Hi,
Sorry about the late response, I've been incredibly busy lately.
Please see below for some comments:
Rex of around 10^-18 (=nearly zero)
sigma_ex = Rex / omega**2
I extract Rex with the following command, I guess this is then the
already field-corrected value?
value.write( param = 'rex', file = 'example/rex.txt')
This is clearly a bug! For example on an 800, you should multiply
1e-18 with the value of ~2.6e17. Could you file a bug report for
this? A value of 1e-18 should give a significant, yet low, Rex value
of 0.15-0.3 rad.s^-1.
I'm not sure what kind of data / dump would be helpful as an attachment.
I'm also not sure if this is really a bug?
I'll have to think about this one. In the documentation for the
value.set user function
(http://www.nmr-relax.com/manual/value_set.html) it says that you need
to input the Rex value as the field strength independent value, so you
need to scale with:
value = rex / (2.0 * pi * frequency) ** 2
The value.read user function also says this
(http://www.nmr-relax.com/manual/value_read.html). So maybe it makes
sense that if the field strength independent value is input, then this
value should also be output. But the grace.write user function scales
to the first spectrometer frequency. This is confusing and I don't
know a solution yet. Maybe I just need to add a description of the
field strength independent Rex value to the value.write user function
documentation (and value.display). Do you have any ideas?
chi^2 and AIC values do not converge but differ by only a factor of
10^10 [in fact, the fluctuations are small, and in the range of approx
1e-10] from each other in the last ~20 rounds.
How many rounds is it up to? If it runs infinitely, then maybe you
have run into a chaotic system. Now that would be fascinating!
Theoretically anyway, biologically it would be irrelevant.
Hooray chaos! I finally killed it after 190 rounds. That poor workstation
would have worked forever I guess.
I'm guessing you mean 1e-10. Can you see which models are changing?
Models are not changing at all.
Link to a plot of assigned models over the iterations (every plot
corresponds to a single residue, y axis corresponds to models 0-9):
https://gna.org/support/download.php?file_id=17287
Can you find any chi2 or AIC values which match between the rounds?
If you make a table of total parameter number, chi2, and AIC, can you
see any patterns?
Have a look for yourself: I can't see any patterns. Looks like random to me.
Link to tab-seperated table of parameters:
https://gna.org/support/download.php?file_id=17284
Link to plot of parameters over iterations:
https://gna.org/support/download.php?file_id=17286
Link to plot of parameters over iterations, zoomed in:
https://gna.org/support/download.php?file_id=17287
There are some clear patterns in the chi-squared value. However this
is a bizarre problem! It looks like it's jumping around chaotically
with the chi2 value and Rex value varying significantly above machine
precision, but not significantly from a dynamics perspective. I don't
know if this is caused by strangeness in the system you are studying
or strangeness due to the CPU architecture + C library version +
Python version + numpy version. You could look at these plots and
judge where you think the changes are no longer significant, and then
use the maximum iteration argument in the dauvergne_protocol
auto-analysis to terminate early.
However I don't
think I've seen a problem which runs forever - that would just be
theoretically weird.
*sigh*
You can say that again!
Those are quite interesting plots. Though I'm not sure why m0 is
selected so often. I've never seen such a phenomenon.
Maybe you should come around for a visit, talk to Peter Schmieder, Hartmut
Oschkinat and Phil Selenko and then you can crush my dreams of doing
anything useful with our system.
That would be fun.
Lol, that might be an option. I was wondering if you had the latest
Bruker pulse sequences for R1 and R2 from Wolfgang Bermel? They come
with the latest topspin version, or you can just ask him about them.
My old boss, Paul Gooley and I talked to him and got him to create
sequences with single scan interleaving and temperature compensation
blocks - both in the same experiment. I'm guessing that your pulse
sequences are at this level anyway.
I can only recommend switching to Sparky for this type of analysis.
You can use Topspin to split up the file and create a set of 2D fids.
These can then be used for processing in nmrPipe, if you like, and
converted to Sparky format.
I process with Topspin, (zero-fill for 8k, baseline correction, forward
prediction, set the right nc_proc, etc etc), convert to ucsf with
bruk2ucsf, corrected for the sfo1 with ucsfdata and imported the spectra
one-by-one into Sparky. Then copied my reference peak list onto all single
spectra and saved the resulting peak heights.
https://gna.org/support/download.php?file_id=17290
https://gna.org/support/download.php?file_id=17291
https://gna.org/support/download.php?file_id=17292
https://gna.org/support/download.php?file_id=17293
(some of the plots are truncated for outliers)
I don't see an indication that there are significant differences between
picking with sparky or picking with ccpn ...
If you calculate the correlation coefficient for the plot of
https://gna.org/support/download.php?file_id=17291, I think you'll
notice some bias there. And some of the outliers appear to have value
differences up to 10 rad.s^-1. These differences are not
insignificant and could influence the model-free results.
For Sparky (and Topspin for that matter), when extracting such precise
peak height data data I would recommend zero filling to 16k*16k. You
may need to turn off polynomial baseline corrections when zero-filling
so much. And why do you use forward prediction? Linear prediction
should be avoided when you want accurate peak heights.
Regards,
Edward
Re: m0 models