Dear Wayne
This is, however a serious issue !
The random sampling procedure will in a way result in random noise.
Usually, noise should be independent of a certain number of parameters, but the way of doing it in ccpnmr makes it dependent, as mentioned before, on peak intensity, spectral width (ratio between "populated" and "unpopulated" spectral regions), number of residues of your protein (translated into peak number) etc. This is not meaningful.
In addition, this "random" noise is then translated by the non-experienced user in "absolute" noise estimation of such critical values as Heteronuclear NOE which may then be entered with the "randomly" estimated noise in quite quantitative analyses such as Tensor that will do MonteCarlo simulations taking into account the STD of the data (that comes from the randomly determiend ccpnmr noise). Therefore I am afraid that, starting from two values that vary by a factor of 2 as found in the original 6 numbers I cited and which, when compared to the noise estimation of other NMR Programs such as NMRView or NmrPipe, are much too elevated, the final result will be quite approximative. I did not go through the whole procedure as my current project does not allow a full Tensor analysis (I do not have a structure) but if somebody is interested to do this, it may be quite interesting.
However, as I understood that you are looking for a better way to do the noise estimation, I will just repeat what I heard in a group seminar this morning (presented by Adrien Favier).
an iterative thresholding algorithm
is applied by first defining an initial threshold as the mean
plus three times the standard deviation using all the points
in the spectrum. Next, a new threshold is calculated in the
same manner but this time using only the spectral data
points below the first threshold. This iterative process is
repeated until no new points exceed the final threshold during
an iterative step.
This procedure is based on an article published by:
Cobas et al., Journal of Magnetic Resonance 183 (2006) 145–151.
The interest compared to the method currently used by ccpnmr is that peaks (=regions with no noise) are discarded in an iterative procedure. This approach gives results comparable to noise calculation within NMRPipe.
Adrien implemented it in a python script, seems to be easy to do (according to him).
Hope this helps
Beate
|