Hi Vineet,
Please see below:
I ran another run, this time using the pdb file containing all the 15
lowest energy structures. Analysis of the same relaxation data with this
structure gives me an ellipsoid diffusion model with following parameters..
If you look at the logs, which you can save if you use the --tee
command line option (the Mac OS X App excluded, you will probably see
that only one of the structures is loaded.
Diffusion type ellipsoid
tm (s) 4.63094e-09
Diso (rad/s) 3.59898e+07
Da (rad/s) 2.4722e+07
Dr 0.475806
Dx (rad/s) 1.59863e+07
Dy (rad/s) 3.9512e+07
Dz (rad/s) 5.24711e+07
alpha (rad) 2.59643
beta (rad) 2.087
gamma (rad) 0.438009
Fixed flag True
I remember reading an old post where Ed mentions that relax is not set up
to carry out analysis for residues with multiple NH bond vectors since it
lies in the model-free theory itself.
That would be the post at
http://thread.gmane.org/gmane.science.nmr.relax.user/1561/focus=1562.
In the same post it is also mentioned that if that is the case then relax
should give an error. Now, about 5 residues at the N terminus and 2 at the
C terminus of my 43 residue protein are unstructured and show high backbone
RMSD. It would be great if i get some help regarding the understanding of
why in such a case, there was no error given from relax. And also, what
measures are to be taken to run relax analysis on such a protein using the
15 structure ensemble as an input structure. Is this change from oblate to
ellipsoid a case of over-estimation of the diffusion properties occuring
due to highly flexible termini in the protein?
It is true that you cannot execute a model-free analysis when multiple
bond vectors are loaded into the relax data store. The problem is
that model-free theory is derived with the assumption that the XH bond
vector is centred in the distribution of motions. There is nothing in
the base Lipari and Szabo theory for handling multiple bond vectors.
If you would like to use multiple structures and are good with theory,
you could try to derive a new theory where the bond vector
distribution is considered as a subset of the true distribution of
bond vectors. This would be quite revolutionary though and would be a
paper in itself.
If you manage to successfully perform an analysis with multiple bond
vectors present, then something strange will be happening. The
results are not to be trusted. If so, could you please create a new
bug report at https://gna.org/bugs/?func=additem&group=relax. This
situation must be caught by relax and a RelaxError raised which
explains the problem and prevents the analysis from going forward.
If you also solve how to best handle multiple bond vectors in the
model-free protocol, rather than in the original theory (these are
quite different, see the protocol at
http://www.nmr-relax.com/api/3.1/auto_analyses.dauvergne_protocol-module.html
and
http://www.nmr-relax.com/api/3.1/auto_analyses.dauvergne_protocol-pysrc.html),
then this will also be worthy of publication.
However the best you can currently do is to perform the analysis on
multiple structures independently and come up with a consensus. You
should also compare results to the base relaxation data to see if it
makes sense.
Cheers,
Edward
Re: