Re: [modeller_usage] The selection of models

[J B Procter <>]

BinBin - this is an even tougher question !

On Sun, 28 Mar 2004 13:10:07 +0100
Binbin Liu <> wrote:

> 100 models were generated based on one template. Their atomic RMSDs
> all are less than 0.4 Angstroms. Does any body know any other way I
> can find a representative model from the 100 models.

The low variation in RMSD is an indicator that you gave Modeller a good
set of alignments and templates to work with.  From your last email,
this variation seems appropriate for an essentially well
determined backbone with an ensemble of different side-chain
conformations. What you do now all depends on what the model is for.

If you have some biochemical criteria (ligand binding, metal ion site
conservation, etc), then you would want the model that has the right
conformation for the protein's function. This can be quite hard to do if
you don't really know what conformation is necessary for function,
though!

If you just want to pick the most likely model, and you do not have any
more information, then all that you can do is trust in statistics, and
molecular mechanics. The modeller objective function is one guide
to quality (as mentioned by Ben and Erick in another
posting "Re: OBJECTIVE FUNCTION"). Sort all the models by energy and
look at the lowest energy models (see the second line of each structure
file to get the energy). 

The alternative is to assess the structural quality of each model using
a tool like Procheck:

http://www.biochem.ucl.ac.uk/~roman/procheck/procheck.html

You could apply this to each model and pick the one which has the best
quality, based on all the Procheck analysis'.  There is also a very easy
to use analysis program called ERRAT, available from here :
http://www.doe-mbi.ucla.edu/Services/ERRAT/

Hope this helps,

j.
_______________________________________________________________________
Dr JB Procter:Biomolecular Modelling at ZBH - Center for Bioinformatics
Hamburg       http://www.zbh.uni-hamburg.de/staff.php