Re: problem loading R1, R2 and nOe files for modelfree
Hi Ravi, Which version of relax are you using and which operating system are you running it on? Are you using the newest 3.2.2 version? Could you run 'relax -i' or, in the GUI, select the 'Tools-System information' menu item and the copy and paste the output into the body of a new email? Note that due to too much abuse of the system, attachments are no longer allowed on the relax mailing lists to avoid excessive strain on the open source infrastructure. All attachments are filtered, i.e. stripped from the message. See your archived message at http://article.gmane.org/gmane.science.nmr.relax.user/1695 for example. If you would like to share a file in such a situation, the best solution is to create a bug report using the link https://gna.org/bugs/?func=additemgroup=relax and to attach the file there. There are also support requests for non-bug related issues where files can also be attached (https://gna.org/support/?func=additemgroup=relax). Cheers, Edward On 27 June 2014 00:50, Ravi Pratap Barnwal ravi1...@gmail.com wrote: Hi, I am encountering problem while loading R1,R2 and nOe values calculated for two different fields into relax (precompiled version 3.2.2) for mac 10.8.5. When i try to add the data and try to change the free format file settings for respective file, it is kind of hanging and does not proceed anywhere. My files are in the following format, 3 GLY 1.814795542 0.090739777 4 SER 2.267446596 0.11337233 5 MET 2.621469801 0.13107349 6 ASP 2.751590887 0.137579544 Am i doing something wrong? I am following the protocol from this link, http://www.nmr-relax.com/manual/d_Auvergne_protocol_GUI_mode_data_pipe_initialisation.html I have attached a screen-shot for the problem. Any help is highly appreciated. regards ravi ___ 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 ___ 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
Re: Side chain dynamics
Hi Mengjun, Sorry for not getting back to you earlier. The papers you mention are: Oscar Millet and Anthony Mittermaier and David Baker and Lewis E. Kay (2003) The Effects of Mutations on Motions of Side-chains in Protein L Studied by 2H NMR Dynamics and Scalar Couplings. Journal of Molecular Biology, 329, 551-563 (http://dx.doi.org/10.1016/S0022-2836(03)00471-6) Eric Johnson and Walter J. Chazin and Mark Rance (2006) Effects of Calcium Binding on the Side-chain Methyl Dynamics of Calbindin D9k: A 2H NMR Relaxation Study. Journal of Molecular Biology, 357, 1237-1252 (http://dx.doi.org/10.1016/j.jmb.2006.01.031). The key here is that if these papers use the standard model-free theory with no modifications, which I think they do though you need to carefully check this, then this will work fine in relax where the standard theory is implemented. If the theory has been modified and hence is not the original equations, then it might be time to learn some computer programming ;) But if you look at equation (1) of the Rance paper, you will see that they measure 5 quadrupolar relaxation rates. If you have measured these, then you cannot use them in relax yet. There is currently no support in relax for quadrapolar relaxation rates. However it would be relatively easy to add. Note that all of the required infrastructure is in place. The model-free spectral density functions are all located in lib/spectral_densities. Also present are all model-free models, the highest quality optimisation infrastructure, powerful model selection techniques, model elimination infrastructure for failed models, the automated protocol that I came up with for solving the convoluted diffusion tensor vs. internal motion problem (though I don't know if this protocol would work if you only measured Me deuterium relaxation and do not combine it with backbone relaxation), powerfully data visualisation infrastructure, etc. The only work required is: 1) Extending the list of input relaxation data types in the relax_data.read user function. This is trivial - the hard part is deciding what to call these rates. 2) Adding relevant functions to the relax library lib/auto_relaxation/ modules to handle the additional rates. This is where the real work is. 3) Added the quadrupolar coupling constant Q (or e2qQ/h) as a model-free parameter that can be set via the value.set user function (see http://www.nmr-relax.com/api/3.2/specific_analyses.model_free.parameter_object-pysrc.html#Model_free_params.__init__). This requires one line of code. The hardest part would be obtaining published test data. That would only require emailing one of the corresponding authors and explaining what you would like to do, and the data in the Rance paper, especially figure 4 appears to be the best for this. The test data is then turned into a system test - this is a mini-analysis used to check that everything runs as expected. I could do this in ~2 weeks of solid work, but I don't have the time to allocate to such a task. This would also require the gradients and Hessians (1st and 2nd partial derivatives) to be calculated for the relaxation rates: http://www.nmr-relax.com/manual/Ri_theta_gradients.html http://www.nmr-relax.com/manual/Ri_theta_Hessians.html http://www.nmr-relax.com/manual/Ri_prime_theta_gradients.html http://www.nmr-relax.com/manual/Ri_prime_theta_Hessians.html This is rather trivial in Maxima, wxMaxima, Mathematica, or other symbolic algebra systems. However it is so simple that it could be derived by hand in a few minutes. If you would be interested in adding this to relax, you could then mention this in your paper - that you implemented quadrapolar relaxation rates in relax. This could then be added to the reference list in the citations chapter (http://www.nmr-relax.com/manual/Model_free_analysis_references.html) as well as a new section for citations at the start of the model-free chapter (http://www.nmr-relax.com/manual/Model_free_analysis.html) so that people using relax with such rates should then cite your paper. Regards, Edward On 12 June 2014 14:35, mengjun@mailbox.tu-berlin.de wrote: Hi Edward, Thank you so much. I would like to replicate the analysis in the publication The Effects of Mutations on Motions of Side-chains in Protein L Studied by 2H NMR Dynamics and Scalar Couplings and Effects of Calcium Binding on the Side-chain Methyl Dynamics of Calbindin D9k: A 2H NMR Relaxation Study , where models LS2 and LS3 were used for fitting. Best regards, Mengjun Quoting mengjun@mailbox.tu-berlin.de: Hi Edward, There is some script in Relax software which can be used for analyzing side chain dynamics (R1 and R1r of deuterium in CH2D isotopomers)? Thank you. Best regards, Mengjun ___ 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,
Re: R1rho RD analysis
Dear Andrew. Thank you for your valuable comments! It becomes clear for me, that I should sit down with theory to grasp, why I actually do: 1) matrix exponential 2) matrix power in all the numerical models. Then I would be able to select which best method to apply. For: ns_cpmg_2site_star.py = 2x2 matrix, ns_mmq_2site.py = 2x2 matrix. I should write up the closed form. So, the corresponding characteristic polynomial is quadratic so the eigenvalues can be expressed in closed form in terms of the matrix elements. M = [ [a b], [b c] ] discriminant: D = sqrt(a**2 + 4b**2 - 2ac + c**2= Eigenvalues, lamda1, lamda2. l1 = 0.5*(a + c - D), l2 = 0.5*(a + c + D) Define M: M = V^-1 * [ [l1 0], [0 l2] ] * V V11 = (a - c - D)/2b V12 = 1 V21 = (a - c + D)/2b V22 = 1 V = [ [v11 v12], [v21 v22] ] For: ns_mmq_3site.py = 3x3 matrix, ns_cpmg_2site_3d.py = 7 X 7 matrix, ns_r1rho_2site.py = 6x6 matrix, ns_r1rho_3site.py = 9x9 matrix. I also asked on the Scipy mailing. (Their mail server link is not working, so here a short reference). ### Charles R Harris: Are you solving a differential equation? If so, an explicit solution may not be the best way to go. ### Eric Moore: There are expansions for the matrix exponential in terms of Chebyshev and Laguerre matrix polynomials. (The first of which does get use a bit for calculating propagators in NMR, for instance, spinevolution and simpson both can/do use it.) However, your matrices seem quite small, even if there are a large number of them, so it may be very difficult to speed this up by much. ### Stephan Hoyer: My experience with solving very similar quantum mechanics problems is that using an ODE integrator like zofe (via scipy.integrate.ode) can be much faster than solving the propagator exactly. ### Michael Kreim: I agree with Stephan and Charles that it my be worth trying to solve the ODE using any suitable ODE solver. In Matlab I made the experience that it is very problem depended if the ode solver or the matrix exponential method is faster. Also you can have a look at expokit which is a very good implementation of matrix exponential solvers (using Krylov subspace methods): http://www.maths.uq.edu.au/expokit/. Unfortunately there is no Python implementation but maybe you can call the fortran or C++ version from python. And you can have a look at operator splitting methods. If it is possible to split your problem in two (or more) sub problems, then these methods often speed up the numerical computations. The problem usually lies in finding a suitable splitting. The implementation is not very complicated, if you have numerical methods for the sub problems. ## So here, it sounds like I should figure out to do a ordinary differential equation, for magnetisation evolution. Would this be the right way to go? I am very happy for any comments. Best Troels 2014-06-26 13:46 GMT+02:00 Andrew Baldwin andrew.bald...@chem.ox.ac.uk: I am looking for a solution I can apply to all numerical models :-) The 'close formed' solutions I describe are eigenvalue decompositions. So that's just a slightly smart way of doing that rather than asking a minimiser to do it for you. No general closed form solutions exist for the eigenvalues (hence eigenvectors) I think above 3x3. Incidentally, the theoretical treatments of the R1rho (needing minimum of 6x6) relies on being able to approximate the smallest real eigenvalues of the six. So the theoretical R1rho theoretical treatments are in essence methods for approximating one of the six eigenvalues (which is hard enough). Getting the other five is far from straightforward and might not even be possible. This is also why a general CEST expression isn't going to happen: to get a complete treatment of the CEST experiment you need all six eigenvalues, and no-one has (yet) come up with a way of getting them all. If the matrix is sparse then you can get some speedups. But in my hands at least, those algorithms only tend to work well on huge matrices that are really sparse (eg100x100 with not much over 10% of elements having entries). On 26/06/2014 09:44, Troels Emtekær Linnet wrote: Hi Andrew. I am looking for a solution I can apply to all numerical models :-) The current implementation, is the eigenvalue decomposition. http://svn.gna.org/svn/relax/branches/disp_spin_speed/lib/dispersion/matrix_exponential.py It the bottom, I provide the profiling for NS R1rho 2site. You will see, that using the eig function, takes 50% of the time: That is i little sad, that the reason why Numerical solutions is so slow, is numpy.linalg.eig(). They differ a little in matrix size. http://svn.gna.org/svn/relax/branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_3d.py 7 X 7 matrix. http://svn.gna.org/svn/relax/branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py 2x2 matrix http://svn.gna.org/svn/relax/branches/disp_spin_speed/lib/dispersion/ns_mmq_2site.py 2x2 matrix.
Re: R1rho RD analysis
Hi Troels, For reference for the relax users reading this, the abbreviations Troels used for the relaxation dispersion models can be decoded using the relax wiki page http://wiki.nmr-relax.com/Category:Relaxation_dispersion. Just a little heads-up. Within a week or two, we should be able to release a new version of relax. This update has focused on speed, recasting the data to higher dimensional numpy arrays, and moving the multiple dot operations out of the for loops. So we have quite much better speed now. :-) Troels, you just gave away the surprise of the relax 3.2.3 or 3.2.4 release! Note that relax was not particularly slow before these changes, but that you have taken far greater advantage of the BLAS/LAPACK libraries behind the numpy functions to make the dispersion models in relax super fast, especially when spins are clustered. It's a pity we don't have the means to compare the optimisation target function speed between relax and other software to show how much faster relax now is, as the relax advantage of having a grid search to find the initial non-biased position for optimisation (a very important technique for optimisation that a number of other dispersion software do not provide) will give many relax users the incorrect impression that relax is slower than the other software. Though if relax is used on a cluster with OpenMPI, this is a non-issue. But we still have a bottleneck with numerical models, where doing the matrix exponential via eigenvalue decomposition is slow! Do any of you any experience with a faster method for doing matrix exponential? These initial results shows that if you are going to use the R1rho 2site model, you can expect: -R1rho 100 single spins analysis: DPL94: 23.525+/-0.409 - 1.138+/-0.035, 20.676x faster. TP02: 20.191+/-0.375 - 1.238+/-0.020, 16.308x faster. TAP03: 31.993+/-0.235 - 1.956+/-0.025, 16.353x faster. MP05: 24.892+/-0.354 - 1.431+/-0.014, 17.395x faster. NS R1rho 2-site: 245.961+/-2.637 - 36.308+/-0.458, 6.774x faster. Cluster of 100 spins analysis: DPL94: 23.872+/-0.505 - 0.145+/-0.002, 164.180x faster. TP02: 20.445+/-0.411 - 0.172+/-0.004, 118.662x faster. TAP03: 32.212+/-0.234 - 0.294+/-0.002, 109.714x faster. MP05: 25.013+/-0.362 - 0.188+/-0.003, 132.834x faster. NS R1rho 2-site: 246.024+/-3.724 - 33.119+/-0.320, 7.428x faster. -CPMG 100 single spins analysis: CR72:2.615+/-0.018 - 1.614+/-0.024, 1.621x faster. CR72 full: 2.678+/-0.020 - 1.839+/-0.165, 1.456x faster. IT99:1.837+/-0.030 - 0.881+/-0.010, 2.086x faster. TSMFK01: 1.665+/-0.049 - 0.742+/-0.007, 2.243x faster. B14: 5.851+/-0.133 - 3.978+/-0.049, 1.471x faster. B14 full:5.789+/-0.102 - 4.065+/-0.059, 1.424x faster. NS CPMG 2-site expanded: 8.225+/-0.196 - 4.140+/-0.062, 1.987x faster. NS CPMG 2-site 3D: 240.027+/-3.182 - 45.056+/-0.584, 5.327x faster. NS CPMG 2-site 3D full:240.910+/-4.882 - 45.230+/-0.540, 5.326x faster. NS CPMG 2-site star: 186.480+/-2.299 - 36.400+/-0.397, 5.123x faster. NS CPMG 2-site star full: 187.111+/-2.791 - 36.745+/-0.689, 5.092x faster. Cluster of 100 spins analysis: CR72:2.610+/-0.035 - 0.118+/-0.001, 22.138x faster. CR72 full: 2.674+/-0.021 - 0.122+/-0.001, 21.882x faster. IT99:0.018+/-0.000 - 0.009+/-0.000, 2.044x faster. IT99: Cluster of only 1 spin analysis, since v. 3.2.2 had a bug with clustering analysis.: TSMFK01: 1.691+/-0.091 - 0.039+/-0.000, 43.369x faster. B14: 5.891+/-0.127 - 0.523+/-0.004, 11.264x faster. B14 full:5.818+/-0.127 - 0.515+/-0.021, 11.295x faster. NS CPMG 2-site expanded: 8.167+/-0.159 - 0.702+/-0.008, 11.638x faster. NS CPMG 2-site 3D: 238.717+/-3.025 - 41.380+/-0.950, 5.769x faster. NS CPMG 2-site 3D full:507.411+/-803.089 - 41.852+/-1.317, 12.124x faster. NS CPMG 2-site star: 187.004+/-1.935 - 30.823+/-0.371, 6.067x faster. NS CPMG 2-site star full: 187.852+/-2.698 - 30.882+/-0.671, 6.083x faster. There are a number of ways of computing the matrix exponential. Avoiding eigenvalue decomposition is the essential key. The Pade approximation is probably the best, followed by one of the Taylor series expansion approximations. As I mentioned in a different post to the relax-devel mailing list, this was used earlier in relax via Scipy. But I removed this and wrote my own algorithm using eigenvalue decomposition as the Scipy
Re: problem loading R1, R2 and nOe files for modelfree
Hi Edward, I am using version 3.2.2 on mac 10.8.5 (mountain lion). I am using GUI mode to run the relax. Here is the report from Tools- System Information relax 3.2.2 Molecular dynamics by NMR data analysis Copyright (C) 2001-2006 Edward d'Auvergne Copyright (C) 2006-2014 the relax development team This is free software which you are welcome to modify and redistribute under the conditions of the GNU General Public License (GPL). This program, including all modules, is licensed under the GPL and comes with absolutely no warranty. For details type 'GPL' within the relax prompt. Assistance in using the relax prompt and scripting interface can be accessed by typing 'help' within the prompt. Processor fabric: Uni-processor. relax sys_info() Hardware information: Machine: x86_64 Processor: i386 Processor name: Intel(R) Xeon(R) CPU W3530 @ 2.80GHz Endianness: little Total RAM size: 2048.0 Mb Total swap size: 11264.0 Mb Operating system information: System: Darwin Release: 12.5.0 Version: Darwin Kernel Version 12.5.0: Sun Sep 29 13:33:47 PDT 2013; root:xnu-2050.48.12~1/RELEASE_X86_64 Mac version: 10.8.5 (, , ) x86_64 Distribution: Full platform string:Darwin-12.5.0-x86_64-i386-64bit Python information: Architecture:64bit Python version: 2.7.2 Python branch: Python build:default, Apr 5 2012 18:46:15 Python compiler: GCC 4.2.1 (Apple Inc. build 5666) (dot 3) Libc version: Python implementation: CPython Python revision: Python executable: /Applications/relax.app/Contents/MacOS/python Python flags:sys.flags(debug=0, py3k_warning=0, division_warning=0, division_new=0, inspect=0, interactive=0, optimize=1, dont_write_bytecode=0, no_user_site=0, no_site=0, ignore_environment=0, tabcheck=0, verbose=0, unicode=0, bytes_warning=0) Python float info: sys.float_info(max=1.7976931348623157e+308, max_exp=1024, max_10_exp=308, min=2.2250738585072014e-308, min_exp=-1021, min_10_exp=-307, dig=15, mant_dig=53, epsilon=2.220446049250313e-16, radix=2, rounds=1) Python module path: ['/Applications/relax.app/Contents/Resources/lib/python27.zip', '/Applications/relax.app/Contents/Resources/lib/python2.7', '/Applications/relax.app/Contents/Resources/lib/python2.7/plat-darwin', '/Applications/relax.app/Contents/Resources/lib/python2.7/plat-mac', '/Applications/relax.app/Contents/Resources/lib/python2.7/plat-mac/lib-scriptpackages', '/Applications/relax.app/Contents/Resources/lib/python2.7/lib-tk', '/Applications/relax.app/Contents/Resources/lib/python2.7/lib-old', '/Applications/relax.app/Contents/Resources/lib/python2.7/lib-dynload', '/Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip', '/Applications/relax.app/Contents/Resources/lib/python2.7/site-packages', '/Applications/relax.app/Contents/Resources/lib/python2.7/site-packages'] Python packages and modules (most are optional): Name InstalledVersion Path minfx True 1.0.6 /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/minfx bmrblibTrue 1.0.3 /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/bmrblib numpy True 1.6.1 /Applications/relax.app/Contents/Resources/lib/python2.7/numpy scipy True 0.10.1 /Applications/relax.app/Contents/Resources/lib/python2.7/scipy wxPython True 2.9.3.1 osx-cocoa (classic) /Applications/relax.app/Contents/Resources/lib/python2.7/wx matplotlib False mpi4py True 1.2.2 /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/mpi4py epydoc True 3.0.1 /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/epydoc optparse True 1.5.3 /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/optparse.pyc readline False profileTrue /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/profile.pyc bz2True /Applications/relax.app/Contents/Resources/lib/python2.7/lib-dynload/bz2.so gzip True /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/gzip.pyc io True /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/io.pyc xmlTrue 0.8.4 (internal) /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/xml/__init__.pyc xml.dom.minidomTrue /Applications/relax.app/Contents/Resources/lib/python2.7/site-packages.zip/xml/dom/minidom.pyc relax
