A novel approach to denoising ion trap tandem mass spectra

1
A novel approach to denoising ion trap tandem
mass spectra

• by Jiarui Ding, Jinhong Shi, Guy Poirier, and
  Fang-Xiang Wu
• University of Saskatchewan, Canada
• Proteome Science 2009
• Presenter Kyowon Joeng

2
Why this paper?

• Related to my work (spectral pre-processing)
• A good summary on features of spectrum
• EASY

3
Outline

• Spectral pre-processing
• What they did
• Result
• Some features of spectrum
• Conclusion/criticism/discussion

4
Spectral pre-processing

• To increase the number of identified peptides
• Spectrum clustering (Frank, J proteome Res 08
  Tabb, Anal Chem 03)
• Precursor charge correction (Klammer, IEEE CSBC
  05 Na, Anal Chem 08)
• Denoising (Zhang, RCM 08)
• Quality assessment (Na, J proteome Res 06 Bern,
  Bioinformatics 04)
• Need to be simple and fast
• Need to be generic otherwise, need to have a
  killer application

5
What they did

• Denoising of spectrum
• signal peaks peaks from y or b ions
• noisy peaks other peaks
• Intensity normalization
• Using interrelation features to assign Score to
  each peak
• New intensity original intensity Score
• Peak selection
• Use morphological reconstruction filter
• Select local maxima peaks

6
Intensity normalization feature selection

• Score of a peak p is decided by 5 interrelation
  features
• F1 of peaks p such that p-p an a.a.
  mass (Good diff fraction)
• F2 of peaks p such that pp precursor
  mass (Complementary peaks)
• F3 of peaks p such that p-p H2O or NH3
  mass (Neutral loss)

7
Intensity normalization feature selection

• F4 of peaks p such that p-p CO or NH
  mass (Neutral loss)
• F5 of peaks p such that p-p isotope
  mass (Isotope)
• F1-F5 are normalized to have zero mean and one
  variance.

8
Intensity normalization scoring

• Score w0w1F1w2F2w3F3w4F4w5F5
• w0 5 Offset for non-negative score
• w1 w2 1 Good diff complementary
• w3 w4 0.2 Neutral losses
• w5 0.5 Isotope
• The weights are decided by referring to Sequest
  scoring function

9
Peak selection

• After intensity normalization, it is likely that
  signal peaks are local maxima.
• To select the local maxima, morphological
  reconstruction filter is adopted

10
Morphological filter

• State of the art filter in image processing
• Everyone used it at least one time not so many
  knows it is the morphological filter.
• Flood Fill color tool morphological filter

11
Morphological filter

• Given marker signal (or curve) and mask signal
• Dilate mask signal repeatedly until contour of
  dilated mask signal fits under marker signal.
• In each dilation, each point of marker signal
  takes the maximum value of its neighborhood.

12
Morphological filter
13
Dataset

• ISB ESI ion trap 37,044 spectra
• TOV LCQ DECA XP ion trap 22,576 spectra
• Database ipi.Human protein database
• Mascot is used to evaluate denoising

14
Mascot parameters
15
Number of identified spectra

• Spectrum is identified if its Mascot ion score is
  larger than the identity threshold (no target
  decoy FDR is derived)

16
Number of identified spectra
17
False positive rate

• A spectrum in ISB dataset is false positive if it
  is identified in ipi.HUMAN database but it is not
  from the known 18 proteins.

18
Intensity normalization vs. peak selection
19
Features of spectrum

• Number of peaks
• Total ion current (total intensity of a spectrum)
• Good-Diff fraction
• Total normalized intensity of peaks with
  associated isotope peaks
• Complements
• Water losses
• Signal to noise ratio

20
Features of spectrum

• The average intensity of the peaks
• Total number of peaks having relative intensities
  greater than x of TIC
• Among them, only features considering m/z
  differences between peaks turned out to be
  significant. (Bern, Bioinformatics 04)

21
Conclusion

• A denoising algorithm that uses features of
  spectrum is introduced.
• It is simple and improves quality of spectrum
• 15-30 more spectra were identified by Mascot
  after denoising

22
Criticism method

• Intensity normalization is too heuristic.
• Among used features, neutral losses are often
  observed in noisy peaks (e.g., precursor peaks).
• Features were manually selected, and no new
  feature was introduced.
• The benefit of morphological filter is not clear.

23
Criticism result

• Standard target-decoy analysis was not shown.
• It is about denoising, but the result of
  denoising is not directly shown.
• Proposed scheme may not suitable for other tools.
• The running time of their algorithm is not shown
  only Mascot search time was shown.

24
Complement peaks associated with their
intensities?

• For Discussion