Hi, You're welcome. As for changing the auto_analyses/dauvergne_protocol.py file, this will have the desired effect. It is quite easy to modify such a protocol, however the validity of any changes would need to be thoroughly tested. In this case, as I said in the last mail, there are fundamental theoretical problems with this change (http://marc.info/?l=relax-users&m=135522152004674). Also if the protocol is changed, it should then be labelled as a new protocol.
Regards, Edward On 11 December 2012 13:05, Romel Bobby <[email protected]> wrote: > Hi Edward, > > Thank you for your quick response and pointing me into the right direction. > Indeed, I'll need to have a closer look at the second of the Lipari and > Szabo paper. > And also, I'll to follow your suggestion and implement synthetic data into > relax to see what I get. > > With respect to the AICc model selection technique, after reading your > response, I'm not sure if I've done the right thing. I believe I used the > AICc method to select the best model-free model at the end of each round for > the individual four diffusion tensors (and the local_tm). Either way, I'll > rerun the model-free analysis without fiddling the code this time. > But to answer your question, in the auto_analyses/dauvergne_protocol.py file > (at line 763), I changed the line from: > > self.interpreter.model_selection(method='AIC', modsel_pipe=modsel_pipe, > bundle=self.pipe_bundle, pipes=self.pipes) > > to: > > self.interpreter.model_selection(method='AICc', modsel_pipe=modsel_pipe, > bundle=self.pipe_bundle, pipes=self.pipes) > > Cheers, > > Romel > > > On 11 December 2012 23:24, Edward d'Auvergne <[email protected]> wrote: >> >> Hi Romel, >> >> The nanosecond motions you are seeing are interesting. Whether or not >> they are real is a valid question. However like most NMR >> spectroscopists, you are completely wrong in your assumption ;) You >> just need to carefully read the second of the Lipari and Szabo papers, >> something many people do not do, wherein they check the validity of >> internal motions slower than the global tumbling of the molecule. >> >> You can also perform your own tests by setting up synthetic models in >> relax - just load a PDB file, set a diffusion tensor to the values you >> like, set the internal model-free parameters, and then back calculate >> the relaxation data. Then perform a simple single model model-free >> fit and see what happens. I.e. what they do in the second Lipari and >> Szabo paper (well, without relax). You will see that with no >> experimental noise, you can easily extract internal motions slower >> than the global correlation time. Think of it as follows: After one >> time period of the global tumbling tau_m, the exponential total >> correlation function has not hit zero! It takes a few periods of >> tau_m before statistical zero is reached. Therefore other exponential >> functions can be present in the auto-correlation function. In the >> case of ellipsoidal diffusion, there are 5 exponential functions mixed >> in together. So why can, for example, the isotropic diffusion with a >> single exponential not be mixed in with another exponential coming >> from the internal motions rather than the global tumbling process? >> >> The answer, which is present in the original Lipari and Szabo papers, >> is that it most definitely can. The problem of parameter extraction >> is experimental, not theoretical! As soon as you add noise to the >> synthetic relaxation data, it becomes harder and harder to extract the >> slow internal motions reliably. Monte Carlo simulations can show this >> too. Again this can be tested by adding white noise to the back >> calculated relaxation data. I highly recommend you perform these >> tests yourself so that you get a good idea of what you can obtain - it >> will help you understand your system better. You can even use your >> protein system for the synthetic data tests. Note that if the two >> correlation times are identical, you will have problems. Just test >> yourself and see what happens! By learning from these tests, you will >> be able to formulate much stronger conclusions from your experimental >> results. >> >> As for your experimental data, another check would be to see if these >> motions are present in the local tm models. I think I discussed this >> in my second 2008 paper >> (http://www.nmr-relax.com/refs.html#dAuvergneGooley08b or >> http://dx.doi.org/10.1007/s10858-007-9213-3). You also have to >> carefully think if a single diffusion tensor is valid for the entirety >> of your system, an assumption which may not hold. >> >> One question I have though is how did you change the model selection >> protocol in the dauvergne_protocol.py auto-analysis? This is >> hardcoded into relax, and the use of both AICc and BIC is invalid for >> the searching of the universal solution (see >> http://thread.gmane.org/gmane.science.nmr.relax.user/1356/focus=1357 >> and my 2007 paper at >> http://www.nmr-relax.com/refs.html#dAuvergneGooley07 or >> http://dx.doi.org/10.1039/b702202f). You can only use model selection >> techniques where the fundamental derivation is based on the >> Kullback-Leibler discrepancy >> (http://en.wikipedia.org/wiki/Kullback%E2%80%93Leibler_divergence) or >> the Fisher information metric. AIC and ICOMP model selection are >> suitable, for example, but AICc and BIC are not as the first is a >> deliberate divergence from the KL discrepancy and the second arises >> from quite different Bayesian assumptions. To use AICc or BIC, the >> universal solution equation of my 2007 paper would need to be >> reformulated to match the very different fundamental concepts behind >> these techniques and the model-free protocol redesigned to fit. >> >> Regards, >> >> Edward >> >> >> P. S. Because of the number of bugfixes since relax 2.0.0, I would >> recommend updating to relax 2.1.2 >> (http://marc.info/?l=relax-announce&m=135070664825024). >> >> On 11 December 2012 10:28, Romel Bobby <[email protected]> wrote: >> > Dear Ed and relax users, >> > >> > I have recently used relax (ver 2.0.0) to obtain the model-free dynamics >> > of >> > a small protein with a molecular mass of 6 kDa. In fact, it's a protein >> > in >> > complex with a small peptide, but only the protein is 15N isotope >> > labelled. >> > R1, R2 experiments were recorded at 600, 800 and 900 MHz, and the >> > steady-state NOE at 800 and 900 MHz. Temperature calibration was >> > performed >> > before running the experiments for all the fields. >> > >> > I used the fully automated analysis (dauvergne_protocol.py) and that >> > went >> > well. At the end of the run, the selected diffusion tensor had the form >> > of >> > an oblate spheroid with a global rotational correlation time of 3.2 ns. >> > >> > Now, I have observed that some residues display slow internal >> > correlation >> > times (ts) that are in the range of 1-4 ns. Considering the tm is only >> > 3.2 >> > ns, this would mean, that motions slower than the overall tumbling were >> > captured. However, my understanding is and please correct me if I'm >> > wrong, >> > that except for Rex contributions, relaxation measurements are >> > insensitive >> > to motions on a time scale equal to or slower than the overall tumbling. >> > I'm a bit puzzled about the validity of these motions, that is to say, >> > are >> > they physically meaningful? >> > Is it possible that some of the motional models were not adequately fit >> > and >> > poorly chosen? Also, I've used the AICc selection method to reduce the >> > probability of overfitting. >> > >> > Cheers, >> > >> > Romel Bobby >> > Biomolecular NMR Research Group >> > School of Chemical Sciences/School of Biological Sciences >> > The University of Auckland >> > +64 (09) 3737599 Ext 83157 >> > >> > _______________________________________________ >> > relax (http://www.nmr-relax.com) >> > >> > This is the relax-users 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-users >> > > > > > > -- > Romel Bobby > Biomolecular NMR Research Group > School of Chemical Sciences/School of Biological Sciences > The University of Auckland > +64 (09) 3737599 Ext 83157 _______________________________________________ relax (http://www.nmr-relax.com) This is the relax-users 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-users
