Using Modular Ontologies to Capture Causal Knowledge contained in Bayesian Networks

Abstract

A Bayesian Network (BN) is a popular framework for causal studies. Causal relationships and interactions can be captured in the topology of a BN, creating a Causal Bayesian Network (CBN). This framework enables us to reason under uncertainty and capture the strength of causal links as conditional probabilities. However, there currently is no quick and efficient way to utilize the causal knowledge contained within a CBN, once it has been learned from data. In this paper we will examine a novel, conceptual approach that uses a modular ontology to store the probabilistic relationships among variables contained within a CBN. We will demonstrate this conceptual approach for patients of depression. This will be done by learning CBN structures from pre-existing National Institutes of Mental Health (NIMH) study on Sequenced Treatment Alternatives to Relieve Depression (STAR*D) patient dataset. Once we have a CBN, we will capture and standardize the patient variables in a modular ontology based on the structure and definitions found in the Medical Dictionary for Regulatory Activities Terminology (MedDRA).