Blanche Francheterre
Interpretable Methods in Unsupervised and Supervised Learning for Multisource analysis of Exposome Data and Prediction of cancer outcomes (2024-xx, 50%)
Blanche’s thesis deals with variable selection and dimension reduction methods for pan cancer data with multiple outcomes and observed through multiple omics, including exposome data. It is supported by the H2020 project DISCERN and co-supervised with Marc Chadeau-Hyam at Imperial College.
Summary
Blanche’s PhD develops high-dimensional variable selection and network inference methods for proteomic cancer data, as part of Work Package 5 of the European DISCERN project. Her first paper, from her Master’s research, applied high-dimensional variable selection to plasma proteomic data to identify biomarkers distinguishing stages of human papillomavirus (HPV)-related cervical carcinoma. Building on this applied work, her thesis now focuses on comparing molecular networks across conditions rather than single biomarkers: she introduced Data Shared Neighborhood Selection (DSNS), a joint graphical model that decomposes each condition’s network into a shared component and condition-specific perturbations, extending the Data Shared Lasso to neighborhood selection. She is currently pursuing the development of additional models that attempt to disentangle adverse effects, in the form of latent factors, from variables directly associated with a specific cancer subtype.