Characterizing spatial functional microniches with SpaceTravLR 2260916

Abstract

Abstract Introduction The advent of spatial omics has revolutionized our understanding of tissue biology. While in-silico perturbation methods and foundation models aim to model the impact of genetic perturbations, these methods are limited to single-cell approaches lacking spatial resolution. We address this major unmet need by developing SpaceTravLR, a novel interpretable machine learning approach that generalizes across tissues and species, uncovering spatial microniches linked to functional outcomes. Methods SpaceTravLR predicts how single or combinatorial genetic perturbations rewire signals across the tissue neighborhood, defining novel spatial niches across a range of tissues at different scales of organization, disease, and developmental contexts. All predictions are made solely based on context-specific spatial omic data with no prior knowledge, yet they align closely with findings from mechanistic experiments. Critically, interpretable machine learning enables the generation of mechanistic hypotheses underlying identified niches. Results SpaceTravLR discovered a novel mechanism for CCR4 that could not be captured by existing approaches that drives the compartmentalization and spatial location of a pathogenic population of Th2 cells in allergic asthma, which was validated in a murine model. Conclusion Overall, SpaceTravLR provides a novel interpretable and experimentally validated framework for uncovering how genes act individually and combinatorially through cell-intrinsic and cell-extrinsic circuits to shape spatial tissue organization and function. Funding Source n/a Topic Categories Computational and Systems Immunology (COMP)