Hi, I've answered below: On 24 February 2010 20:53, Carl Diehl <Carl.Diehl@xxxxxxxxx> wrote:
Hi. Well, the mail server said that it tried sending three times before giving up. Anyway, sofar I've used relax 1.2.XX both for fitting and modelfree-selection. Is there a any major difference in the fitting routines between the 1.2 and the 1.3 line?
No, this should be exactly the same! If you use the full_analysis.py script in both lines, you should expect the same. That should be a useful test for bugs actually!
So far I've used a modified full_analysis.py script (in relax 1.2.14) for analysis which works well and is able to accurately reproduce older modelfree-fits. I haven't made the transition into 1.3.4, since I would have to change from runs into pipes.
If you using just the full_analysis.py script, switching to 1.3.4 (or the 1.3 line) is very easy. It will require the modification of ~20 lines of variables at the start of the script. For implementing this, I highly recommend using the most up to date copy of the 1.3 line so that your changes can eventually be useful for others. If you make this work for the frozen 1.2 line (no new features will be added there, it has been in maintenance mode for almost 3 years), there will be no path to implement it to the 1.3 line.
My basic idea is to rerun the my analysis script, only this time using the Palmer model-selection instead of AIC, which I've been using sofar.
If you've been using full_analysis.py, this will not be possible. The reason is because it implements a new concept - it searches for the solution within universal set of all combinations of model-free models and diffusion models (the universal solution). It uses a very specific target function for this. This function is not compatible with the Mandel et al. model selection. This concept is described in: d'Auvergne E. J., Gooley P. R. (2007). Set theory formulation of the model-free problem and the diffusion seeded model-free paradigm. Mol. Biosyst., 3(7), 483-494. (http://dx.doi.org/10.1039/b702202f) The implementation and the new protocol that full_analysis.py implements is given in: d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR dynamic models II. A new methodology for the dual optimisation of the model-free parameters and the Brownian rotational diffusion tensor. J. Biomol. NMR, 40(2), 121-133. (http://dx.doi.org/10.1007/s10858-007-9213-3) This new protocol specifically starts with the local_tm models (the Mandel et al. model selection cannot handle these!!!). These are optimised and the best for each spin selected. Then the local_tm parameter is removed and the diffusion tensor optimised on top of the fixed model-free parameter values. Then the diffusion tensor is fixed and the normal model-free models are optimised. Model elimination is performed (this is missing from the Mandel et al. protocol so you may want to remove this?). Then AIC model selection. And finally diffusion tensor optimisation. This is repeated until convergence (~6 to 15 iterations). The Mandel et al. protocol starts with a diffusion tensor. The model-free models are optimised (my thesis extends the protocol for models m6-m9, see http://eprints.infodiv.unimelb.edu.au/archive/00002799/ or search on Amazon). Model selection is performed. Then the diffusion tensor is optimised. This tensor is taken as a starting point, and the procedure is repeated ~3 times. To implement this protocol, you should look at the mf_multimodel.py and modsel.py scripts. This will give you most, but not all of the tools needed to implement the Mandel et al. protocol.
From an old fit using Modelfree4 and Fast-MF (1field only), I know that there are only small differences in S2 compared to the 2field-data fitted using relax.
For highly mobile systems, the differences are more pronounced. You'll should see most differences in loops or other mobile regions - the interesting parts in terms of dynamics and function.
I can easily redo my 2field fits using Modelfree4 and Fast-MF, although I would prefer to use relax. Time-wise, the full_analysis scripts seems to be much faster than the Fast-MF way.
It sounds like you will learn quite a lot about the model-free protocols if you start digging into these scripts ;)
I'll take a look at the repository and try to see if I can get it to work.
Cheers, Edward