scDisentangle: sequential disentanglement for single-cell perturbation response prediction

scDisentangle predicts single-cell transcriptional responses to perturbations by decomposing expression into observed attributes plus sequential residual components, enabling counterfactuals that preserve source cell identity.

• Why it matters: Accurate perturbation-response prediction is a key goal for single-cell modeling, but existing approaches often do not beat simple baselines and provide limited insight into how biological variation is structured.
• What they did: The method sequentially disentangles latent residual components to generate counterfactuals, and it was evaluated on cytokine stimulation, liver infection, and unseen combinatorial genetic perturbations.
• The result: It outperformed current models and naive baselines across those perturbation settings, and in an immune checkpoint inhibitor-associated myocarditis cohort it predicted donor-specific intracardiac CD8+ T cell states from peripheral blood.

AI summary of the abstract · may contain errors · not medical advice · read the paper.