On Wed, 2006-11-01 at 03:08 +1100, Edward d'Auvergne wrote:
Oh, and one last thought. It would be nice to automatically scale the
tensor size for the size of the protein, so we don't have to do
trial-and-error adjustments of the scaling factor. The current fixed
default scaling factor means larger proteins will have smaller tensors
and vice-versa. This is something I could have a go at implimenting if
others think its a godd idea?
The reason I have used a scaling factor is twofold. Firstly for
comparing two states or systems, you need to have exactly the same
scaling in both analyses. Secondly the scaling factor should probably
be given with the figure. The value is important - it is the
diffusion rate per Angstrom within the figure. I should add this to
the user function docstring.
What I had in mind was a default auto-scaling that can be overiden as
required. ie.
def pdb.create_tensor_pdb(run, file, scaling):
if scaling == None:
scaling = autoScale()
...
In our experience of using this sort of functionality with tensor, its
most common use is for a quick look to check that the tensor looks
reasonable, rather than for quantitative comparisons. Given this, a
default auto-scaling will give the most commonly desired behaviour with
minimal effort, but can easily be overriden when quantitation is
required.
How about the accepting the scale argument value 'auto'. I prefer it
not to be the default so that user knows that they will get a
diffusion rate of 1.8e6 s^-1 per Angstrom. And I would recommend that
all figures of the diffusion tensor be labelled with the diffusion
rate per Angstrom. Hence the auto-scaling should report the scale
value in an obvious way.
An important question is how would you define 'auto'? A number of
algorithms would be required. Would you loop over all atoms of all
residues and pick the atom furthest away from the centre of mass?
Probably the easiest would be to scale relative to the molecular mass,
though a maximum dimension as you suggest, or a radius of gyration might
be more robust. I can do a bit of playing to see what works. Tensor
seems to do a good job of it, so it cant be too hard.
How
would you handle multi-domain systems? If there are multiple tensors,
then they will have to be all scaled by the same amount. Or what
about complexes?
For these rarer and more complicated systems, the user might have to
take some responsibility for getting a good result.
Chris
Re: PyMOL is now supported.