AbM problem?

A.H. Henry mbahh at s-crim1.dl.ac.uk
Mon Jul 4 06:59:31 EST 1994


[I posted this from Bath on Friday, and it had not appeared at
 Daresbury by Monday.  Apologies if you see this twice  - AHH]

In the referenced article, philstas at vub.ac.be (Stas Philippe) writes:

>I am trying to get Oxford Molecular's AbM V2.0 to work on an SG Indigo
>machine.  It runs, and seems to do most of its job, but invariably
>crashes in the final run of congen/eureka screening... Its very
>frustrating to get so far and then dump out with "unknown error"
>messages... I surmise that the modelling parameters need to be changed
>in order to reduce the number of possible structures generated by
>congen, since I think it is a memory problem that is killing the
>program, but as a beginner modeller, I don't know what and to what
>extent to change the variables... 

I'm one of the postgrad's in Tony Rees's lab.  I didn't write any of
AbM, but I'm pretty familiar with it.

I'm pretty sure that you're not running out of memory.  Each of the
Eureka screens is basically independant, so it doesn't whether you get
one or a thousand conformations - they will all use a similar amount of
memory.

Can you give me the abm_config* files with the sequence in ?  What
machine and version of the operating system are you using.

>I'd like to hear from anyone who has
>experience with AbM, or news of alternative packages that can turn
>sequence data into a reasonable .pdb file that can be displayed in
>Pimms or RasMol etc...

For an antibody, I think AbM is going to be your best bet.
It has been tested with the seqeuences of 30-35 known
structures.  When its inaccurate, at least you can get a good
idea of where the inaccuracies are likely to be.  If you run
a test to see how AbM performs on a known structure, either
use a newly released one, or you can make a file which with
a list of the structures you _don't_ want AbM to use.  I can't
remember what this file should be called.

I'm not exactly sure just how much of this information is
included with AbM.  We're planning to show some of our
testing data at the Oxford Protein Engineering conference
in September if that is any consolation.  One of the criticisms
that has been levelled at AbM is that it is a tool that
produces a pdb file and that it doesn't really give you
much of an idea of what to do with it after that.

Whilst its fine to have an automated procedure
to make a model, you should check it to see if is similar
to the known antibody structures.  These are the "raw
materials" for a homology model.  If the model disagrees
strongly with the known structures, its quite possibly wrong.

There is a lot of structural conservation between
antibodies.  These areas are modelled the most accurately
by AbM.  However, you could often get the  same "useful"
information by looking at similar crystal structures.
The sidechains for most framework residues and many CDR
residues all tend to point in a similar direction in
similar crystal structures.  If your sequence is similar in
a particular region to 5 structures, all of which have a
very similar structure, you can have a reasonable bet that
your structure would also be similar.

I think this approach is much better than using a
relatively modelling procedure, where you never know just
how accurate parts of your structure are.

Its a big help to be able to load up several structures at
once to see where they are similar and where they differ.
You can't do this with RasMol at the moment, although it is
coming, and I've made a conscious effort to avoid all
contact with Pimms, so I don't know whether this will be
easy for you to do.


-- 
Andrew Henry           | Read the rec.autos.sport List of Frequently
A.H.Henry at bath.ac.uk   | Asked Questions v1.0 available by anonymous
University of Bath, UK | ftp at mgu.bath.ac.uk      _/Sempre  Gilles



More information about the Molmodel mailing list