Hi,
On Nov 9, 2007 4:20 PM, Sebastien Morin <sebastien.morin.1@xxxxxxxxx> wrote:
Hi Ed,
Your comments are interesting... but nevertheless trouble me...
My comments being troubling is a common occurance ;)
I talked with the guy who makes the MD simulation with my protein. He
noticed that some NH vectors indeed have motions well defined by a cone
which is fine if we want to analyze things with the model-free approach.
However, some NH vectors have motions more like a stretched cone (which is
not surprising at all, since motions inside a protein are restricted by the
available space as long as with the position of charges, etc), More worrying
are some motions seen as two separate cones, which is then bad, since an
average position doesn't have any meaning and can even be impossible due to
steric or charge problems...
This is all true, but not something that I have addressed in my
studies. This affects more that model-free analysis though - it
touches many different parts of NMR (RDCs are a good example).
Ok. The reason I would not use a crystallographic structure is that maybe
some NH vectors could be badly oriented because of packing modifications in
the crystal (at interfaces)... An NMR structure may be better, but
impossible for a lot of people, like me...
Most crystallographic structures are not high enough resolution so
that the PDB file will not have the NH protons included. The addition
of the protons is by some subsequent algorithm. Whether NMR or X-ray
structures are better (not generally but on an individual level), I'd
prefer not to get into that.
But what does the individual NH vector orientation really affect ?
For model-free analysis - everything.
Will a
different NH vector orientation give rise to a different selected model or
will it only affect model parameters ?
It will affect the parameters of internal motion (artificial Rex
(Tjandra 1996) or artificial nanosecond motions (Schurr 1994)), which
will then affect model selection and affect the phenomenon of model
failure, and all of this will then influence the global diffusion
parameters (and vice versa).
Also, will some bad NH vector
orientations (let's say less than 10% bad orientations overall) affect the
global diffusion parameters ?
Yes. How much and how significant this is is another question. Are
the bad orientations causing bias or just adding random noise? That
was a rhetorical question :)
In other words, will those possibly bad
orientations produce artifacts, like false Rex or modified tm, for example ?
Yes.
Another question... The NH vectors are described by two coordinates (for H
and N) in the PDB files, but transformed to angles within relax or any
program alike (Am I right ?). Can a user directly input a set of angles
rather than coordinates ? I ask this because I'd like to try to input
average orientation which were computed to give a set of angles rather than
a modified PDB file... Are the coordinates necessary to analyze the global
diffusion (to assess the shape of the protein) ? Are the angles only used
for local motions analysis ?
relax uses coordinates, specifically unit vectors. This is because
dot products, which are very fast, are used in the model-free code.
This way everything is done in the molecular frame and the conversion
between say spherical angles defining the vector in that frame to an
inter-vector angle is unnecessary. It should be easy though if you
have these average angles to create a PDB file with all heteronuclei
at [0,0,0] and the protons placed appropriately. Alternatively using
scripting you can place the unit vectors directly into the relax data
storage structure 'self.relax.data.res' (in relax-1.2) and not use the
PDB reader. For the linear algebra - dot products, etc. - these must
be 'Numeric' arrays though!
Thanks for the tips / opinions / comments / anything !
I hope my answers were useful.
Regards,
Edward
Re: PDB orientation