Mathematics in Independent Component Analysis
Mathematics in Independent Component Analysis
Mathematics in Independent Component Analysis
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
218 Chapter 15. Neurocomput<strong>in</strong>g, 69:1485-1501, 2006<br />
<strong>in</strong>troduced by the statistical analysis procedure, hence denois<strong>in</strong>g deemed necessary.<br />
The algorithms discussed above have been applied to an experimental 2D Nuclear<br />
Overhauser Effect Spectroscopy (NOESY) proton NMR spectrum of the<br />
polypeptide P11 dissolved <strong>in</strong> water. The synthetic peptide P11 consists of 24<br />
am<strong>in</strong>o acids only and represents the helix H11 of the human Glutathion reductase<br />
[21]. A simple pre-saturation of the water resonance was applied to prevent<br />
saturation of the dynamic range of the Analog Digital Converter (ADC).<br />
Every data set comprises 512 Free Induction Decays (FIDs) S(t1, t2) ≡ xn[l]<br />
or their correspond<strong>in</strong>g spectra ˆ S(11, ω2) ≡ ˆxn[l], with L = 2048 samples each,<br />
which correspond to N = 128 evolution periods t1 ≡ [n]. To each evolution<br />
period belong four FIDs with different phase modulations, hence only FIDs<br />
with equal phase modulations have been considered for analysis. A BSS analysis,<br />
us<strong>in</strong>g both the algorithm GEVD us<strong>in</strong>g Matrix Pencil (GEVD-MP) [28]<br />
and the algorithm dAMUSE [33], was applied to all data sets. Note that the<br />
matrix pencil with<strong>in</strong> GEVD-MP was conveniently computed <strong>in</strong> the frequency<br />
doma<strong>in</strong>, while <strong>in</strong> the algorithm dAMUSE <strong>in</strong> spite of the filter<strong>in</strong>g operation<br />
be<strong>in</strong>g performed <strong>in</strong> the frequency doma<strong>in</strong>, the matrix pencil was computed <strong>in</strong><br />
the time doma<strong>in</strong>. The GEVD is performed <strong>in</strong> dAMUSE as described above to<br />
achieve a dimension reduction and concomitant denois<strong>in</strong>g.<br />
4.3.1 Local ICA denois<strong>in</strong>g<br />
For denois<strong>in</strong>g we first used the LICA denois<strong>in</strong>g algorithm proposed above<br />
to enhance the reconstructed prote<strong>in</strong> signal without the water artifact. We<br />
applied the denois<strong>in</strong>g only to those components which were identified as water<br />
components. Then we removed the denoised versions of these water artifact<br />
components from the total spectrum. As a result, the additional noise is at<br />
least halved as can also be seen from figure 7. On the part of the spectrum<br />
away from the center, i.e. not conta<strong>in</strong><strong>in</strong>g any water artifacts, we could estimate<br />
the <strong>in</strong>crease of the SNR with the orig<strong>in</strong>al spectrum as reference. We calculated<br />
a SNR of 17.3 dB of the noisy spectrum and a SNR of 21.6 dB with apply<strong>in</strong>g<br />
the denois<strong>in</strong>g algorithm.<br />
We compare the result, i.e. the reconstructed artifact-free prote<strong>in</strong> spectrum of<br />
our denois<strong>in</strong>g algorithm to the result of a KPCA based denois<strong>in</strong>g algorithm<br />
us<strong>in</strong>g a gaussian kernel <strong>in</strong> figure 8. The figure depicts the differences between<br />
the denoised spectra and the orig<strong>in</strong>al spectrum <strong>in</strong> the regions where the water<br />
signal is not very dom<strong>in</strong>at<strong>in</strong>g. As can be seen, the LICA denois<strong>in</strong>g algorithm<br />
reduces the noise but does not change the content of the signal, whereas the<br />
KPCA algorithm seems to <strong>in</strong>fluence the peak amplitudes of the prote<strong>in</strong> resonances<br />
as well. Further experiments are under way <strong>in</strong> our laboratory to <strong>in</strong>vestigate<br />
these differences <strong>in</strong> more detail and to establish an automatic artifact<br />
21