|
From: Edward d'A. <ed...@do...> - 2014-12-12 09:25:23
|
Hi Sean, Welcome to the relax mailing lists! I'll start by explaining how relax can handle your single field strength data. Firstly, I need to point out that relax can do everything Modelfree4 does and it can replicate the results of Modelfree4 exactly (well, almost, as real constraint algorithms are used in relax and that used by Modelfree4 will never be added to relax). I used this replication of Modelfree4 and Dasha behaviour as the basis for my 2008 papers (http://dx.doi.org/10.1007/s10858-007-9214-2, http://dx.doi.org/10.1007/s10858-007-9213-3), for understanding why these two softwares produce sometimes quite different results. I don't think this was mentioned in the papers or supplement as is it so basic in relax to do - you simply pick the same optimisation algorithm as the other software and decrease the optimisation precision and maximum number of optimisation iterations to match (see table 1 of the first paper for these). So, to perform your analysis, you need to note the difference between the theory and optimising the model-free models verses a full analysis protocol. These are very different and it confuses a lot of people. The former is fast and only takes minutes to an hour. The later is super-iterative and much slower, taking a day to 1-2 weeks. There is a lot of software out there that implements the former (Modelfree4, Dasha, Tensor2, DYNAMICS), and then you are expected to manually take the steps required for fulfilling the full protocol yourself (some people don't know this and end up publishing very questionable dynamics). So what are the protocols? The first full protocol was published in the Mandel et al, 1995 paper: http://www.nmr-relax.com/manual/The_methodology_of_Mandel_et_al_1995.html When implemented correctly, this takes an incredible amount of time as expensive Monte Carlo simulations are performed at each step and each model to obtain the chi2 and F-test statistics. As I published in my 2003 model selection paper (http://dx.doi.org/10.1023/A:1021902006114), this should not be used. This introduces, deliberately I should note, or the original paper notes, bias towards the simpler models and hence less motion. Another major problem is the F-test. The F-test is useful for comparing models, but there is fine print the most people don't know about. The F-test is only statistically valid when the two models compared are nested! This is mentioned in the Numerical Recipes book series, for example. As The next protocol I designed based on this one and it handles single field strength data: http://www.nmr-relax.com/manual/The_diffusion_seeded_paradigm.html It is similar in that you need and initial diffusion tensor estimate. However it replaces the chi-squared, F-tests, and chi-squared value empirical cut-offs with statistical AIC model selection, introduces the concept of model elimination (http://dx.doi.org/10.1007/s10858-006-9007-z), removes the need for intermediate Monte Carlo simulations, and simplifies the flow of operation. The last protocol I'll mention here is the one I devised to eliminate the problems associated with the initial diffusion tensor estimate: http://www.nmr-relax.com/manual/The_new_model_free_optimisation_protocol.html http://www.nmr-relax.com/manual/Model_free_analysis_in_reverse.html This is the protocol from my second 2008 paper (http://dx.doi.org/10.1007/s10858-007-9213-3). If you are using the relax model-free auto-analysis (called the dauvergne_protocol, see http://www.nmr-relax.com/api/3.3/auto_analyses.dauvergne_protocol-module.html) or the graphical user interface (GUI), you will be using this protocol. I have implemented it as a fully automated auto-analysis which is essentially just a complicated relax script. Although this protocol solves all of the earlier problems with model-free analyses (protocols), it absolutely requires data at two fields. So, how do you implement the protocol of http://www.nmr-relax.com/manual/The_diffusion_seeded_paradigm.html for single field strength data? I have answered this many times on the relax mailing lists, and it has been summarised in the following relax wiki article: http://wiki.nmr-relax.com/Model-free_analysis_single_field You will find the links to the discussions there highly informative! For the protocol described therein, most steps are present as individual scripts in the sample_scripts/model_free/ directory. I have not merged these together into an automated analysis yet, mainly due to the horrendous artefacts of using single field strength data that are often interpreted as real dynamics. The first step is to obtain the initial diffusion tensor estimate. There is no script for this as most people use Tensor2 for this. Neither Modelfree4 nor Dasha implement this. However this is not ideal as there is no clear mapping of the spherical angles and Euler angles between Tensor2, Modelfree4, Dasha, DYNAMICS, and relax (due to lack of documentation). For the rest, I'll let you read http://wiki.nmr-relax.com/Model-free_analysis_single_field#Protocol. For the problems associated with single field strength data, see http://wiki.nmr-relax.com/Model-free_analysis_single_field#Warning. Don't hesitate to ask if you have any questions. Regards, Edward On 11 December 2014 at 03:26, Sean Leo Moro <sm...@do...> wrote: > Hello Users, > > I am new to the world of model-free analysis, I have my first data set of > R1, R2, and NOE at 800mHz, at 35C and 50C. I would like to use relax as the > program for analysis, as it seems both robust and user-friendly. > Unfortunately, it would be quite difficult for me to acquire data at a > lower field since my protein is ~40kDa and the fact that the 600mHZ > spectrometers at my university can only decouple, but not detect, 15N. > > I am curious, what catastrophic event would occur if I were to try to use > relax with only data sets from 800mHz? I've read through the references, > but could someone make it abundantly clear to me why this is a horrible > idea? Otherwise, I will have to use modelfree, and after reading through > the d'Auvergne/Gooley 2008 J Biomol NMR papers, I feel like that won't be > so great either. > > Thank you, > Sean Moro > PhD candidate > University of California, Irvine. > _______________________________________________ > relax (http://www.nmr-relax.com) > > This is the relax-users mailing list > rel...@do... > > 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 |
