New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...
New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...
New Statistical Algorithms for the Analysis of Mass - FU Berlin, FB MI ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
3.3. PREPROCESSING 35<br />
Figure 3.3.7: This shows <strong>the</strong> results <strong>of</strong> <strong>the</strong> TopHat Filter: <strong>the</strong> TopHat opening (see<br />
text) is subtracted from <strong>the</strong> raw signal (shown on <strong>the</strong> left) yielding <strong>the</strong> filtered version<br />
(on <strong>the</strong> right). Note that this method does not generate negative signals.<br />
3.3.6 Normalization<br />
Inter-spectrum normalization is <strong>the</strong> process <strong>of</strong> removing systematic variations<br />
between spectra. Many different techniques exist such as “Inverse Normalization”<br />
(Petricoin et al., 2002) or “Logarithmic Normalization” (Li et al., 2004).<br />
Our implementation follows <strong>the</strong> idea <strong>of</strong> <strong>the</strong> most frequently used method which<br />
is global normalization with respect to <strong>the</strong> average total ion current (TIC 2 ) 3<br />
(Fung and Enderwick, 2002; Baggerly et al., 2003) with an important extension:<br />
from <strong>the</strong> set <strong>of</strong> spectra to be normalized all TIC values are computed,<br />
outliers removed and <strong>the</strong> remaining highest value (instead <strong>of</strong> <strong>the</strong> average) is<br />
used <strong>for</strong> <strong>the</strong> actual computation.<br />
3.3.7 Computational Complexity <strong>Analysis</strong> <strong>of</strong> Preprocessing<br />
The complexity <strong>of</strong> <strong>the</strong> single steps <strong>of</strong> <strong>the</strong> preprocessing<br />
Denoising: O(n log n) (Besbeas et al., 2004)<br />
Baseline reduction: O(n) (Gao et al., 2003)<br />
Normalization: O(n)<br />
(n being <strong>the</strong> number <strong>of</strong> input samples) yields a total complexity <strong>of</strong> O(n log n).<br />
2<br />
The TIC is <strong>the</strong> sum <strong>of</strong> <strong>the</strong> area <strong>of</strong> all peaks in a spectra.<br />
3 “TIC <strong>of</strong> spectrum”<br />
The normalization by <strong>the</strong> ratio R =<br />
is reported to be superior<br />
“Average TIC <strong>of</strong> all spectra”<br />
to o<strong>the</strong>r methods tested (Norris et al., 2005; Sauve and Speed, 2004).