AUTOMATED PEAK IDENTIFICATION IN A TOF-MS SPECTRUM - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 115

blood serum, it is estimated that there may be up to 10,000 proteins with
concentrations ranging over at least 9 orders of magnitude 9;10 . However,
the dynamic range of MS instruments is only 34 orders of magnitude 11 ,
thus careful biochemical sample preparation is critical. As more eort has
been devoted to improving the performance of MS instruments to provide
more detailed information about the sample, to increase resolving power,
and to lower the detection limit, the resulting mass spectra have inevitably
become more complicated. Very often, as in the biomarker discovery for
disease detection, it is not clear beforehand which peak is important. Thus
as many peaks as possible must be detected and characterized. This is also
true for protein identication. Compounding the problem of dense data
sets, roboticized sample preparation and computerized data collection allow
researchers to generate dozens, or hundreds, of such spectra in a few hours.
The analysis of such large raw data sets produced by survey mass spec-
trometers creates a bottleneck in the research process. To overcome this
bottleneck, the rst step is to simplify a spectrum that contains thousands,
even millions, of data points down to only the essential information about
peaks, i.e., positions and intensities. In this way, a spectrum can be reduced
to only a few hundreds points that represent peak positions, intensities and
uncertainties in the peak positions and intensities.
We should emphasize that not only mass spectrometry faces this peak
detection problem. In fact, peak identication is a quite general problem in
many analytical instruments. A good automated peak detection procedure
should run rapidly, and give repeatable and accurate results. It should nd
all signicant peaks in a spectrum but not report false peaks. For some
biological samples, the concentration is very small and the spectrum has a
low signal-to-noise ratio, hence nding peaks is dicult. Missing peaks in a
spectrum, and reporting false peaks, can both potentially lead to discovery
of false \biomarkers." This could lead to wasted further investment, which
could potentially be costly and time consuming.
A good peak detector should give accurate peak position assignments
and peak intensity estimations. The importance of accuracy in peak posi-
tion is obvious, it has signicant inuence on the database searching results.
Accurate peak intensity estimations are also important when quantitative
analysis is required. For example, when looking for proteins that are associ-
ated with disease, it is very possible that the proteins we are looking for are
common in both healthy and sick people but are overexpressed or under-
expressed. In this case, it is not a \yes or no" problem, but rather a \more
or less" problem, and the correct peak intensity estimation is essential.