B1653
Title: Latent space models for the hierarchical clustering of complex data in precision medicine
Authors: Giulia Capitoli - University of Milano-Bicocca (Italy) [presenting]
Maria Francesca Marino - University of Florence (Italy)
Stefania Galimberti - University of Milano-Bicocca (Italy)
Monia Lupparelli - University of Florence (Italy)
Abstract: Atypical pathologies, interobserver variability in the evaluation of tumour subtypes, and the absence of clear evidence of malignancy at a morphological level characterize the difficulty of the pathological diagnosis of thyroid cancer. Mass spectrometry imaging is an emerging technology that maps various biomolecules within their native spatial context, revealing hidden information beyond morphological analysis. The aim is to develop latent variable models for hierarchical networks able to capture unobserved heterogeneity and allow for the identification of patients' sub-groups sharing similar molecular profiles and, therefore, a similar risk of developing the disease. Substantially, a model for a multi-layer network is proposed where nodes of the network represent proteins and layers identify different patients. Observed relationships between nodes detect similarities in terms of protein expressions. The intent is to exploit the observed network structures across layers that, in principle, may be different to provide a joint classification of patients and their biomolecules. This methodology is expected to support the clinician in diagnosing thyroid cancer and measure the uncertainty level in the clinical evaluation.
