Re: [ccp4bb]: homology modelling?

[Karsten Suhre <Karsten.Suhre@igs.cnrs-mrs.fr>]

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

Hi!

In reply to Randy Read's question: Yes, we have recently solved a structure 
using homology modelling in a situation where molecular replacement failed 
(i.e. the two potential templates had less than 25% sequence identity). A 
second case also would have worked, but as we had MAD data available for that 
protein at that time we did not proceed any further on it. 

However, it did take some time to get there, that is testing different 
alignments (T-COFFEE-SAP worked best) and scoring numerous different models 
(using MODELLER) with AMoRe before we finally got one that worked (there were 
4 molecules in the asymmetric unit, each 200 residues long). We also erased a 
badly aligned loop and some of the N-terminal residues, but they could be 
reconstructed using ArpWarp and Buster lateron. 

Note that D.T. Jones proposed a similar approach in Acta Cryst. (2001) D57, 
1428-1434.

Kind regards, Karsten.

On Thursday 21 November 2002 10:59, Randy Read wrote:
> Following up on what Bart has said, I've been asked several times recently
> whether or not one should use homology models for molecular replacement, in
> preference to the original crystallographic model with the wrong sequence.
> My (also biased?) impression is that homology modeling makes the models
> worse for molecular replacement, and I have been suggesting that it's
> probably best to stick to the original model, omitting only parts that are
> clearly very different, and perhaps downweighting loops and side-chains
> that are likely to fit poorly (as done in MolRep).
>
> Of course, at some point the techniques of homology modeling will have
> improved enough that the resulting models are actually better when the best
> methods are applied carefully.  So the question is whether we're reaching,
> or even nearing, that point yet.  It would be interesting if people could
> supply clear-cut examples of where homology models were better for
> molecular replacement than the original structures.  Does anybody out there
> have any such examples?  Is my impression still correct, or is it finally
> out of date?

> On Wednesday 20 November 2002 18:53, Bart Hazes wrote:
> > Basically the question is whether current software
> > will create a model that is more similar to the real structure than the
> > template the model was based on. In the past my (biased?) impression was
> > NO and the paper above seems to confirm that. So unless the YASARA force
> > field is an acronym for "Your Atomic Structure Always Reliable &
> > Accurate" I think modeling should be approached with a properly sceptical
> > attitude. The question shouldn't be "what is the easiest way to apply a
> > homologous sequence to a known crystal structure" but what is the most
> > accurate method and will it be sufficient for the questions you have.
> >
> > I have been recommending people not to make homology models at all.