Highly-multiplexed imaging technologies with single-cell resolution, such as imaging mass cytometry or multiplexed ion beam imaging, are revolutionising the field of biomedical research. These new technologies generate high-throughput, highly-multiplexed in-situ data, enabling the interrogation of biological systems at an unprecedented level of detail. In comparison to current single-cell technologies that analyse suspen- sions of single cells, extracted from cell cultures or tissues, these technologies do not only elucidate cellular compositions of tissues, but also recapitulate actual cell-cell interactions in their native microenvironment. Information about this cellular interactome creates new opportunities to study biological heterogeneity from a spatial perspective and to better understand disease mechanisms, paving the way for the exploration of new therapeutic approaches.
Similar to the transition from bulk sequencing to single-cell sequencing, these technologies introduce new computational challenges in terms of how the generated data is handled and analyzed. To fully exploit these data sets, novel computational methods and data standards need to be developed to accommodate the wide range of resolutions, multiplexing and modalities (e.g., genomic, transcriptomic, proteomic and metabolomic) produced by di↵erent technologies.
In this thesis, we develop a computational framework to visualise spatial omics data and characterize biological heterogeneity in a spatially informed way. Our framework introduces a new data structure, the SpatialOmics class, to store spatial data in a technology-agnostic way. The capabilities of the framework are demonstrated on a publicly available human breast cancer data set. We assess the methods provided in the toolbox on artificially generated data and report high robustness with respect to cell type distributions and high sensitivity regarding cell type interaction strength. Finally, the framework is employed to extract spatial heterogeneity patterns, observed in human breast cancer. However, prediction of various patient features based on spatially informed heterogeneity measures was not successful.