Project 1: The idea of this project is to develop a statistical methodology to exploit the information about the molecular structure of the data. In particular, the novel methodology aims at accounting for groups of biomarkers in high-dimensional survival models, namely when biological knowledge suggests that some of the candidate biomarkers belong to molecular pathways. Among other possible approaches, penalized regression methods based on adaptive lasso and sparse-group lasso will be explored.
Project 2: An extremely important tool in biomedical applications is the so-called meta-analysis, in which information from different studies is combined together to provide more evidence for or against a particular result. Unfortunately, not all scientific studies are published, most commonly because researchers and editors are reluctant to publish null results and results with the wrong sign. This phenomenon is captured by the term publication bias. For the meta-analysts, publication bias is a serious problem, as it causes systematically biased effect size estimates. An approach that takes into account the effect of the publication bias in a meta-analysis study has been theoretically developed by Jonas Moss. The aim of this project is to further develop the approach and derive a model which can be effectively used in biomedical applications.