Network Structure Learning Under Uncertain Interventions

Gaussian Directed Acyclic Graphs (DAGs) represent a powerful tool for learning the network of dependencies among variables, task which is primary interest in many fields and specifically biology. Different DAGs may encode equivalent conditional independence structures, implying limited ability, with observational data, to identify causal relations. In contexts however, measurements are collected under heterogeneous settings where variables subject exogenous interventions. Interventional data can improve structure process whenever targets an intervention known. However, these often uncertain or completely unknown, as context drug target discovery. We propose Bayesian method dependence structures from interventions on unknown system. Selected features our approach include DAG-Wishart prior DAG parameters, use variable selection priors express uncertainty targets. provide theoretical results correct asymptotic identification derive sufficient conditions Bayes factor posterior ratio consistency graph structure. Our applied simulations real-data world settings, analyze perturbed protein assess antiepileptic therapies. Details MCMC algorithm proofs propositions provided supplementary materials, together more extensive studies. Supplementary materials this article available online.