Open Collections

Lee, Eric

University of British Columbia

In the course of tumour progression, normal and malignant cells of various kinds of cell types engage in complex patterns of cell-cell interactions creating the tumor microenvironment. The dynamics in the tumour microenvironment are tumour-driven and fosters cellular heterogeneity to modulate cancer behavior. With an accurate classification of the cell type composition and a deep investigation of the cell-cell interactions, we can characterize the heterogeneity in the tumor microenvironment and potentially elucidate mechanisms of immune invasion, tumour growth, and metastases. Analyses of single cells in suspension using mass cytometry is a current approach used to characterize previously unknown phenotypes, yet the data generated by this approach are disaggregated and do not retain the spatial structure of tumours. Emerging high-throughput spatial expression profiling technologies, such as imaging mass cytometry allow for spatially aware profiling of single cell expression in high-parameter space. Various cell type classification methods have been proposed for disaggregated data, however there is a need for spatially aware clustering methods. We present SpatialSort, a scalable joint approach for spatially aware clustering of cell types and estimation of cell-cell interactions in the tumour microenvironment. This computational approach leverages a Markov random field model to allow spatially proximal cells linked in a neighbour graph to influence the cluster assignment of their neighbouring cells. Markov chain Monte Carlo sampling is employed to approximate the posterior distribution and perform probabilistic cell type identification. Cell to cell interactions will be encoded as interpretable model parameters representing the affinity between different cell types. Through spatially aware clustering, we hope to characterize patient-specific phenotypic heterogeneity better than our current methods. As heterogeneity promotes therapeutic resistance, an improved understanding of cellular composition and cell-cell interaction profiles can potentially provide better prognosis for cancer patients.

Text

2023-01-31

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/80453

Bioinformatics

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/