Hi, Is this the paper at http://dx.doi.org/10.1073%2Fpnas.0507370102 ? If so, do you know if the data for methyl group protons needs to be treated differently? Or does the TROSY data need a special treatment? Are a different set of Bloch-McConnell equations used? For the data analysis, the key sentences are:
"Chemical shift differences were initially estimated from spectra recorded at 0.5°C and then optimized based on the extracted exchange parameters from fits of the dispersion data. This optimization is accomplished in an iterative manner whereby the fitted exchange parameters are used, along with estimated shift differences, to obtain the true shift differences from the eigenvalues of the free precession evolution matrix." It sounds like Remco used a numeric model. I do not know why he did not perform a standard optimisation - that is not explained in the paper. With relax we would perform a direct optimisation - first the standard grid search to find a rough initial position followed by the Nelder-Mead simplex optimisation (which requires no expensive gradients or Hessians to be derived or numerically approximated, http://en.wikipedia.org/wiki/Nelder%E2%80%93Mead_method). The low temperature shifts are only used as a starting estimate of dw. I think that with relax using a grid search for determining the initial parameter values prior to optimisation that this initial estimate is not necessary. Maybe they found it necessary due to the prohibitive amount of time required for a comprehensive grid search. This is where the trick used in the auto-analysis in relax would be very powerful - these initial parameter estimates could come from the analytic MQ CR72 model (the multiple quantum Carver and Richards 1972 model) rather than from a grid search. Maybe the synthetic Korzhnev data would be a better match for your aims (http://dx.doi.org/10.1021/ja049968b), as the dw value measurements/assumptions are not there. If we know that we can optimise both the dwH and dwN (or dwC) correctly, that might be of higher value. The lower temperature assumptions for an initial parameter estimate are not needed. But we could try to perform a direct optimisation of the Sprangers data and see what happens. Regards, Edward P. S. I will be on holidays for 3 weeks starting from this weekend, possibly with no internet access for most of the time, so I will not be able to respond to emails very quickly during that time. On 6 September 2013 17:00, Paul Schanda <[email protected]> wrote: > Hi Edward, > > I got some data from Remco Sprangers, a PNAS paper with Lewis Kay. > The fit seems to take the delta-omega values from direct observation of the > peaks at lower temperature (slow exchange). > > Paul > > > On 06.09.13 16:57, Edward d'Auvergne wrote: >> >> Hi Paul, >> >> For the literature data, which paper did you choose? Are you asking >> Dmitry Korzhnev about his 2004 JACS paper >> (http://dx.doi.org/10.1021/ja049968b)? Having both the base data, >> ideally in peak list form, together with the fitted parameters or >> original parameters for simulated data would be great. I would like >> to create a relax system test which analyses the SQ + MQ CPMG data >> simultaneously, and then I would write the code to make the test pass. >> >> Cheers, >> >> Edward >> >> >> >> On 4 September 2013 20:30, Paul Schanda <[email protected]> wrote: >>> >>> Hi Edward, >>> >>> I managed to find some literature data that are probably suitable. I will >>> get them from the authors in a week or so. >>> So that probably solves point 3). >>> >>> I agree that the infrastructure needs to be revised for MQ data. This >>> might >>> in fact be the opportunity to re-think the structure. One might want to >>> jointly fit all kinds of data, e.g. SQ+MQ, or CPMG+R1rho. You should also >>> think of the possibility of adding in the future EXSY data. In the slow >>> to >>> intermediate exchange, one might have EXSY + CPMG to be fit together. >>> If relax would be designed now in a way that is sufficiently flexible to >>> handle all this, it would be great. >>> >>> We will have a look at the code that Mathilde and Dominique have >>> implemented; this might also take a few days. >>> >>> best - >>> >>> >>> Paul >>> >>> >>> On 04.09.13 16:47, Edward d'Auvergne wrote: >>>> >>>> The following are reference development notes for how to modify relax >>>> to handle multi-quantum (MQ) data. Currently relax supports >>>> single-quantum (SQ) data but not MQ data. There are three aspects to >>>> this problem, detailed below. Note that the changes for part 2) will >>>> likely break the code of the relax_disp branch for a short time. >>>> >>>> >>>> 1) The MQ models. >>>> >>>> Firstly the specific MQ model(s) must be added to relax (following the >>>> steps at >>>> >>>> http://wiki.nmr-relax.com/Tutorial_for_adding_relaxation_dispersion_models_to_relax). >>>> This part is relatively trivial, especially if the code already >>>> exists, a published reference exists, and equations can be copied from >>>> a paper directly into the relax manual. For example the 'MQ CR72' >>>> model supported by GUARDD (http://dx.doi.org/10.1021/ja049968b). >>>> >>>> >>>> 2) Infrastructure changes. >>>> >>>> There is one infrastructure change required to support this. The >>>> reason is because most people with MQ data will want to analyse it >>>> simultaneously with SQ data (well, that is my guess from the >>>> literature data). Support for optimising both SQ and MQ data >>>> simultaneously needs to be added. Currently relax only handles one >>>> type of dispersion data per data pipe. This should not be too hard to >>>> change. Such a change could allow CPMG-type and R1rho-type data to be >>>> analysed simultaneously as well. In each individual data combination >>>> case, a specific combined model such as 'SQ+MQ CR72' would need to be >>>> created. This model would be independent of the 'CR72' and 'MQ CR72' >>>> models, and the user will have to explicitly select it. >>>> >>>> The only change from the perspective of the user would the >>>> relax_disp.exp_type user function. This user function currently sets >>>> the experiment type for all data in the current data pipe. This needs >>>> to be changed to be similar to the relax_disp.cpmg_frq, >>>> relax_disp.relax_time, relax_disp.spin_lock_field, etc user functions, >>>> where the experiment type is associated with a spectrum ID. The GUI >>>> would be changed so that this user function is not presented when >>>> setting up a new analysis, but rather when the peak lists are being >>>> loaded. The GUI element displaying the experiment type would need to >>>> be removed and the data presented in the peak list GUI element >>>> instead. >>>> >>>> The backend changes would be more extensive. The following generator >>>> functions would need to be added to >>>> specific_analyses.relax_disp.disp_data: >>>> >>>> loop_exp() - for new usages. >>>> loop_exp_frq() - to replace calls to loop_frq(). >>>> loop_exp_frq_point() - to replace calls to loop_frq_point(). >>>> loop_exp_frq_point_time() - to replace calls to loop_frq_point_time(). >>>> >>>> The the code in specific_analyses.relax_disp and >>>> target_functions.relax_disp need to be modified around this new >>>> concept. I.e. a new top level looping needs to be inserted. The >>>> looping hierarchy from highest level to lowest would become: >>>> experiment type, magnetic field strength, dispersion point (CPMG >>>> frequency or spin-lock field strength), and finally relaxation time. >>>> The only difference to now is the highest level experiment type >>>> looping. >>>> >>>> The cdp.exp_type data structure will need to be converted into a >>>> dictionary where the keys are the spectrum IDs. >>>> >>>> >>>> 3) Test data. >>>> >>>> More of an issue is data! Test data is essential to be sure that the >>>> code is functional. And in relax it needs to be incorporated into a >>>> system test. The most important point is that the data needs to be >>>> independent of relax - the same code for fitting cannot be used to >>>> generate synthetic data. The more independence between relax and this >>>> test data, the better. >>>> >>>> Literature data for MQ data appears not to be available. It looks >>>> like corresponding authors would need to be emailed for access to the >>>> base data and numerical results. An alternative is provided by the >>>> Matlab software GUARDD. It has a module for simulating dispersion >>>> data >>>> >>>> (http://code.google.com/p/guardd/#Optimization_and_education_with_RD_Simulator). >>>> This could be used to generate the test data to be included with >>>> relax. Some test data is also provided by the program. This could >>>> also be used, but it suffers from the fact that the real parameter >>>> values are not known. But it can be used to directly compare relax >>>> and GUARDD. >>>> >>>> Regards, >>>> >>>> Edawrd >>>> >>> >>> -- >>> 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 >>> [email protected] >>> http://www.ibs.fr/groups/biomolecular-nmr-spectroscopy?lang=en >>> > > > -- > 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 > [email protected] > http://www.ibs.fr/groups/biomolecular-nmr-spectroscopy?lang=en > _______________________________________________ relax (http://www.nmr-relax.com) This is the relax-devel mailing list [email protected] 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
