WAVELETS AND PROJECTING SPECTRUM BINNING FOR PROTEOMIC DATA PROCESSING - Quantitative Medical Data Analysis Using Mathematical Tools and Statistical Techniques - page 162

The paper is organized as follows: mass spectrometric data preprocess-
ing procedures, GG procedure { Gaussian smoothing and Genetic algorithm
for nal peak binning, WW procedure { Wavelet denoising and Window-
based peak alignment, as well as the WG procedure { Wavelet denoising
and Genetic algorithm for peak binning, will be briey reviewed in the next
section. In section 3, the projecting spectrum binning (PSB) method will
be introduced and a new MS data preprocessing procedures using wavelet
denoising and PSB, named as WPSB procedure is proposed. Comparison
results are shown in Section 4. Final remarks and questions for future study
are discussed in the nal section.

2. Review of MS Data Processing Procedures

2.1. The GG procedure
The GG procedure was based on Gaussian smoothing and Genetic algo-
rithm for nal peak binning, which was adopted by Vanderbilt Medical
Center and can be described as follows.
Step 1, Raw data smoothing and calibration: Use a Gaussian smoothing
function provided by the Data Explorer software (Applied Biosystems, Fos-
ter City, CA). After smoothing, each spectrum was internally calibrated to
minimize the inevitable mass shifts with a single sample as well as between
dierent samples. The spectra were calibrated using four xed m=z values.
Step 2, Baseline correction: A piecewise linear function interpolates the
lowest intensity within a series mass windows. The baseline corrected spec-
trum is generated by subtracting the area under the t.
Step 3, Normalization: Compare each spectrum to one pre-specied ref-
erence spectrum. For each comparison, the common peaks between two
spectra are identied. Calculate the \intensity ratio" for each peak in com-
mon. Fit a linear equation to these ratios. Remove those ratios bigger that
two standard deviations from the t. Then t the remaining intensity ratios
again. Rescale the intensity for each spectrum by multiplying the correspon-
dence value of the tted equation.
Step 4, Peak selection: Based on a pre-specied S=N ratio, decide the
number of peaks for each spectrum 27 .
Step 5, Peak alignment (binning): Genetic Algorithm (GA) is cus-
tomized to search the best bins for combining the peaks represent the same
protein. A search started to nd the optimal window width and location
if it can maximize

P

n 2 , n is the number of peaks within that window,
while the maximum width of a window is constrained by 3 + 0:001Mass