PersiST: Robust Identification of Spatially Variable Features in Spatial Omics Datasets via Topological Data Analysis

Abstract: Spatial transcriptomics studies are becoming increasingly large and commonplace, necessitating the analysis of a large number of spatially resolved variables. With spatial transcriptomics data sets typically containing data on thousands of different genes, on increasingly large numbers of samples, there is a need for bioinformatics tools that enable the comparison of spatial structure across large numbers of variables. Here we present PersiST, an exploratory tool that uses topology to automatically compute a continuous measure of spatial structure for each gene in a spatial transcriptomics sample. This quantification can be used for analytical tasks such as spatially variable gene identification, or searching for spatial differences in the expression of a gene between samples. We use PersiST to derive biologically meaningful insights into two public spatial transcriptomics data sets, and we experiment with applying PersiST to a spatial metabolomics data set, making use of PersiST's non-parametric approach to enable application across different measurement types. Our work showcases the advantages of using a continuous quantification of spatial structure over p-value based approaches to SVG identification, the potential for developing unified methods for the analysis of different spatial `omics modalities, and the utility of persistent homology in big data applications.