CauTrigger: deciphering biological system state transitions by hierarchical causal decomposition.

Biological systems exhibit complex hierarchical regulation, with interactions across multiple layers governing transitions between states. Identifying the causal architecture underlying these state transitions remains challenging because confounding can produce spurious associations in observational data. Here we present CauTrigger, a hierarchical deep latent variable model that integrates do-calculus to identify potential causal triggers and predict perturbation outcomes across multiple regulatory layers. By disentangling causal and spurious latent factors, CauTrigger recursively reconstructs the underlying regulatory hierarchy and enables interpretable reasoning from observational data. Across synthetic and real-world datasets with multilayer regulatory structures, CauTrigger outperforms existing methods in identifying causal triggers and forecasting system-level responses. Applied to single-cell multi-omics and gut microbiome datasets, CauTrigger prioritizes candidate causal triggers and uses in silico perturbations to generate testable hypotheses about biological state transitions and inform potential strategies for state navigation. Together, these results establish CauTrigger as a general-purpose framework for causal discovery in complex biological systems.