gerard at rigel.bmc.uu.se (Gerard Kleijwegt) writes:
>As any crystallographer who uses X-PLOR knows, MD is the quickest way
>to screw up any perfectly good structure if you switch off the xray
>pseudo-energy term; a few ps at room temperature will do it.
No-ones asking you to do that though!
>|> And what was that about needing 80% identity to get a reasonable
>|> comparative model - is it my imagination, or is your estimate of the
>|> required sequence simliarity going up!
>???
I thought I'd seen a post a while back where you'd said that something
like 60% identity was needed - but maybe it was someone else.
>|> You can easily build "OK" models in these cases.
>- "OK" for what purpose ? Getting it published ?
OK for giving you insights into the way the protein might work i.e.
rationalising all kinds of structure-function studies (mutatgenesis,
protein chemistry, mechanistic studies etc)!
> Of course, xtallographers use "homology models" all the time, namely
> when they make search models to be used in Molecular Replacement
> exercises. Except, rather than assuming they know enough about
> nature to be able to confidently "fill in the blanks" (substitutions,
> deletions & insertions), they usually *strip* the homologous structure
> of any atoms that can be expected to be different (side chains, loops,
> areas of insertions or deletions), and with success !
Didn't Alwyn Jones "invent" database searching for loops of unrelated
known structures that might fit into a new structure?!
> May I take the opportunity to suggest a simple test to see if homology
> models are better than random ?
NO! Well OK then...
<stuff about molecular replacement deleted>
If you've got enough phasing power from the backbone of the
homologous protein to find the "real" structure in the data, then that's
all you want isn't it? But you might save yourself some bother of
modelling interior side-chain conformations by using state-of-the-art
prediction methods, once you've adjusted the tracing of the main-chain so
that it's pretty much correct for the "real" protein (or even just
Calphas would do)?
>- even if the fold can be predicted, this usually carries very little
> biologically-relevant information.
If this were true, people should stop wasting space in front
line journals describing the folds of proteins. But it isn't!!
Obviously some models will be more accurate than others, and I agree
with you that large insertions can't be modelled well. But you can
sometimes do quite nice predictions with deletions, and small insertions -
and these can be just as important as large insertions.
Homology models are better than nothing - honest! And like others
have said, the technique is often much quicker than X-ray or NMR.
Although I'd grant you the fact that if you're doing a structure by
molecular replacement, and your data diffracts to 1 A, then you could
knock it out in a weekend ;-) But us poor NMR people are stuck working
away for a few months, minimum, to get a structure better than a good
homology model.
_________________________________________________________________________
|| ,_ o Simon M. Brocklehurst,
| / //\, Oxford Centre for Molecular Sciences
| \>> | Department of Biochemistry, University of Oxford,
| \\, Oxford, UK.
| E-mail: smb at bioch.ox.ac.uk|________________________________________________________________________
