B1614
Title: Biologically-informed gene clustering for spatial transcriptomics
Authors: Davide Risso - University of Padua (Italy) [presenting]
Andrea Sottosanti - University of Padova (Italy)
Sara Castiglioni - University of Padova (Italy)
Abstract: Key biological processes depend on the physical proximity of cells and the spatial organization of tissues. In recent years, technological advances have made it possible to quantify the mRNA expression of large numbers of genes while preserving the spatial context of tissues and cells. In many applications, for instance, a pathologist annotation is available in tumour samples. It can be used as external knowledge to identify genes that show interesting spatial variability within each of the annotated tissue areas (e.g., within a tumour or stroma). A statistical model is presented that clusters the spatial expression profiles of the genes according to a partition of the tissue; this partition can either be learned from the data or given by a domain expert annotation. This is accomplished by modelling the spatial dependency of gene expression across the tissue with an isotropic spatial covariance function. Given the high dimensionality of the problem, the approach has a large computational complexity; to speed up computation, a strategy is considered based on nearest neighbour Gaussian process models.
