Celcomen: spatial causal disentanglement for single-cell and tissue perturbation modeling

Abstract

Abstract Celcomen leverages a mathematical causality framework to disentangle intra- and inter-cellular gene regulation programs in spatial transcriptomics data through a generative graph neural network. It is a first step towards perturbation models of Virtual Tissues and can generate post-perturbation counterfactual spatial transcriptomics, thereby offering access to experimentally inaccessible samples. We validated its disentanglement, identifiability of causal structure, and counterfactual prediction capabilities through simulations and in clinically relevant human glioblastoma, human fetal spleen, and mouse lung cancer samples. Celcomen provides the means to model disease- and therapy-induced changes allowing for new insights into single-cell spatially resolved tissue responses.