Primary Submission Category: Applications in Health and Biology
Causal Modeling and Discovery of Brain Network Interactions
Authors: Ang Li,
Presenting Author: Ang Li*
Understanding the causal organization of brain networks is a central problem in cognitive neuroscience, as disruptions in these networks are linked to disorders such as Parkinson’s disease, Alzheimer’s disease, and epilepsy. Describing how communication between brain regions breaks down can help explain disease-related symptoms and inform markers of progression and potential intervention strategies. More broadly, identifying causal relationships among brain regions is essential for understanding brain function and cognition.
Most analyses of functional brain imaging data rely on correlation-based measures, which cannot distinguish true causal influence from spurious associations. We address this limitation by applying the Structural Causal Model framework and Probabilities of Causation to large-scale fMRI data collected without an explicit task. This allows interventional and counterfactual reasoning about brain network interactions, including counterfactual questions related to disease-associated changes.
We also introduce an expert-guided causal discovery strategy that incorporates expert knowledge into discovery based on conditional independence tests. When statistical tests leave edge directions unresolved, plausible orientations are propagated through the graph and evaluated for contradictions with expert knowledge, allowing implausible causal structures to be ruled out.