Celcomen is a newly developed computational framework that utilizes causal disentanglement and generative graph neural networks to analyze spatial transcriptomics data. The model is designed to separate intrinsic gene regulation within a single cell from extrinsic signals coming from the surrounding tissue environment. By establishing a mathematically identifiable causal structure, the tool can predict counterfactual outcomes, allowing researchers to simulate how tissues respond to genetic perturbations like knockouts. Scientists validated the model's accuracy using simulated data and real-world samples from human glioblastoma, fetal spleen, and mouse lung cancer. Ultimately, Celcomen serves as a robust foundation for building Virtual Tissues and gaining deeper insights into how diseases and therapies alter cellular communication.

References:

• Megas S, Chen D G, Polanski K, et al. Celcomen: spatial causal disentanglement for single-cell and tissue perturbation modeling[J]. Nature Communications, 2026.