Biomedical Engineering Reference
In-Depth Information
tra takes two rounds: the rst round only considers the set of peaks with
S=N greater than a pre-specied value m
1
, pool the list of detected peaks
and combine peaks that dier in location by no more than 7 clock ticks or
in relative mass by 0.003, thus the peaks classied within one range cor-
responds to the same protein. The second round is going back to add the
peaks greater than another pre-specied value m
2
(m
2
S=Nm
1
) to
above list only if they fall within the same range limits (7 clock ticks or
0.003 relative mass) of above peaks just identied.
2.3. The WG procedure
The WG procedure is a relatively new method by using wavelets for de-
noising and genetic algorithm (or PSB) for nal peak alignment. The WG
procedure combines dierent portion of WW and GG procedures and the
rationale for proposing it are explained below.
First of all, denoising (smoothing) is an especially important step for de-
tecting true peaks. The GG procedure used Gaussian smoothing, a default
machine software for smoothing and it may have problems, for example,
closely overlapping peaks may not be distinguished
27
. On the other hand,
the WW procedure applied wavelet smoothing, a very exible and powerful
method for the signal process like MS data. Wavelet denoising normally
starts by transforming from the time domain to the wavelet domain and
then estimating the variability of the coecient. Then it sets up a threshold
parameter and applies either soft or hard thresholding, and nally, trans-
forms back to the time domain. More descriptions on advantages of wavelets
methods for MS data processing and medical data analysis can be found in
[5, 8, 14, 15].
For the whole MS data processing procedure, the nal peak alignment
is an inevitable step due to the drift in the locations of spectral peaks from
on experiment to another even though they represent the same biochemical
substance across dierent spectra. Nevertheless, the quality of alignment
method directly aects the nal pn matrix for further statistical anal-
ysis. The alignment (binning) idea of WW method (combining the peaks
that dier by no more than a certain clock tick or a certain relative mass)
sounds reasonable, but in practice, it might be problematic. For example,
consider 20 consecutive peaks found by peak selection. If any two adja-
cent peaks meet the above criteria for combining peaks, we might end up
combining all 20 peaks together. In applications (see section 4), it actually
shows the implementation algorithm by the WW procedure having trou-
Search WWH ::
Custom Search