Multiscale Processing of Mass Spectrometry Data

Timothy W. Randolph, University of Washington
Yutaka Yasui, Fred Hutchinson Cancer Research Center

Abstract

This work addresses the problem of extracting signal content from protein mass spectrometry data. A multiscale decomposition of these spectra is used to focus on local scale-based structure and provide an unambiguous definition of scale-specific features. An objective quantification of features/peaks is accompanied by an efficient method for calculating the location of features that avoids ad hoc decisions regarding signal-to-noise ratios or bandwidths. Scale-based histograms serve as spectral-density-like functions indicating the regions of high density of features in the data. These regions provide bins within which features can be quantified and compared across samples. As a preliminary step, the locations of dominant features within coarse-scale bins are used for registration of spectra. The multiscale decomposition, the scale-based feature definition, the calculation of feature locations and subsequent quantification of features is carried out by way of a translation-invariant wavelet analysis.