Network models derived from genomic profiles of cancer tumours can provide insight into genetic drivers behind a disease, provide biomarkers for diagnostic prediction and be used to propose new treatment regimes. Graphical lasso estimates are widely used; they can be adapted to model disease subtypes and to incorporate prior information. However, it is a well known that large-scale estimates are highly variable which limits their interpretable value. In addition, the data correlation matrices that the models are derived from can be dominated by structures/processes that aren't very interesting from a disease-predictive standpoint. Recently, researchers have looked into approaches that improve estimation precision for large-scale models (see this article
In this project, you would apply the root-based graphical lasso to cancer data and investigate if this enhances disease-relevant components in the generated networks.
Obs! För GU-studenter räknas projektet som ett projekt i Matematisk Statistik (MSG900/MSG910).
Examinator Maria Roginskaya, Marina Axelson-Fisk
Institution Matematiska vetenskaper