*** For details on how to be removed from this list visit the ***
*** CCP4 home page http://www.ccp4.ac.uk ***
...
However, one can also use rmsd's to guide the refinement process
itself. For example, at low resolution, we can't distinguish
between bond lengths of 1.4 or 1.6Å. Should we therefore not allow
the bond lengths to vary much more than at high resolution? Target
rmsd's should therefore be higher at low resolution than at high
resolution, not the other way around. They should go parallel with
the coordinate error. The argument is that using constraints (i.e.
low rmsd's) at low resolution gives the impression of a precision
that is way too high compared to the information content of the
data. This assumes using rmsd's as error models, which we have to
match to experimental errors.
> ...
Here, I disagree: as you stated above, the model should make
physically and chemically sense. If you allow a larger deviation
from expected chemistry only because the lower resolution of the
observed data doesn't allow for a more precise model, you will
violate that physical/chemical reasoning. The underlying problem
is, as Gerard has pointed out, that there is no good single model
that would explain low resolution data. In principle, one should
describe these data with multiple models, but this is prohibitive
given the low amount of data. I also think that the best that one
can currently do with low resolution data is to use a single model
tightly restrained to expected chemistry together with TLS to
describe at least partially the model's rigid body flexibility.
Quite plausible, or not? In fact, we violate physical/chemical
reasoning all the time in other parameters
To let Tassos' nightmare come true (perhaps), the rmsd's indeed look
like swash buckets for all kinds of things we don't quite know how to
model. As Gerard pointed out, lower resolution means more disorder
(some dynamic, a lot of static), probably for the largest part due to
crystal imperfections. As a consequence, we have large uncertainties
in the positions of the atoms in the model. It makes sense to reduce
the number of parameters at low resolution, but changing rmsd targets
does accomplish that. We are still refining atomic positions. Unless
there is some ingenious scheme that converts tight restraints into a
reduced number of parameters (there might very well be), one would
really have to move over to some form of constraints.
Anyway, the physical reality is that the uncertainties in the atomic
positions are not due to varying in the bond lengths and should
therefore ideally not be represented as such. Non-structural
biologists, who are not familiar with the statistical peculiarities
of X-ray crystallography have problems with varying bond lengths, and
rightfully so. That's one of the reasons I think rmsd's are a rather
poor way of describing the error in a model. One could very well use
the same bond length parameters derived from quantum mechanics
throughout all resolution ranges (because they represent physical
reality, and reality doesn't change with resolution) and use other
parameters for absorbing the uncertainties. The usual suspects come
to mind (B values, TLS), but as long as we continue to absorb errors
in our data with parameters like geometric rmsd's, there seems to be
a need to come up with a new concept. Alternatively, we could, as
Eleanor suggest, simply use Rfree as a guide and forget about rmsd's.
To me, they seem to be as meaningful or meaningless as B values. We
will have to fight reviewers again, justifying our weird rmsd's.
Better to come up with a new parameter that doesn't have anything to
do with a particular physical phenomenon, but still describes
disorder, and give it some bogus name. No reviewer will complain
about that.
------------------------------------------------------------------------
--------
Mischa Machius, PhD
Associate Professor
UT Southwestern Medical Center at Dallas
5323 Harry Hines Blvd.; ND10.214A
Dallas, TX 75390-8816; U.S.A.
Tel: +1 214 645 6381
Fax: +1 214 645 6353
