DynGFN: Bayesian Dynamic Causal Discovery using Generative Flow Networks

Abstract

Learning the causal structure of observable variables is a central focus for scientific discovery. Bayesian causal discovery methods tackle this problem by learning a posterior over the set of admissible graphs given our priors and observations. Existing methods primarily consider observations from static systems and assume the underlying causal structure takes the form of a directed acyclic graph (DAG). In settings with dynamic feedback mechanisms that regulate the trajectories of individual variables, this acyclicity assumption fails unless we account for time. We focus on learning Bayesian posteriors over cyclic graphs and treat causal discovery as a problem of sparse identification of a dynamical system. This imposes a natural temporal causal order between variables and captures cyclic feedback loops through time. Under this lens, we propose a new framework for Bayesian causal discovery for dynamical systems and present a novel generative flow network architecture (DynGFN) tailored for this task. Our results indicate that DynGFN learns posteriors that better encapsulate the distributions over admissible cyclic causal structures compared to counterpart state-of-the-art approaches.

WENDY: Covariance Dynamics Based Gene Regulatory Network Inference

“…For single-cell level time series data types with joint distribution of different time points (Scenarios 5, 10, 15), there have been numerous GRN inference methods for Scenario 5, while Scenarios 10 and 15 generally do not have extra information that support specific inference methods. For Scenario 5, most inference methods are similar to those for Scenario 2, especially those methods on regression [75], tree-based feature selection [29, 78], or more advanced machine learning tools [47, 3]. The difference is that the direction of a regulatory relationship can be determined in Scenario 5: if the level of V i at time t can predict the level of V j at time t + Δ t , then one knows that V i regulates V j , not the inverse.…”