A1076
Title: Uncovering latent molecular patterns in mass spectrometry imaging via spatially-constrained graphical mixture models
Authors: Giulia Capitoli - University of Milano-Bicocca (Italy) [presenting]
Abstract: Mass spectrometry is a class of imaging techniques that measure molecular abundance in tissue samples at cellular resolution, while preserving the spatial structure of the tissue. In particular, mass spectrometry imaging has the capability to differentiate regions that are indistinguishable to pathologists at the microscopic level. A central goal in mass spectrometry data analysis is to identify molecules with similar functions within the analyzed biological system, enabling a better understanding of abnormal molecular mechanisms. The aim is to identify relevant biomolecules associated with cancer cells and the tumor microenvironment, thereby expanding biological knowledge. A Gaussian graphical mixture model is proposed to address unobserved heterogeneity and segment tissue sections into regions based on distinct molecular profiles. Specifically, the aim is to identify groups of molecules with similar activation patterns, investigate their spatial mapping within cancer tissue samples (e.g., renal and/or thyroid neoplasm), and discover clusters of molecules whose activation is linked to specific biological mechanisms. To model this heterogeneity, underlying molecular graphs are reconstructed from the data using sparsity constraints and spatial dependencies between neighboring pixels are incorporated. To account for the spatial nature of the dataset, hidden Markov random fields are utilized, ensuring that the spatial structure is effectively captured.
