Conditional independence relations among biomarkers: an extension of a published case study

We address the issue of conditional independence relationships among biomarkers in a published case study on molecular bioprofiling of breast cancer. The structure of Bayesian networks has been learned by scoring candidate structures with the BDe metric. A further characterization of patients was obtained by using a Naive Bayes clustering model that we fitted through a Markov Chain Monte Carlo simulation. In the five markers analysis we found that ER plays a key role, for example an important marker like P53 is conditionally independent from all the others given ER. Moreover the structures identified within clusters of patients suggest that heterogeneities might exist as regards conditional independence relations within population subgroups.