Analysis of mass spectrometric data - peak picking and map alignment

Abstract

We study two fundamental processing steps in mass spectrometric data analysis from a theoretical and practical point of view. For the detection and extraction of mass spectral peaks we developed an efficient peak picking algorithm that is independent of the underlying machine or ionization method, and is able to resolve highly convoluted and asymmetric signals. The method uses the multiscale nature of spectrometric data by first detecting the mass peaks in the wavelet-transformed signal before a given asymmetric peak function is fitted to the raw data. In two optional stages, highly overlapping peaks can be separated or all peak parameters can be further improved using techniques from nonlinear optimization. In contrast to currently established techniques, our algorithm is able to separate overlapping peaks of multiply charged peptides in LC-ESI-MS data of low resolution. Furthermore, applied to high-quality MALDI-TOF spectra it yields a high degree of accuracy and precision and compares very favorably with the algorithms supplied by the vendor of the mass spectrometers. On the high-resolution MALDI spectra as well as on the low-resolution LC-MS data set, our algorithm achieves a fast runtime of only a few seconds. Another important processing step that can be found in every typical protocol for labelfree quantification is the combination of results from multiple LC-MS experiments to improve confidence in the obtained measurements or to compare results from different samples. To do so, a multiple alignment of the LC-MS maps needs to be estimated. The alignment has to correct for variations in mass and elution time which are present in all mass spectrometry experiments. For the first time we formally define the multiple LC-MS raw and feature map alignment problem using our own distance function for LC-MS maps. Furthermore, we present a solution to this problem. Our novel algorithm aligns LC-MS samples and matches corresponding ion species across samples. In a first step, it uses an adapted pose clustering approach to efficiently superimpose raw maps as well as feature maps. This is done in a star-wise manner, where the elements of all maps are transformed onto the coordinate system of a reference map. To detect and combine corresponding features in multiple feature maps into a so-called consensus map, we developed an additional step based on techniques from computational geometry. We show that our alignment approach is fast and reliable as compared to five other alignment approaches. Furthermore, we prove its robustness in the presence of noise and its ability to accurately align samples with only few common ion species.