Combining machine learning and metabolic modelling for yeast bioprocessing

Recent research has shown that metabolic modelling utilising gene expression profiling combined with genetic information is more predictive for target metabolites than the latter alone [Zampieri et al., 2017]. This new field of research has gained notable attention [Phys.org, 2017] and is an extremely promising area, where machine learning techniques can aid in discerning genetic and media composition to maximise bioprocess yield.

Our objective is to propose and combine a novel machine learning 'kernel trick', based on a newly-defined metabolic distance with a poly-omic analysis on yeast cells. The resulting poly-omic machine-learning method will identify whether target cells have optimal conditions for a targeted protein growth, and predict genetic and media modifications that will increase the yield. We will train the method using gene expression data and the most up-to-date model of Saccharomyces cerevisiae metabolism.