Dear K,
I do think there is much of a hornet's nest to be stirred up by your
questions.
I hate to advertise it, but chapter 12 of BMC has a thorough discussion of
the
subject.
Ad riding: ADDING riding hydrogens is not exactly the same as MODELLING
some stuff one has no clue about (like parts with no density). Covalently
bonded
hydrogens are not really 'modeled' in a sense of parameterization,
they are almost certainly there, and add no parameters to a model. They
just make certain computations of restraints and validation work better when
included.
Propositions about other non-H parts not visible in density, are an entirely
different beast. The crystal is made of trillions of molecules, which
when they have almost identical conformation in almost all the scattering
molecules
will show distinct density. So if no density is visible, what are you
modeling?
One possible conformation of many? The most probable of the
non-evidence-bearing ones?
Much better to fess up and leave the residues/atoms out of the model.
Increased B
should be in any case a result of refinement and not arbitrarily set (and
let's not go into
details of B-restrains which I believe are to this date a challenge to
properly implement)
while adjusting occupancy is not an option for a main chain.
In addition, setting occupancy to 0 for conjured parts is bad because most
display programs and users will ignore that column. In practice (as theory)
omitting parts that have no density = high B exponential = negligible
contribution
to scattering factor rarely has a significant effect on refinement
parameters,
but the model with absent parts is a more truthful, parsimonious model.
BR
-----Original Message-----
From: CCP4 bulletin board [mailto:[log in to unmask]] On Behalf Of
SIPPEL,KATHERINE H
Sent: Tuesday, April 13, 2010 5:40 PM
To: [log in to unmask]
Subject: [ccp4bb] To model or not to model...
Dear Crystallographic Community,
Dr. Holton made a comment today that got me thinking on the issue
of modeling. This has been a hotly debated topic in our own lab
but I would like to hear the current opinions of the community as
a whole. It is a question of two parts.
First, what do you think about modeling into regions of poor
density? Do you (A) model something in as best you can while
conforming to ideal geometry/chemistry with full disclosure about
b-factors in the region, (B) reduce the occupancies of the poorly
modeled loops/side chains, or (C) truncate your model, removing
loops and side chains from the model at the cost of statistical
numbers? From a crystallographers point of view we can assess the
quality of the model and make informed decisions as to what
conclusions to draw, however most of the greater scientific
community has no way to judge this. They do not know what a
b-factor is or where to look for an occupancy. Are we doing a
disservice to science by emphasizing the minimization of Rwork and
Rfree over full disclosure of what we can legitimately see?
Secondly, on a similar vein, what is the community's opinion on
modeling hydrogens? I have read a lot on the subject and can see
both sides of the argument. From a crystallographer's point of
view these are very helpful in maintaining geometry and ensuring
the model makes chemical sense. I can also see the necessity of
submitting them to the pdb so that the statistics can be
recreated. On the other hand most biologist has no comprehension
of the concept of riding hydrogens. They assume that if the
hydrogen is in the pdb, that the crystallographer saw it and use
that information to develop experiments.
I realize that I may be kicking a hornet nest here but I would
genuinely like to know what people think.
Thanks,
Katherine Sippel
SIPPEL,KATHERINE H
Ph. D. candidate
Department of Biochemistry and Molecular Biology
College of Medicine
University of Florida
|