Re: Pulse miscalibration, R1A and R1B in the fitting_main_kex.py script. -- July 17, 2013 - 11:32

Hi,

There is one more 'trick' I am using to speed everything up.  The
initial grid search required prior to optimisation can be quite
expensive for the models with the R20A and R20B parameters (i.e. R2G
and R2E).  So I have created simpler or reduced versions of the models
whereby R20A = R20B = R20 in the target function (the relax library
code derived from the fitting_main_kex.py script is not touched).  In
the automated analysis presented to users, the simpler model is
optimised first.  Then instead of performing the expensive grid search
on the more complex model, the parameters of the simpler model are
taken as the starting point for the more complex model.  The R20A and
R20B values are initialised to the same R20 value, but these then
diverge.  Note that this has no effect on Monte Carlo simulation
speed.  This also is not used if the standard relax script interface
is used and the auto-analysis is avoided.  I will write an email once
all of the models are taken over into relax and all the reduced models
have been created.

Regards,

Edward


P. S.  The relax auto-analyses are simply large relax scripts bundled
with the program and presented to the user in a simple interface in
both the script/prompt UI and graphical user interface.



On 16 July 2013 18:13, Paul Schanda <paul.schanda@xxxxxx> wrote:
> Hi Edward,
>
> Sounds great.
> I will be very interested to see how much speed you gain, and also how the
> different fitting procedures (2D version, the Maple-derived numerical one,
> and the 3D version) compare.
> If all your optimizations bring an order of magnitude or so speed-up, it
> will be a big advantage for the Monte Carlo runs.
>
> best -
>
> paul
>
>
> On 16.07.13 17:40, Edward d'Auvergne wrote:
> > Cheers,
> >
> > I'll leave all the functionality intact then, but just default to
> > zeros.  As the R2eff calculation and matrix construction is in the
> > relax library and a clean separation between it and the target
> > function code (for optimisation), the relaxation dispersion
> > auto-analysis (to simplify the analysis for users), GUI, and the
> > specific analysis code, the functions can be directly imported into
> > simulation scripts and used as is.  I have now simplified and speed up
> > most of this code.  Most speed ups are due to a large reduction in the
> > number of multiplications per target function call, simply by
> > performing the multiplication at the highest loop level and storing
> > the value as a variable.  As well as by minimisation numpy matrix
> > creation and destruction.  Additional large speed ups in the analysis
> > of experimental data are due to a switch from the scipy optimisation
> > algorithms to those of minfx (http://gna.org/projects/minfx/).
> >
> > Regards,
> >
> > Edward
> >
> > On 16 July 2013 17:09, Dominique Marion (IBS) <Dominique.Marion@xxxxxx>
> > wrote:
> > > Hello,
> > >
> > > We have found useful to carry out some simulations with the same
> > > equations
> > > to figure out the influence of some parameters.  You are right, the train
> > > of
> > > 180º pulses can be miscalibrated and the rf coil can be highly
> > > inhomogeneous.  If the CPMG pulses differ from 180º, then the spins spend
> > > some fraction of time along the z-axis. In the case of slow tumbling
> > > proteins, the difference between R1 and R2 may become quite large.
> > > Whether
> > > the small oscillations observed for some CPMG profiles originate from
> > > imperfect 180º remains to be demonstrated.
> > >
> > > Cheers,
> > >
> > > Dominique
> > >
> > >
> > >
> > > On 16/7/13 16:35 , Edward d'Auvergne wrote:
> > >
> > > Hello,
> > >
> > > I'm now copying in the function returned by MakeFunction when a==1
> > > into relax.  I have noticed that this has the ability to handle
> > > miscalibrated pulses via the R180_3Dx() function, however the pi-pulse
> > > angle is 180 + a degrees, where a is hardcoded to 0.0 degrees.  Is
> > > this a planned future feature?
> > >
> > > It is also not clear how the R1 relaxation rates for states G and E
> > > (or A and B) are used.  Is the aim to have R1G (per spin), R1E (per
> > > spin), and the pulse angle (all per magnetic field strength) as
> > > optimisable parameters of the model?  For the current implementation,
> > > I will just set these all to hardcoded values of zero, as neither
> > > relax nor the fitting_main_kex.py script appear to use these.
> > >
> > > Cheers,
> > >
> > > Edward
> > >
> > >
> > >
> > > _______________________________________________
> > > relax (http://www.nmr-relax.com)
> > >
> > > This is the relax-devel mailing list
> > > relax-devel@xxxxxxx
> > >
> > > To unsubscribe from this list, get a password
> > > reminder, or change your subscription options,
> > > visit the list information page at
> > > https://mail.gna.org/listinfo/relax-devel
>
>
> --
> Paul Schanda, Ph.D.
> Biomolecular NMR group
> Institut de Biologie Structurale Jean-Pierre Ebel (IBS)
> 41, rue Jules Horowitz
> F-38027 Grenoble
> France
> +33 438 78 95 55
> paul.schanda@xxxxxx
> http://www.ibs.fr/groups/biomolecular-nmr-spectroscopy?lang=en