Machine learning approach to recognition of detrimental cellular states in synthetic organisms

Synthetic biology employs engineering principles to design biological systems that do not exist in nature. For example, a genetic circuit can be inserted into a host organism to produce materials with desired properties. To optimise synthetic circuits performance, it is essential to identify harmful cellular states in engineered host organisms, as these can lead to a lower yield. Recent advances in omics technologies have resulted with vast amounts of high-throughput datasets, enabling the use of data mining techniques to explore the roles of various molecules within cells and in identification of high-level cellular states.

In this talk, I will introduce the machine learning methods we developed to identify minimal transcriptional biomarker panels that can detect different detrimental states in E. coli and B. subtilis (two of the most commonly used organisms in synthetic biology). I will first focus on identifying biomarkers specific to a load stress state induced by heterologous gene expression in E. coli. Then, I will move on to recognising various stress states that may be present in B. subtilis grown under diverse conditions. Finally, I will describe our novel computational strategies for prioritizing a biomarker solution when multiple candidate solutions are available.