Hi Edward,
Sorry for the confusion. The local_tm models were generated and handed over
to the next step of the analysis. However, the local_tm.txt was empty because
all local_tm values were estimated higher than global tm. S2 values were
estimated at first in the same size order I also see if I use the HYDRONMR
diffusion model as a starting point (rather close to one except of two
flexible loops). However, during the diffusion model selection s2 values
become smaller and both local and global correlation times bigger. I think
this could be due to over parameterization, or? I already increased the
experimental error by factor 2 but this seemed not to help.
I did not remove the protein tag after purification (TEV site + StrepTag,
about 20 residues). Because it is also not part of the used Xray crystal
structure, it might induce a bias also I expect it to be highly flexible and
move independently from the folded domain.
Regards,
Jonas
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: Donnerstag, 21. Januar 2016 10:57
To: Jonas Hanske
Subject: Re: Confidence in Model-free output
Hi,
This sounds like a very strange bug! Which version of relax are you using?
Looking at the most up to date code:
http://www.nmr-relax.com/api/4.0/auto_analyses.dauvergne_protocol-pysrc.html#dAuvergne_protocol.execute
(line 718)
I can see that the list of diffusion models comes from the list of
subdirectories in the results directory:
# The contents of the results directory.
dir_list = listdir(self.results_dir)
The list of data pipes used in model selection, self.pipes, is created by
checking the names in dir_list. Therefore I can only guess that this is a
file system error (or the directory was accidentally renamed or moved). The
analysis will create the local_tm directory - this is an essential first step
in this protocol - so it simply cannot find that directory. Did you notice
such a problem? Or did you temporarily run out of hard disk space? Do you
have a listing of the results directory and can you see the local_tm/
directory there? This is really, really bizarre!
As for outputting the diffusion tensor information to a text file, you could
maybe run relax with the --tee flag to output all messages to a file, and
then load each results file and run diffusion_tensor.display().
Alternatively you can access the object directly with a relax script, and
then print out the elements:
relax> pipe.create('test', 'mf')
relax> diffusion_tensor.init((1.340e7, 1.516e7, 1.691e7, 0.000e7,
0.000e7, 0.000e7), param_types=3)
relax> cdp.diff_tensor.Da
2630000.0
relax> print("Da: %s" % cdp.diff_tensor.Da)
Da: 2630000.0
relax>
Note that the objects in cdp.diff_tensor are automatically generated on
initialisation and Python will raise an AttributeError if you ask for a
parameter that does not exist for that diffusion tensor type (see the full
parameter list at
http://www.nmr-relax.com/api/4.0/data_store.diff_tensor-pysrc.html#DiffTensorData
).
Regards,
Edward
On 21 January 2016 at 10:00, Jonas Hanske <Jonas.Hanske@xxxxxxxxxxxx> wrote:
Hi Edward,
Thanks for coming back to me. The input data should be fine. Temperature
calibration was performed with MetOH in H2O. I eliminated all overlapping
peaks and fitted the curves with the functions you also provide in relax
and performed consistency testing which worked well with only three
outliers. I eliminated the outliers from the analysis and run your
protocol again. However, I am worried about model selection since the chi2
values are very close and I am not confident if the selection procedure
reflects the true diffusion model. I calculated the diffusion parameters
with HYDRONMR which gave a ellipsoid model which also came out in the last
mf runs. I fed this as starting point manually to the program and will now
perform several manual runs to see if the results are similar to the
automated.
The local_tm files were always empty in my analyses.
Is there a quick way to write the diffusion tensor parameters to a .txt
file? I am new to python...
Regards,
Jonas
-----Original Message-----
From: edward.dauvergne@xxxxxxxxx [mailto:edward.dauvergne@xxxxxxxxx] On
Behalf Of Edward d'Auvergne
Sent: Donnerstag, 21. Januar 2016 09:48
To: Jonas Hanske
Cc: relax-users@xxxxxxx; Christoph Rademacher
Subject: Re: Confidence in Model-free output
Hi Jonas,
Welcome to the relax mailing lists! From the output, I guess you are using
the automated model-free analysis protocol (the one I developed and
published). Or are you running a custom analysis protocol? Such results are
not uncommon and are usually an indication of a bigger problem. Often this
comes back to inadequate temperature calibration and/or control - have a
close look at this text I wrote on the
subject:
http://www.nmr-relax.com/manual/Temperature_control_and_calibration.html
Other times it can be due to a bias in the fitting of relaxation curves.
Which software did you use to calculate the relaxation rates from the base
data? Another reason could be the failure of the diffusion tensor
optimisation due to the specific tensor not being an adequate description of
reality. This happens if you have large scale internal motions, partial
dimerisation (even with 5-10%), long flexible tails, etc. A good test is to
use the local_tm global model, which is strangely absent from the above list
of global models. The local_tm global model will always be too noisy and
give some terrible outlier values, but in general it is a good test for the
diffusion tensor approximation. As there is no global diffusion assumed, the
local tm values can adapt to the local diffusion contribution. The single
value may be a good enough approximate average value for the spread of 3, 5,
10 or more global diffusion coefficients. Therefore if you see the same
trend in S2 values in the local_tm models as you do in the oblate spheroid,
then you know that the trend is real.
Whether that real trend is due to the protein system or due to bad input data
is another question. Hopefully this helps.
Regards,
Edward
On 6 January 2016 at 10:14, Jonas Hanske <Jonas.Hanske@xxxxxxxxxxxx> wrote:
Hi Edward,
I am a last-year grad student working at a Max Planck Institute in Potsdam,
Germany. My thesis is concerned with a Ca2+ binding protein. I collected
full set of backbone relaxation data of the apo and holo form at two fields
(600 and 750) and saw a slight difference in R2 (about 10% decreased in
holo form), some difference in R1 (slightly increased in holo form), and no
change in NOE. However, after I run model free analysis (using the holo
structure for both apo and holo form since no apo structure is available,
all 9 local_tm models, and all diffusion models), I saw a global tm
difference of more than twofold and tremendous changes in s2 values in both
forms with up to 50% decrease for the holo form. Is this a likely outcome?
I do not know if I should trust model selection especially since the oblate
model that was chosen needed about 200 iterations to converge. I used the
Newton minimizing method. Below, I copied the AIC output from the console
and the corresponding global tm values.
I'd be happy for any advice!
Thanks in advance,
Jonas
AIC model selection for apo form
Global model - all diffusion tensor parameters and spin specific model-free
parameters.
# Data pipe Num_params_(k)
Num_data_sets_(n) Chi2 Criterion
sphere - mf (Wed Dec 30 14:23:37 2015) 197 468
526.81736 920.81736
prolate - mf (Wed Dec 30 14:23:37 2015) 185 468
538.26716 908.26716
oblate - mf (Wed Dec 30 14:23:37 2015) 197 468
512.99733 906.99733
ellipsoid - mf (Wed Dec 30 14:23:37 2015) 197 468
518.69107 912.69107
The model from the data pipe 'oblate - mf (Wed Dec 30 14:23:37 2015)' has
been selected.
Global tm = 1.5247e-08 s
AIC model selection for holo form
Global model - all diffusion tensor parameters and spin specific model-free
parameters.
# Data pipe Num_params_(k)
Num_data_sets_(n) Chi2 Criterion
sphere - mf (Tue Jan 5 18:26:27 2016) 206 504
2305.05273 2717.05273
prolate - mf (Tue Jan 5 18:26:27 2016) 209 504
2299.94808 2717.94808
oblate - mf (Tue Jan 5 18:26:27 2016) 209 504
2281.81543 2699.81543
The model from the data pipe 'oblate - mf (Tue Jan 5 18:26:27 2016)' has
been selected.
Global tm = 6.8584e-09 s
_______________________________________________
relax (http://www.nmr-relax.com)
This is the relax-users mailing list
relax-users@xxxxxxx
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: Confidence in Model-free output