Incorporating prior expert knowledge in learning Bayesian networks from genetic epidemiological data

Abstract

We consider the applicability of Bayesian networks (BNs) for discovering relations between genes, environment, and disease. Most state-of-the-art BN structure learning algorithms are not capable of learning structures from data containing missing values, which is a norm in genetic epidemiological data. In addition, there exists a wealth of existing prior knowledge which could be incorporated to improve computational efficiency in BN structure learning. To address these challenges, we applied a Markov chain Monte Carlo based BN structure learning algorithm to data from a population-based study of bladder cancer in New Hampshire, USA. A large improvement in computational efficiency is achieved under this approach.