dwc@aipna.ed.ac.uk (Dave W. Corne) (04/04/91)
Hello everyone,
My brother wants to write a short paper suggesting how his
research group can use AI methods to find peaks in NMR data,
and has asked me in particular about neural net methods --
He's thinking of using NNs to learn the differences between real
peaks and noise; the problem is this:
"Perhaps the best way in which I can describe the NMR data
is to draw the analogy with Ordnance Survey maps. A 2D NMR
spectrum is actually a "relief map" with (x,y) points which
have an intensity (height): this is represented graphically as
either a contour map (as per Ordnance Survey) or as a 3D
plot (relief map). The raw data are usually in the form
of a 2D matrix (typically, 512 x 512 or 1024 x 512) of
intensities - given the number of data points in the x- and
y- directions and the total width in the x- and y-directions the
(x,y) coordinates of each of the intensities in the matrix may be
calculated:
e.g. a small peak in a much larger matrix may be
1 0 1 0 9 0 0 0 0
0 1 9 0 1 9 0 0 1
0 0 0 9 0 9 1 1 0
0 1 1 9 9 0 1 0 1
which represents a shape like :
~
~ ~
~ ~
~ ~
and the 0's and 1's are "noise".
Hopefully, noise peaks will have, in general, lower intensities,
and real peaks will have particular shape characteristics.
Does distinguishing between the peaks and the noise lend itself to
NN methods? "
If anyone out there has any thoughts, comments, ideas and/or
references that would help my brother out, we would be very pleased
to receive them. I`m sure Simon will be happy to discuss his research
in more detail if anyone is interested.
Could you please send comments to simon@uk.ac.leeds.bio.vax, or me
at dwc@uk.ac.ed.aifh, or both, rather than (or in addition to )
posting them here, as Simon doesn't have news access.
Thank you very much for reading this far, and we look forward
to hearing some suggestions on this matter.
Dave: dwc@uk.ac.ed.aifh
on behalf of
Simon: simon@uk.ac.leeds.bio.vaxtorda@igc.ethz.ch (Andrew Torda ) (04/06/91)
In article <4253@aipna.ed.ac.uk>, dwc@aipna.ed.ac.uk (Dave W. Corne) writes: [...] > research group can use AI methods to find peaks in NMR data, > and has asked me in particular about neural net methods -- > > He's thinking of using NNs to learn the differences between real > peaks and noise; the problem is this: [... nmr description here ...] > If anyone out there has any thoughts, comments, ideas and/or > references that would help my brother out, we would be very pleased 1. A reference to this being done rather successfully "Identification of 2D 1H NMR peaks using neural networks" Kjaer, M. and Poulsen, F.M. Carlsberg Lab, Dept of Chemistry, Gamle Carlsberg Vej 10, DK-2500 Valby, Copenhagen, Denmark, fax +45 31211366 Proceedings of Computational aspects of the study of biological macromolecuoles by NMR, A NATO Advanced Research Workshop, Il Ciocco, Italy, June 3-8, 1990. Last I heard, the proceedings were not finished, but they should be soon. 2. Having been at this conference, I thought these guys had done such a neat job, I went and wrote a simple n. net. for 1D spectra as a demo program for first year students. The method seems to work well. The code is not documented well, and is in C++ using the InterViews library for a graphical front end. -- Andrew Torda, ETH, Zurich