Peak picking in mass spectra

Abstract

High throughput analysis of proteins using mass spectrometry requires an efficient signal preprocessing which reduces the amount of data with minimal loss of information. Therefore the peaks that belong to a peptide have to be detected, and important features like their peak centroid position, the height and the area have to be determined. Here we present a peak picking algorithm which detects peaks in a noisy mass spectrum using its continuous wavelet transform. By adapting the wavelet to the theoretical peak shape, each peaks centroid position is precisely identified; thus, overlapping peaks (e.g. caused by electronspray ionization) are well separated. Representing each peak not only by its centroid position and area but also by the parameters of the best fitting asymmetric lorentzian or hyperbolic secant squared functions, yields valuable information about the original peak shape for further analysis. Given a mass spectrum with 100,000 raw data points representing 200 peaks, our algorithm stored 5 parameters per peak, and was able to reduce the memory requirement down to 1 percent.