Reproduce “Celcomen: spatial causal disentanglement for single-cell and tissue perturbation modeling” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Stathis Megas, Daniel G. Chen, Krzysztof Polanski, Moshe Eliasof, Carola-Bibiane Schonlieb, Sarah A. Teichmann
Celcomen introduces a generative graph neural network that mathematically disentangles intrinsic and extrinsic gene regulation from spatial transcriptomics and single-cell data. The model accurately predicted transcriptomic changes in mouse lung cancer lesions following in silico gene knockouts, correlating significantly with experimental knockout data.
Key insights
- A mathematically identifiable framework is essential for robustly disentangling intrinsic and extrinsic gene regulatory programs and inferring causal mechanisms from spatial omics data.
- Gene interactions can be successfully disentangled into intra- and inter-cellular components without prior knowledge of gene localization, with the model implicitly learning these distinctions.
- In silico perturbation modeling, when built upon a causally robust framework, can accurately predict spatially resolved transcriptomic changes in response to genetic interventions in complex tissues.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.