[phenixbb] rotamers

Dale Tronrud det102 at uoxray.uoregon.edu
Fri Mar 25 12:58:01 PDT 2011

On 3/25/2011 8:53 AM, Donnie Berkholz wrote:
> On 12:29 Thu 24 Mar     , Dale Tronrud wrote:
>> Of the possible ways to represent this common situation I think
>> placing a side chain with occupancy set to zero is the worst.  For the
>> naive user (and that is not necessarly pejorative) it will lead them
>> to view the placement of those atoms as unwarrantably confident.  For
>> those with a deep understanding of the meaning of the parameters it is
>> flat wrong - it says that I know that this atom is never in this
>> location.
>> If you want to be precise for the knowledgeable user you need to put
>> in a non-zero occupancy but assign it a very large sigma on your
>> SIGATM record for that atom.  Of course when naive users see SIGATM
>> records I expect their heads will explode.
> Has anyone thought about putting them into a second MODEL record
> instead? That way, they're present for people who want them, but aren't
> shown by default in programs like PyMol. No clue how refinement programs
> would deal with this.

    That solution has two problems.  First, it would violate the definition
of MODEL -- Each MODEL is supposed to be a complete, alternative model fitting
the data.    Second, what do you do when you do have alternative models,
each with floppy bits?  There isn't a second level of MODEL.

    The more I think about it the better I like SIGATM cards.  Their usage
isn't as I suggested in my previous posting, of increasing the sigma of
the occupancy.  Instead you place the atom somewhere reasonable and give
the xyz a large sigma.  It has the advantage of using only long established
PDB syntax and using it for exactly for what it was intended.

    In addition, it is the right answer.  The main argument against my
previously favorite idea of not building floppy atoms was that the absence
of these atoms in the model indicate that I have no idea where they are.  In
fact I do.  They can't wander far from the ordered atoms they are covalently
bonded to.  Put each atom in a reasonable place and give their positional
parameters a 5 A sigma.

    Of course these sigmas would have to be calculated for all the atoms in
the model.  George Sheldrick would advise you on the calculation but, if
you have an aversion to inverting big matrices, Cruickshank came up with
an empirical formula about 10 years ago that might be useful.

    The PDB format does not have to be changed, but we would have to crusade
to get software to begin to pay attention to SIGATM.  A program like Coot
could have the option of showing spheres whose size is related to the atom's
uncertainty in location.  An electrostatic calculation program would have
a direct indication of how far it could move each charged atom to optimize
the electrostatics while still being consistent with the original x-ray model.

    This seems to be a win-win solution.

Dale Tronrud

More information about the phenixbb mailing list