 12 parameters describe hydrogen atoms
is almost beside the point.  Suppose you had added 3 water molecules -
that also is 12 parameters - and saw Rfree go down by 1%.  Would you
have posted to the bulletin board asking whether you were justified in
keeping your three new waters?

Oh, and for another discussion on the utility of adding hydrogens
to medium-resolution structures you might want to have a look at
the Richardson lab web pages on model validation.

        http://kinemage.biochem.duke.edu/validation/valid.html

The key notion here is not one of R or R-free, but simply that any
correct structural model for the heavy atoms must leave room for
the attached hydrogens. If there is no room, then something must
be wrong, or at least non-optimal. So adding hydrogens can be
used as a structure validation tool.


Dave Schuller

1) your protein actually has hydrogens, that is why adding hydrogens to
the model brings it closer to reality, i.e. lowers R and Rfree.

2) as mentioned in your subject header, these are "riding hydrogens".
they 'ride' in a set position relative to the other atoms.  the positions
of the hydrogens themselves are not independently refined. that is why
your number of refined parameters has not gone up, which might have
led to over-refinement and increase of the R : Rfree split.



William G. Scott

If they lower the R, it may be just because you have more adjustable
parameters.
If they lower the R-free, it is either (a) real or (b) you have more
than one molecule in the asymmetric unit
and your test and working sets are not therefore strictly independent,
so having more adjustable parameters will also lower R-free.

In addition, each hydrogen has 1 electron that scatters X-rays, so
modeling in all the electrons in your molecule and accounting better for
their
distribution will justifiably lower R-free.

Now if you really want to freak everyone out, truncate your data at 3.0
Angstroms and then refine with a single TLS group that includes all
atoms in your structure (only 9 additional adjustable parameters) and
watch what happens to R-free.


Bart Hayes

Since they are riding hydrogens, I would expect that the coordinate error
in
the hydrogens is similar to that of the carbon and nitrogen atoms to which
they are attached. Clearly, the coordinate error will be greater for a
lower
resolution structure but at the same time, coordinate errors have less
impact
on R/Rfree at lower resolution. So I would think adding hydrogens still
improves the model even though the effect on R/Rfree will be less.

People should also keep in mind that adding 8 hydrogens and therefore 8
electrons does not have the same effect as adding 1 oxygen that also has 8
electrons. The effect is only the same if all 8 hydrogen electrons scatter
in
phase and of course they don't. You actually have to add 64 (8**2)
hydrogens
to get the effect of 1 oxygen (probably even more since, as Ethan pointed
out,
the atomic scattering factor for hydrogen falls off faster with
resolution).

Another practical point concerns the treatment of the hydrogen atoms by
the
refinement program. Are they used for pairwise interaction calculations,
are
they used to determine the solvent mask. If so that can effect the
refinement
as well.


Garib N Murshudov

One of the ideas of adding hydrogens in their riding positions was that
their
addition in some way is validation of accuracy of the refined atoms.
Hydrogens are not refined. But they are used for structure factor
calculations and geometry (bond lengthes, angles, antibumping). Their
positions are "idealised" during refinement. Usually their addition
improves
overall geometry like torsion angles. Sometimes they are useful in
spotting
"problem areas". Look at the list of vdw outliers.

As name suggests only those hydrogens for which position can be preicted
from
their parents are included. Other hydrogens have 0 occupancy and do not
contribute.


Thank you everyone for the input,


Jason