A0726
Title: Enhance constraint spatial partitioning for spatial omics data
Authors: Pratheepa Jeganathan - McMaster University (Canada) [presenting]
Rajitha Senanayake - McMaster University (Canada)
Abstract: Spatial omics data, when preprocessed, identifies cell types across multiple tissues, providing crucial insights into the tumor microenvironment (TME). Traditional methods, such as constraint spatial hierarchical clustering and spatial latent Dirichlet allocation (sLDA), have characterized TME by classifying it in patients and associating it with clinical factors like survival time and treatment type. However, a significant limitation of these methods is their inability to uncover spatially separated partitions with similar cell-type distributions. To address this limitation, the enhanced constraint spatial partitioning (ECSP) approach is proposed, which leverages a soft constraint hierarchical clustering framework to merge similar partitions in cell-type distribution. Additionally, sLDA is demonstrated to be improved by incorporating a relational topic model framework to generate links after identifying cell partitions. The methods enhance the interpretability of the TME by effectively integrating spatial context. Their effectiveness is demonstrated using multiplexed ion beam imaging-time of flight (MIBI-TOF) data, highlighting their potential to improve TME characterization and clinical associations. These novel deterministic and probabilistic approaches promise to refine the analysis of spatial omics data, providing a deeper understanding of the spatial distribution and interaction of cell types within the tissues.
