Hi Martin,
From my experience, the best way to handle this is as follows:
- For the first spectrum in the time series, shift the peak list to the tops of the peaks (i.e. 'pc' in Sparky). - Copy this 1st spectrum list onto all spectra, shifting the peaks to the top/centre. - When the peak disappears into the noise, leave it at its current position and do not type 'pc' or equivalent. - Once all spectra are shifted, calculate an average peak list. - Copy this average peak list onto fresh copies of all spectra. - Measure peak heights using this averaged peak list. This methodology will be discussed in a paper which is in preparation. This is a special technique that I personally came up with (big citation hint ;) which is designed to minimise the white-noise bias talked about in the Viles et al., 2001 paper, though not solved there. As the noise often decreases with the decrease in total spectral power, using the tops of the peaks means that you are actually measuring the real peak height plus noise in all cases. As this additional noise contribution may not be constant across spectra, you often end up with a double exponential in the measured data. The technique above eliminates this as you then measure the peak height +/- noise, rather than just + noise. It also nicely solves your problem. Where the peaks disappear, you then are measuring the pure baseplane noise. This is ok, as these white-noise data points centred at zero will help with your subsequent exponential fit in relax. Regards, Edward On 23 May 2012 14:46, Martin Ballaschk <ballaschk@xxxxxxxxxxxxx> wrote:
Hi Edward, I have a general question regarding how to pick signals prior to relaxation rate fitting. Following the rationale of [1] fitting the data with a two-parameter exponential decay is preferable. With relax, I can choose to do so. The question remains if "picking noise" – i.e. picking signal intensities at the position of the reference peaks, although the signal has decayed already – is a good idea. To me it seems only logical that "picking noise" should not be done. The signal decays to zero whatsoever (given the pulse sequence is properly set up) and by picking signals where no signal should be I may pick artifacts that are not distributed evenly over the base plane of the spectrum. Which means, I may be introduce artificial offsets. However relax seems not be happy when I try to use peak lists for T1/T2 relaxation fitting if the lists have different lengths. I always have to do "noise picking", otherwise already the grid_search fails when encountering a incomplete time series. Is the behaviour intended, do I need a complete time series of each amino acid? Or am I missing something? Or am I completely misguided with my not-picking-noise approach? Cheers Martin [1] Viles, J. H., Duggan, B. M., Zaborowski, E., Schwarzinger, S., Huntley, J. J., Kroon, G. J., Dyson, H. J., et al. (2001). Potential bias in NMR relaxation data introduced by peak intensity analysis and curve fitting methods. Journal of biomolecular NMR, 21(1), 1–9. -- Martin Ballaschk AG Schmieder Leibniz-Institut für Molekulare Pharmakologie Robert-Rössle-Str. 10 13125 Berlin ballaschk@xxxxxxxxxxxxx Tel.: +49-30-94793-234/315 Büro: A 1.26 Labor: C 1.10 _______________________________________________ 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