Machine learning techniques to unveil and understand Pseudomonas aeruginosa survival mechanism in nutrient depleted water

Abstract

Pseudomonas aeruginosa is an organism notable for its ubiquity in the ecosystem and its resistance to antibiotics. It is an environmental bacterium that is a common cause of hospital-acquired infections. Identifying its survival mechanism is critical for designing preventative and curative measures. Also, understanding this mechanism is beneficial because P. aeruginosa and other related organisms are capable of bioremediation. To address this practical problem, we proceeded by decomposition into multiple learnable components: the optimal regulatory network models—static and temporal—of the survival mechanism, the functional interplay, the determination of the viability maintenance genes, and the identification of the bacterium’s states. With unlabeled data collected from P. aeruginosa gene expression response to low nutrient water, a Bayesian Networks Machine Learning methodology was implemented to model a static regulatory network of its survival process. Subsequently, node influence techniques were used to infer a group of genes as key orchestrators of the observed survival phenotype. Further, we proposed Dynamic Bayesian Networks for temporal modeling, clustering for the functional interplay, and hierarchical classification for the bacterium’s states identification.