Hi Soumya,
I've had a close look at your relaxation data and model-free results
attached to https://gna.org/support/?3127, and have some comments
below:
I've conducted model-free analysis on relax using relaxation data (T1, T2
adn
NOE) collected on 600 and 800 MHz instruments.
Before I get to the other issues, I can see problems with the base
data. How did you calculate the T1, T2, and NOE? And more
importantly how did you calculate the errors? I would recommend you
try to calculate these again using relax, and see if you get the same
results. It is clear that your steady-state NOE errors are too small,
they are not even visible. If you have duplicated or triplicated
spectra, relax can use these to obtain an error estimate.
Or you can perform the normal procedure of measuring the RMSD of the
baseplane noise. The 'rm' command in Sparky is the most powerful for
this. Do not use the estimates that spectral software give for the
entire spectrum, that is not accurate enough for a model-free analysis
(you must avoid peaks and the water signal, and it is unknown how
these blackbox estimates do that, or even if they do that at all).
You need to draw many boxes in empty parts of the spectrum, but close
to your peaks in the centre where it is noisier, and use an average
RMSD value from that. You can then input the RMSD into the NOE
analysis in relax and obtain the real NOE errors.
I used the GUI with the default
settings except for estimated rotational correlation time (for me it is 5
ns).
Note that model-free analysis consists of a complex iterative
protocol. Here is the protocol used in the GUI:
http://www.nmr-relax.com/manual/Model_free_analysis_in_reverse.html
You can read about it in the GUI by clicking on the 'About' button in
a model-free analysis tab. I published this protocol in the paper:
- d'Auvergne, E. J. and Gooley, P. R. (2008). Optimisation of NMR
dynamic models II. A new methodology for the dual optimisation of the
model-free parameters and the Brownian rotational diffusion tensor. J.
Biomol. NMR, 40(2), 121-133.
(http://dx.doi.org/10.1007/s10858-007-9213-3).
The key to this protocol is that the chicken or egg problem - which
comes first, the diffusion tensor or internal motion - is reversed
compared to how protocols were constructed in the past. This protocol
optimises the internal motion first and then the diffusion tensor. At
no point do you specify a diffusion tensor - the protocol will find
the correct one by itself. Correlation time estimates from other
biophysical techniques are always different to that found in the NMR
sample, due to concentration differences and microviscosity, so you
should not use these as starting points (which cannot be done in this
protocol anyway).
My S2 order parameters are the inverse of what I expect. Disordered regions
appear to have values closer to one and regions I know are structured have
values that are closer to zero (see attached). The errors associated with
the
"ordered" NHs are quite large whereas those for dynamic residues are
smaller.
This appears to be the classic problem of inputting the data in the
wrong format - specifically when relaxation times and not relaxation
rates are input into the software. Rates must always be used. The
reason is that all of the fundamental NMR theory going all the way
back to the bible of NMR (Abragam) is based on rates. Spectral
density values add to give rates. Times are not the natural unit for
the physics of the system. If you use relax to recalculate all of the
R1, R2, and NOE data, then this problem should disappear.
Here's what I observed. The calculation ran without any errors. It took
about
three days to run with a single processor. Is that normal or is there
something wrong here?
3 days is normal. It can take anywhere between a few hours on a
multiprocessor machine to 1-2 weeks. It depends on the molecular
system being used, the real diffusion tensor, how complicated the
internal dynamics are, and if the system is not perfectly described by
the classic spherical, spheroidal or ellipsoidal diffusion tensors.
It takes a long time because this is a full iterative protocol being
executed. And relax has far higher accuracy than any of the other
model-free software (see http://dx.doi.org/10.1007/s10858-007-9214-2).
If you have a multi-core or hyperthreaded system, you should try to
run Gary Thompson's multiprocessor for relax. If you have 8 cores,
you can run the relax model-free calculations 7 times faster (one of
the 8 is used for the master processor,
seehttp://www.nmr-relax.com/manual/Introduction_multi_processor.html
and the following sections).
I have conducted consistency tests on these data. I guess I'd like to know
what the threshold might be for something that has a Tc = 5 ns.
The consistency tests are independent of the correlation time. You
must however input relaxation rates and not times. There is a good
description of what problems to look for in the manual
(http://www.nmr-relax.com/manual/Consistency_testing_script_mode_visualisation_data_output.html,
though the PDF is of higher quality). Also have a look at Sebastien
Morin's paper on the subject (he implemented the consistency testing
in relax):
- Morin, S. and Gagn ́e, S. (2009a). Simple tests for the
validation of multiple field spin relaxation data. J. Biomol. NMR, 45,
361–372. (http://dx.doi.org/10.1007/s10858-009-9381-4).
My protein behaves as a single domain to the best of my knowledge. It is
actually a tethered complex between two proteins but they bind each other.
The
dynamic region near the C-terminus is the glycine-serine loop that connects
the two entities. I don't believe it comes apart very often as one of the
proteins precipitates if left on its own.
There appears to be no concentration-dependent dimerisation. The HSQC
spectra
of the protein at 0.04 and 0.6 mM look identical.
From the data, it appears as if everything should be fine, apart from
the NOE errors. However I can see that one spin has an NOE value of >
1. This is not physically possible, so you should probably deselect
this spin in the analysis and look back at the spectra to see why this
is the case.
The relaxation data for the 600 and 800 MHz exp are attached. The S2 has
also
been graphed. Does anyone know why my S2 parameters are the opposite to what
I'm expecting?
As I mentioned above, the S2 values look exactly as I would expect if
times and not rates are feed into the program. I hope this
information helps.
Regards,
Edward
Re: [sr #3127] Opposite S2 values to expected