Hello Edward, I understand that the strict convergence criteria are neccessary. When I set the max_iter parameter to just 30, I get very similar results as to when I reduce the opt_tol. I guess the model is just not converged.
You were quite right, the negative te values are very close to zero, their error is many orders of magnitude larger than their own value. The avergae hetNOE values is around 0.7. The measurements were taken at 600 and 700 MHz. I haven't taken them myself, my work concerns the theoretical side of things. The protein is not a membrane protein, it is in solution and the exact molecule given in the crystal structure I used as input for relax was measured (no tag or similar). The local spectrometer does unfortunately not allow for fancy temperature control methods. I feel relatively certain that the set up was in general correct. I tried both your python script and the GUI. The results I reported were taken from the txt-files in the final folder. As for the results of the previous runs. I am not sure how to access them or to read them. They are all zipped xml files where I cannot find a clear results section. Is there a certain program/command I should use? I wanted to use your visualization script xh_vector_dist.py, but I am not sure whoch file should be the input for 'select.read(file=pardir+sep+'rates.txt', change_all=True, res_num_col=2)' I visualized the tensor.pdb which is printed in the final directory. It is huge compared to the protein. It has the shape of a eight of a spheroid. Maybe just a simple question regarding the input. The R1 and R2 input files contained values in Hz (not radian). Is that correct? Regards, Johannes 2016-08-08 16:49 GMT+02:00 Edward d'Auvergne <edw...@nmr-relax.com>: > Hi Johannes, > > Sorry for the late response. I've been quite busy in the last two > months. Please see below: > > > On 8 August 2016 at 13:25, Johannes Dietschreit <dietschr...@gmail.com> > wrote: > > Hi, > > > > thank you so much for your quick and long response. I did not change the > > hard coded value, I decided to leave that one alone, but I followed the > > example of the test_suite and set the convergence criterion to 1e-7. This > > seems still strict, but now the calculation converges and final results > are > > written. > > The "strict" values are quite important for making sure there are no > strange results. You'll see that in: > > d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR > dynamic models I. Minimisation algorithms and their performance within > the model-free and Brownian rotational diffusion spaces. J. Biomol. > NMR, 40(2), 107-119. (http://dx.doi.org/10.1007/s10858-007-9214-2) > > Reading this paper is essential for understanding these cut-off > values, and why these high precision values are important. However if > there is a problem with the input data, then you will see problems. > In your example, I am not sure why this is not stopping after 30 > iterations. If you have a look at the automated protocol: > > http://www.nmr-relax.com/api/4.0/auto_analyses.dauvergne_ > protocol-module.html > http://www.nmr-relax.com/api/4.0/auto_analyses.dauvergne_ > protocol.dAuvergne_protocol-class.html > > you will see the max_iter parameter. Ok, I see that this is set to 30 > in the GUI, but it is not set in the sample script. If you require > more than 30 iterations of the global protocol ( > http://www.nmr-relax.com/manual/Model_free_analysis_in_reverse.html ), > then this is an indication that the input data is problematic. > > > > My question now regards the results in the folder "final". My analysis > was > > performed regarding the classical model free ansatz but I allowed for all > > possible diffusion models. The protein I am dealing with is rather stiff, > > it contains a beta barrel and some flexible loops. However, all residues > > have a S^2 value of about 0.03. I expected the residues in the barrel to > > have much larger values. > > These values indicate a severe problem with the input data. What is > your average HetNOE value? At which field strengths did you measure? > > > > Also the t_e values are somewhat spurious. Most of them are unphisically > > small (~e-24 seconds), I had expected values in the range of ten to > hundred > > pico seconds. And there are a few t_e values with a negative sign. How is > > that possible? > > These should disappear through model selection. If te ~ 0, then the > simpler model without te will be selected, as these models should have > converged to the same result. Unless of course there is a major > failure of the whole analysis. For the te values with negative > values, what are there values? The lower quality cut-off values will > allow for very small minus te values. > > > > Is this a common error? Have I just made a mistake regarding the input? I > > attached my dauvergne_protocol.py file and some input data. > > Note that you cannot attach files when posting to a public mailing > list. This is to avoid major strain on the infrastructure. For > sharing files, please create a support request and attach the files > there ( http://gna.org/support/?func=additem&group=relax ). Be > careful though, as this is public and anyone will be able to access > your data. From the ridiculously large number of iterations of the > global algorithm and the non-physical S2 and te values, I can only > guess that there is a major problem with the input data. This could > either be due to how the analysis was set up, or how the data was > measured. Looking at > http://www.nmr-relax.com/manual/Temperature_control_and_calibration.html > , how did you perform temperature calibration and temperature control? > Do you see any warnings at the start of your log files? When running > relax, it is extremely important to check every warning at the start > to make sure that the results are reasonable. > > Also, if you look at the local tm model results, are the S2 and te > values similar to the final results? Or similar to the spherical > diffusion model? > > Also note that for a rigid beta barrel with flexible loops that the > backbone NH bond vector distribution will not be isotropic. I suggest > running the sample_scripts/xh_vector_dist.py script and visualising > the results. This is the script I created for Figure 4 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) > > Also, are you in a membrane system? If so, then you actually have two > diffusion tensors. The protein will spin inside the micelle/bicelle, > and then the whole system will have an independent global ellipsoidal > or spheroidal diffusion. I am unaware of anyone to date who has > derived the equations for such a system. Nevertheless, the strange > results you see are unlikely to be due to this modelling deficiency. > > Regards, > > Edward > _______________________________________________ relax (http://www.nmr-relax.com) This is the relax-users mailing list relax-users@gna.org 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-users
