Knowledge-based Bayesian network for the classification of Mycobacterium tuberculosis complex sublineages

We develop a novel knowledge-based Bayesian network (KBBN) that models our knowledge of the Mycobacterium tuberculosis complex (MTBC) obtained from expert-defined rules and large DNA fingerprint databases to classify strains of MTBC into fifty-one genetic sublineages. The model uses two high-throughput biomarkers: spacer oligonucleotide types (spoligotypes) and mycobacterial interspersed repetitive units (MIRU) types to represent strains of MTBC, since these are routinely gathered from MTBC isolates of tuberculosis (TB) patients. KBBN provides an elegant and simple way to incorporate existing widely accepted visual rules for MTBC sublineages into a classifier designed to capture known properties of the MTBC biomarkers. Unlike prior knowledge-based SVM approaches which require rules expressed as polyhedral sets, KBBN directly incorporates the rules without any modification. Computational results show that KBBN achieves much higher accuracy than methods based purely on rules, and than Bayesian networks trained on biomarker data alone.