Machine learning strategies for characterising the stress landscape of B.Subtilis

Bacteria almost always experience stresses when growing in all kinds of environments. Typically, hosting synthetic genetic circuits on a chassis will impose load stress which may affect the growth of the host and the desired function of the introduced system. This load stress may be very complex as it usually presents in the form of, not just one, but a number of stress responses. To allow for the design of load stress balancer, it is essential to understand how bacteria adjust their transcriptoms to adapt to various stress conditions. The full potential of this has not been explored for any species of bacteria.

In this talk, I will first introduce the large-scale condition-dependent gene expression dataset we use, then explain the machine learning strategies we applied to identify stress biomarkers and characterise the stress landscape, and lastly illustrate the challenges we encountered in data integrating and analysis.