UC San Diego

Imaging to assess the presence and localization of specific molecules in tissues and cells is central to the study of biological systems. However, most imaging technologies focus on specific molecules of interest. An exciting recent advance is the development of Mass Spectrometry Imaging (MSI), which allows for the generation of topographic 2D maps for various endogenous and some exogenous molecules (e.g., drugs and their metabolites) without prior specification. Advances in MSI have transformative potential, allowing us to answer questions about the localization of proteins, peptides, lipids, metabolites and other molecules. To help MSI realize its potential, we describe several algorithms for the analysis of MSI data from different angles. In a first problem, we start with the premise that we are given a pre -defined region of interest (ROI) based on the morphology of the tissue or organism. We aim to find and identify molecules that are specifically expressed in the ROI. We solve this problem by using a statistics for localization specificity and a novel pipeline for identification. Next, we extend the approach above to segment the MSI dataset into consistent regions of interest, and for each segment, we identify a molecular signature: a collection of peaks that are preferentially expressed in that segment. Our implementation, called AMASS (Algorithm for MSI Analysis by Semi-supervised Segmentation), relies on the discriminating power of a molecular signal instead of its intensity as a key feature, uses an internal consistency measure for validation, and allows significant user interaction and supervision as options. A third problem examines the comparative analysis of many MSI datasets. We describe a new method which, given a set of pertinent query molecules, finds, in each dataset, all molecules that have a similar spatial distribution and clusters the datasets based on the resulting molecular signatures. The approach has the potential to identify unknown relationships between multiple data acquisitions. Finally, we briefly touch on the peptide identification from on- tissue MS/MS data using a spectral library specific to MALDI imaging peptide identification. Our preliminary results highlight the potential of this approach