[Xplor-nih] Distance geometry output

[John Kuszewski]

Hi Jason,

Wow, what a blast from the past.  I didn't think anyone used the  
distance geometry engine any more!

It took me a while to untangle what's giving rise to the error  
messages you're seeing.  I'm not
absolutely positive of my diagnosis, especially since this seems to  
be code that Axel added after
I finished it.

My guess is that these error messages are coming from the limitations  
of triangle bound smoothing.
During metrization, you're picking random distances between the upper  
and lower bounds deduced
from the input constraints (bonds, NOEs, etc) using the triangle  
inequalities.  Unfortunately, the triangle
inequalities are insufficient to guarantee that the bounds they imply  
are embeddable in three dimensions.
Using inequalities involving more than three atoms at a time would  
produce more consistent distance
bounds, but it's prohibitively expensive in compute time (and  
programmer time).

The upshot is that the coordinates that come out of the embedding  
process will look crunched up, because
a bunch of the interatomic distances chosen aren't consistent with  
each other in three dimensions.
Occasionally, these inconsistencies are large enough for the  
metrization code to notice them, and it
produces these error messages.

In practical terms, these error messages are kind of pointless, since  
there's nothing you can do about them.
Using the triangle inequalities alone for bound smoothing (as nearly  
every distance geometry program does)
virtually guarantees geometric inconsistencies like this.  So don't  
worry about them.  I'm sure you're doing
extensive energy minimization afterward, anyway.

A few suggestions:

1.  To save some CPU time (and probably reduce the number of those  
error messages), you should switch
to four-atom metrization.  Change your script to read

mmdg
    metrization (all) 4
end

That will cut the compute time significantly without hurting the  
quality of the embedded structures (such as it is)
or their conformational sampling.  You could look at my Very First  
Paper for more details, if you wish.

2.  Don't use distance geometry!  It's quite slow, and the structures  
it produces require extensive energy minimization
to be usable.  You're much better off just running molecular dynamics  
based simulated annealing, starting from
random coordinates.

Hope this helps.

--JK

On Jun 19, 2007, at 2:02 PM, KWAN,JASON C wrote:

> Hello,
>
> At the moment I am trying to use the distance geometry feature of
> xplor-nih to produce starting structures of a natural product for
> further modelling.  I have put in some distance constraints
> derived from a ROESY experiments.  I get the following output, is
> it something to worry about?
>
>  X-PLOR>flags include noe end
>  X-PLOR>mmdg
>  MMDG>  metrization (All) 156
>  SELRPN:    156 atoms have been selected out of    156
>  MMDG>end
>  MMDG:    156 atoms will be randomly metrized.
>  DGSETNX: found    908 exclusions (1--2, 1--3) for repulsive lower
>
...
> * Geometric inconsistency--lower bound greater than upper.
> Attempted to continue by inverting these bounds.
>  %DISGEO-ERR: There are geometric inconsistencies. Proceed at own
> risk.
>  X-PLOR>
>
> Does the above indicate that I've done something wrong?  At the
> moment I've classified ROESY correlations as strong, medium and
> weak (<2.5A, 2.5-3.5A and 3.5-5.0A respectively).
>
> Thanks for your help
>
> Jason Kwan
>
> --
> KWAN,JASON C
>
> _______________________________________________
> Xplor-nih mailing list
> Xplor-nih at nmr.cit.nih.gov
> http://dcb.cit.nih.gov/mailman/listinfo/xplor-nih