Spatial transcriptomic analysis of diseased and healthy lungs with graph neural networks

Visit our website to learn more - www.mscience.c...
Assoc Prof Davis McCarthy
Head, Bioinformatics & Cellular Genomics
St Vincent's Institute Medical Research
Understanding molecular and structural heterogeneity in tissues is a key component of studying health and disease. Indeed, making progress towards new treatments for a deadly, progressive disease like idiopathic pulmonary fibrosis requires genetic and molecular analysis at high cellular and spatial resolution. Happily, modern ‘omics technologies provide the ability to characterise genetic and other high-dimensional molecular states at single-cell resolution, now also with spatial context. Rich, complex datasets are exciting, but bring with them deep challenges for winnowing the wheat from the chaff to answer biological questions of interest. In this talk, I will discuss using “traditional” statistical and recently developed deep learning approaches to spatial transcriptomic data in diseased and healthy lungs. I will present our use of graph neural network models (among other approaches) to characterise the molecular basis for tissue niche structure in lung fibrosis using 10x Xenium data on 45 lung samples. This analysis offers new insights into the spatial heterogeneity of gene expression in healthy and fibrotic regions of the lung and identifies “early transition” regions from healthy to disease states as the most promising area for clinical intervention.