Computational methods for the analysis of mass spectrometry imaging data
A powerful enhancement to MS-based detection is the addition of spatial information to the chemical data; an approach called mass spectrometry imaging (MSI). MSI enables two- and three-dimensional overviews of hundreds of molecular species over a wide mass range in complex biological samples. In this work, we present two computational methods and a workflow that address three different aspects of MSI data analysis: correction of mass shifts, unsupervised exploration of the data and importance of preprocessing and chemometrics to extract meaningful information from the data. We introduce a new lock mass-free recalibration procedure that enables to significantly reduce these mass shift effects in MSI data. Our method exploits similarities amongst peaklist pairs and takes advantage of the spatial context in three different ways, to perform mass correction in an iterative manner. As an extension of this work, we also present a Java-based tool, MSICorrect, that implements our recalibration approach and also allows data visualization. In the next part, an unsupervised approach to rank ion intensity maps based on the abundance of their spatial pattern is presented. Our method provides a score to every ion intensity map based on the abundance of spatial pattern present in it and then ranks all the maps using it. To know which masses exhibit similar spatial distribution, our method uses spatial-similarity based grouping to provide lists of masses that exhibit similar distribution patterns. In the last part, we demonstrate the application of a data preprocessing and multivariate analysis pipeline to a real-world biological dataset. We demonstrate this by applying the full pipeline to a high-resolution MSI dataset acquired from the leaf surface of Black cottonwood (Populus trichocarpa). Application of the pipeline helped in highlighting and visualizing the chemical specificity on the leaf surface.