mc-ASTRA maps population-level variation to interpretable multicellular tissue organization processes

Abstract

Single-cell and spatial technologies increasingly enable detailed molecular profiling of tissues across biological contexts, creating opportunities to relate variation across samples to the multicellular processes that give rise to differences in tissue organisation and function. However, existing tools typically treat these aspects separately, either representing variation across samples or interpreting specific molecular and spatial features. To bridge this gap, we developed multicellular Analysis of Sample Tissue Representations and Associations (mc-ASTRA; https://github.com/saezlab/mc-astra), an open-source Python package for constructing maps of tissue-state variability that combine diverse tissue descriptors, including molecular, compositional and spatial features, together with complementary sample-level information such as clinical measurements or technical covariates, and biological prior knowledge of processes of interest. By combining available flexible semi-supervised factor models with established tools for the inference of biological processes from omics data, mc-ASTRA makes these tissue-state maps interpretable by quantifying how different tissue descriptors contribute to variation across samples and by resolving the specific coordinated multicellular programs underlying these differences. We illustrate the capabilities of mc-ASTRA across multiple single-cell and spatial omics datasets to infer multicellular programs that are contextualized by technical and biological prior knowledge and to describe distinct trajectories of tissue remodeling. mc-ASTRA provides a framework for building interpretable sample maps that connect tissue organization and multicellular mechanisms with variation observed across samples.