Spatial Omics - leveraging the power of machine learning to identify genes and cell disease markers

Speaker: Quan Nguyen
Affiliation IMB, University of Queensland

Abstract

Spatial Transcriptomics (ST) is an emerging technology that adds spatial dimensionality to the genome-wide transcriptional profiling of cells in undissociated tissues. From a sample, ST can generate three spatial data types, namely morphological imaging, physical distance between spatial data points, and gene expression values. We developed three innovative algorithms, collectively referred to as stLearn, uniquely utilising all the three data types to find tissue organisation(s) and cell types, reconstruct spatiotemporal dynamic patterns within a tissue, and scan for tissue regions with a high likelihood of cell-to-cell interaction. First, stLearn uses tissue morphology and physical distance to correct for technical noise in spatial sequencing data, which significantly improves clustering analysis. Second, we present a novel method, pseudo-time-space (PSTS), to model the spatiotemporal relationship of cellular transcriptional states across a tissue. We validated PSTS in a mouse model of central nervous system injury, reconstructing the spatial trajectory of microglia activation following brain insult. We also assessed the diagnostic potential of PSTS in studying cancer progression. Lastly, we developed an algorithm in which ligand-receptor pairs and neighbourhood information of diverse cell types were combined to find highly interactive regions within a tissue, validating this in skin and breast cancer tissues. Together, the three algorithms that we developed, as implemented in the comprehensive and fast stLearn software, allow for the elucidation of biological processes within healthy and diseased tissues.