Finding Small and Resistance-Predictable Feature Sets for Bacterium Classification by Multi-Objective Optimization

Recently, serious clinical issues have been occurred because of increasing drug-resistant pathogens. Haemophilus influenzae, which is one of the most important pathogens causing respiratory tract infection, has rapidly acquired resistance for the β-lactams by mutating its target gene as β-lactamasenonproducing ampicillin-resistant (BLNAR) strain. Accumulation of bacterial gene analysis data has made it possible to study the relationships between amino acid mutation patterns and drug resistance. We previously analyzed these data manually by focusing on mutations around well-known active sites and obtained a useful strain classification model that explains drug resistance using a small number of mutations[1, 2]. However, manual analysis is time-consuming and has a risk of overlooking an important mutation related to the resistance. In this study, we propose a new method which outputs bacterium classification models using a multi-objective optimization method.