Hi, You should base the clustering on the dynamics, not the selected model! For example the 'CR72' analytic model and the 'NS 2-site expanded' numeric model can both describe exactly the same motional process. Clustering is a much more manual operation and it should not be automated. It is based on human logic and is highly subjective. For example it could be decided that one analysis is performed whereby one motional process is assumed, i.e. one kex value for all exchanging spins. Or it could be decided that there are two motional processes, so two clusters are created, each having their own kex. Some spins with bizarre dynamics may be left out as 'free spins' and not used in the cluster. An automated protocol could be devised, but I would assume that that is not your goal. Though if you just want all spins with Rex to be in one cluster, you could just use all spins where spin.model is not set to 'No Rex'. For the instructions, the sel_residues structure is not very elegant. I would suggest you change this to sel_spins and have it be a list of spin IDs. The major problem is that you are accessing private structures such as spin._res_name. One should never rely on private structures, they may not exist in the next version of relax. There is a option for the spin_loop() generator method to return this information. However you just need spin IDs which are already returned, as the relax_disp.cluster user function only works with spin IDs. The residue level should not be considered. As for the GUI bug, could you submit a bug report with all the details? Cheers, Edward On 28 August 2013 19:11, Troels Emtekær Linnet <tlinnet@xxxxxxxxx> wrote:
I think I got it now. :-) Here is now my entry how to do the clustering. http://wiki.nmr-relax.com/Tutorial_for_Relaxation_dispersion_analysis_cpmg_fixed_time_recorded_on_varian_as_fid_interleaved#Execute_a_clustering_analysis There is a bug, which prevent it do be done through the GUI, so only the script works. Best Troels Troels Emtekær Linnet 2013/8/28 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:Hi, The original pipe actually is called something like 'origin - relax_disp (Wed Aug 28 16:54:13 2013)'. But you should save the program state yourself prior to the first GUI analysis. Otherwise the steps are more complicated from the perspective of the students you are targeting and the logic is more difficult to follow. Also, the 'Previous run directory' should match the 'Results directory' from the previous analysis, and not the "NS 2-site expanded" directory. Regards, Edward On 28 August 2013 16:43, Troels Emtekær Linnet <tlinnet@xxxxxxxxx> wrote:Hi Edward. Allright. So, I should create a copy from the "base pipe" to a new pipe "cluster". The "base pipe" contains the spectra intensities, but nothing else. And the I cluster atoms in this pipe. Then start an auto analysis. Point to "NS 2-site expanded" in "Precious run directory". And then select model "R2eff", "No Rex" and "NS 2-site expanded" Or... :-) Best Troels Troels Emtekær Linnet 2013/8/28 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:No problems! For the clustered analysis, you need to start again from the start. You do not load the final state, this will likely be fatal for the clustered analysis (actually, I don't know what will happen, but don't do it). The auto-analysis is designed to take the pre-run directory name and load the results files for each model itself (not the state file). Each results file will be loaded into a temporary data pipe and the initial parameter values copied from that. You should not load a state or results file yourself for the clustered analysis. Regards, Edward P. S. One exception is that you could have a state file with all data loaded but prior to clicking the 'execute' button. This is essentially a short cut to manually loading everything again. On 28 August 2013 15:36, Troels Emtekær Linnet <tlinnet@xxxxxxxxx> wrote:Hi Edward. Thanks for all the tips! I have a question more. When I open the final_state.bz2, I have several pipes from the different models. It is in the pipe "final". Should I create a new pipe before doing the cluster analysis? For example a copy from "final" or from "NS 2-site expanded". And then do the clustering in that pipe, before an Autoanalysis. Best Troels Troels Emtekær Linnet 2013/8/28 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:Hi, We should add many more of these details to the dispersion chapter of the relax manual! So for the dispersion auto-analysis, there is the 'pre_run_dir' argument. To use clustering, you just define the cluster with the relax_disp.cluster user function, and then set 'pre_run_dir' to the directory of non-clustered results. That's it. The auto-analysis takes care of the rest. Regards, Edward On 28 August 2013 13:43, Troels Emtekær Linnet <tlinnet@xxxxxxxxx> wrote:Hi Edward. Allright, I see your point. :-) Now I am trying to figure out how to do a clustering analysis. I have added to the wiki: http://wiki.nmr-relax.com/Tutorial_for_Relaxation_dispersion_analysis_cpmg_fixed_time_recorded_on_varian_as_fid_interleaved#Inspecting_results_from_the_relax_analysis But I am unsure if I made the clustering correct. It is related to copying of the parameters, and running the minimization with the clustered analysis residues. I looked up in the relax disp manual: "relax disp.parametercopy". (I think the prompt example i wrong?) I can't seem to find a section how to make the minimization to run with a clustering or residues. :-) Thanks :-) Troels Emtekær Linnet 2013/8/28 Edward d'Auvergne <edward@xxxxxxxxxxxxx>:Hi, This model selection technique is called Akaike's Information Criteria (AIC). See http://www.nmr-relax.com/refs.html#dAuvergneGooley03 for details on this. How would you produce a plot in Grace of the models selected? I don't know how this would work for the dispersion models. In a model-free analysis I have deliberately taken the decision to not create such a plot. In model-free, relax supports models m0 to m9, so these could be plotted as 0 to 9 on the Y-axis. But this is not done and will never be implemented. The reason is because the information about which model is selected is worthless. What is important is the dynamics extracted. This is especially the case of nested models. Just because a parameter is missing from a model, this does not mean that it is not present. For example there are model-free models without the Rex parameter. If these are selected, this does not mean that Rex is not present. It means that Rex is statistically insignificant in the current data set. Model differences are an academic question and is not interesting in practice. Sometimes the AIC value for two nested models is very similar and that means that the parameter different between the two is sitting directly on the boarder of statistical significance/insignificant. You could pick either model and it should not make a statistical difference in the dynamics picture. It does not relate to the dynamics, just the judgement of statistically significance for parameters. Therefore for practical reasons, for model-free at least, the interesting plots, text files, and PyMOL/MOLMOL macros are of the parameter values themselves. I think is would be much more of interest to implement PyMOL/MOLMOL macros for the dispersion parameters so you can see the dynamics directly on the 3D structure. This is performed in NESSY and needs to be added to relax. Also note that when you are analysing data, you would probably limit the number of models to 2-3. For example if you know that all residues are experiencing slow exchange, the LM63 fast exchange model does not need to be used. It is interesting to see that sometimes the analytic models are selected and sometimes the numeric models. But this is an academic curiosity, it is probably not a practical question anyone analysing real dispersion data is interested in. The way an analysis would normally be performed is to first decide if the analytic or numeric approach is to be used. For the analytic approach with slow exchange, you only need the 'No Rex' and 'CR72' models. You could add the 'IT99' model if you can see that pA >> pB in the spectra, i.e. the pB peak is tiny. If you take the numeric approach, then the 'No Rex' and 'NS 2-site expanded' models can be used. Once you perform an initial analysis of all residues separately, you can then look at the dynamics parameter values and judge which spins to cluster together to have the same model of dynamics, then re-perform the analysis. Regards, Edward On 28 August 2013 12:02, Troels Emtekær Linnet <tlinnet@xxxxxxxxx> wrote:Hi. After a AIK selection between models CR72/ IT99/ LM63/ NS 2-site expanded/ I get this selected list of models. Would it be possible to add to the grace images in the final directory, the selected model and the residue number? 3 LM63 4 CR72 5 NS 2-site expanded 6 CR72 7 LM63 8 No Rex 9 NS 2-site expanded 10 CR72 11 LM63 12 CR72 13 IT99 14 NS 2-site expanded 15 NS 2-site expanded 16 LM63 17 CR72 18 CR72 20 NS 2-site expanded 21 NS 2-site expanded 22 CR72 23 LM63 24 LM63 25 CR72 26 IT99 27 NS 2-site expanded 28 NS 2-site expanded 29 CR72 30 NS 2-site expanded 31 CR72 32 CR72 33 IT99 34 IT99 35 IT99 36 IT99 37 IT99 38 NS 2-site expanded 39 LM63 40 No Rex 41 CR72 42 No Rex 43 NS 2-site expanded 45 IT99 46 CR72 47 CR72 48 IT99 49 NS 2-site expanded 50 LM63 51 NS 2-site expanded 52 CR72 53 NS 2-site expanded 54 NS 2-site expanded 55 No Rex 56 IT99 57 LM63 58 LM63 59 NS 2-site expanded 60 CR72 61 IT99 62 NS 2-site expanded 63 NS 2-site expanded 64 CR72 65 IT99 66 No Rex 67 CR72 68 No Rex 69 NS 2-site expanded 70 NS 2-site expanded 71 NS 2-site expanded 72 NS 2-site expanded 73 LM63 74 LM63 75 NS 2-site expanded 76 NS 2-site expanded 77 IT99 78 CR72 79 No Rex 80 LM63 81 NS 2-site expanded 82 IT99 83 IT99 84 NS 2-site expanded 85 LM63 86 IT99 Troels Emtekær Linnet _______________________________________________ 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