sosta: a framework to analyse spatial structures from spatial omics data

Author(s): Samuel Gunz,Mark Robinson

Affiliation(s): Department of Molecular Life Sciences, University of Zurich, Switzerland



Understanding the organization of tissue at a molecular level is important for studying the complexity of biological systems. Recent technological advances have made it possible to quantify the abundance of genes, transcripts and proteins not only at cellular resolution but also in their spatial context. Here, we introduce sosta (spatial omics structure analysis) a framework to reconstruct, characterise and compare spatial structures from spatial omics data. The first step in our workflow is to find the optimal reconstruction of structures based on molecular features or cell types in one or multiple samples. Next, based on the obtained reconstruction we calculate a set of metrics for each structure, including shape characteristics (e.g. area, orientation), relevant biological information (e.g. infiltration scores, border scores) or quantifications of marker/gene expression in and between substructures (e.g., gradients). Building on existing R packages for spatial analysis such as spatstat, sf and SpatialFeatureExperiment, sosta facilitates the implementation of additional custom metrics that fit the needs of users (e.g., quantification of specific spatial variation). In the final step of our workflow, we aim to guide users through statistical testing of spatial metrics across samples and conditions. Metrics calculated on spatial structures can harbor multiple levels of dependence, for example due to repeated measurements on slide or patient level. Therefore, it is crucial to select appropriate models and testing regimes to avoid false conclusions, as we show with simulations. We implement sosta as a toolbox to fit in the framework of Bioconductor and other R packages for spatial omics data analysis.