Inferring functionally relevant molecular tissue substructures by agglomerative clustering of digitized spatial transcriptomes

Abstract

Developments on spatial transcriptomics (ST) are providing means to interrogate organ/tissue architecture from the angle of the gene programs defining their molecular complexity. However, computational methods to analyze ST data under-exploits the spatial signature retrieved within the maps. Inspired by contextual pixel classification strategies applied to image analysis, we have developed MULTILAYER, allowing to stratify ST maps into functionally-relevant molecular substructures. For it, MULTILAYER applies agglomerative clustering strategies within contiguous locally-defined transcriptomes (herein defined as gene expression elements or Gexels), combined with community detection methods for graph partitioning. MULTILAYER has been evaluated over multiple public ST data, including developmental tissues but also tumor biopsies. Its performance has been challenged for the processing of high-resolution ST maps and it has been used for an enhanced comparison of multiple public tissue biopsies issued from a cancerous prostate. MULTILAYER provides a digital perspective for the analysis of spatially-resolved transcriptomes and anticipates the application of contextual gexel classification strategies for developing self-supervised molecular diagnostics solutions. Overall, the development of MULTILAYER anticipates the application of contextual gexel classification strategies for developing self-supervised molecular diagnostics solutions.