Public Thesis defense - ICTEAM

Multivariate rare cell mining and biomarker identification from single-cell data by Alexander Gerniers

Identifying rare cell types is an important task to capture the heterogeneity of single-cell data. Indeed, many cell types are present in very small amounts, and are consequently easily missed by classical clustering approaches. As they may play a crucial role in some pathologies (such as cancer) their accurate identification is critical and has prompted the development of clustering methods specifically targeted to spot rare cells. These approaches are often restricted to be univariate due to computational restrictions and scalability concerns. In contrast, this thesis investigates the use of multivariate criteria, specifically, a global sum criterion which is robust with respect to the high technical and biological variability of scRNA-seq data. It jointly identifies rare cells and a corresponding set of genes characterizing them, which proves effective at identifying biologically relevant rare subpopulations of cells.

Different variations of this approach have also been developed. A hybrid procedure allows the use of this multivariate criterion in large-scale single-cell data in a computationally efficient way. The global sum criterion is also robust with respect to the batch effects that appear when combining single-cell data from multiple biological samples (such as multiple donors). Indeed, one generally needs to computationally integrate the different samples into one coherent dataset before applying clustering methods (which thus ignores the initial partition into multiple samples). As an alternative, this thesis shows how the sum criterion can be adapted to explicitly search for shared rare expression patterns across samples, without the need for prior data integration.

Jury members:

Prof. Pierre Dupont (UCLouvain), supervisor

Prof. Siegfried Nijssen (UCLouvain), supervisor

Prof. Charles Pecheur (UCLouvain), chairperson

Prof. Pierre Coulie (UCLouvain), secretary

Prof. Mark Robinson (University of Zurich, Suisse)

Prof. Yvan Saeys (UGent)

Pay attention:

The public defense of Alexander Gerniers scheduled for Tuesday October 22 at 04:30p.m. will also take place in the form of a video conference.

Télécharger l’annonce.