Copula Graphical Modeling of Proteomic Data in Thyroid Lesions

Thyroid nodules are a common occurrence in the general population, yet preoperative assessment of malignancy often remains inconclusive. This uncertainty can lead to surgical decisions that later may prove unnecessary. Identifying potential molecular biomarkers could refine diagnosis and support more informed care. In this setting, proteomic profiling with Matrix Assisted Laser Desorption Ionization Mass Spectrometry Imaging (MALDI-MSI) has shown promise for characterizing tissue at the molecular level. This thesis presents a comparative statistical study of MALDI-MSI proteomic data from thyroid biopsies classified into five diagnostic categories: Follicular Adenoma, Hürthle Cell Adenoma, Papillary Thyroid Carcinoma (PTC), Follicular Variant of Papillary Thyroid Carcinoma (FVPTC), and Noninvasive Follicular Thyroid Neoplasm with Papillary-like Nuclear Features (NIFTP), a recently defined and diagnostically challenging entity. The analysis compares univariate and multivariate statistical approaches to explore molecular differences between diagnostic groups. Elastic Net regression and sparse Partial Least Squares Discriminant Analysis (sPLS-DA) are employed as supervised methods for feature selection and discrimination among diagnostic groups. In contrast, a Copula Graphical Model for Heterogeneous Data is applied in this proteomic setting to characterize conditional dependencies among molecules and to explore how network structure may inform feature relevance across diagnostic categories. The thesis compares the molecules identified by different methods, examining their recurrence across approaches to highlight patterns that may be of interest for future clinical studies.