Machine Learning Enabled Functional Discovery in Yeast Systems Biology

Abstract: Saccharomyces cerevisiae is a well-studied organism, yet roughly 20 percent of its proteins remain poorly characterized. Recent studies also seem to indicate that the pace of functional discovery is slow. Previous work has implied that the most probable path forward is via not only regular automation but fully autonomous systems that can automatically guide and perform high-throughput experimentation. This thesis explores various concepts to accelerate and perform functional discovery of gene and protein functions in Saccharomyces cerevisiae . It does so by combining ideas from artificial intelligence, such as active learning, with highthroughput analytical techniques like mass-spectrometry. The work performed as the basis for this thesis also served to aid in the further characterization of different aspects of yeast systems biology. Specifically, it delved into the diauxic shift and its regulators through the lens of untargeted metabolomics, as well as the regulatory patterns behind genome-wide intracellular proteomic abundances. We find that it is essential not only to develop tools and techniques for facilitating high-throughput experimentation, but also to ensure their optimal utilization of already existing knowledge. It is also of paramount importance to ensure a holistic and encompassing view of systems biology by more fully integrating and using different levels of cellular organization and analytical techniques.