Due to copyright issues, an electronic copy of the thesis must be ordered from the faculty. For the faculty to have time to process the order, the order must be received by the faculty at the latest 2 days before the public defence. Orders received later than 2 days before the defence will not be processed. After the public defence, please address any inquiries regarding the thesis to the candidate.
Trial Lecture – time and place
See Trial Lecture.
Adjudication committee
- First opponent: Associate Professor Geoff Nicholls, University of Oxford, United Kingdom
- Second opponent: Professor Mark van de Wiel, Amsterdam University Medical Center, the Netherlands
- Third member and chair of the evaluation committee: Associate professor Ragnhild Eskeland, University of Oslo
Chair of the Defence
Professor Michael Rory Daws, University of Oslo
Principal Supervisor
Associate Professor Valeria Vitelli, University of Oslo
Summary
Statistical methods for analyzing omics data and other high-dimensional molecular datasets are essential to increase our understanding of complex biological processes. However, analyzing such data typically poses challenges related to its high-dimensionality, heterogeneity, noise and other complexities. Converting the continuous measurements to rankings and instead modeling the ranks can mitigate some of these challenges, as ranks are less sensitive to noise, outliers, and heterogeneity, as well as allowing for easier comparison between different data structures.
Bayesian methods can be particularly well suited for biological contexts, as such methods allow for quantifying uncertainty and incorporating preexisting biological information. No other unsupervised probabilistic method exists for analyzing rank data suitable for an omics context, as rank-based clustering approaches often do not scale to ultra-high dimensions, and one is typically interested in adding to the method specific statistical tasks, such as variable selection and clustering.
This thesis presents novel methodology for performing rank-based analyses on high-dimensional data, based on the Bayesian Mallows model with a focus on three methodological aspects: (i) performing variable selection, (ii) embedding covariate information, and (iii) jointly performing variable selection and clustering. All methods are applied to RNA-seq data from cancer patient samples, to show how the rank-based method can be a valid alternative in unsupervised analyses of complex biological data.
Additional information
Contact the research support staff.