No abstract

Engineering microbes to synthesize molecules of societal value has historically been a time consuming and artisanal process, with the synthesis of each new non-native molecule typically warranting its own separate publication. Because most microbial strain engineering efforts leverage a finite number of common metabolic engineering design tactics, we reasoned that automating these design steps would help create a pipeline that can quickly, cheaply, and reliably generate so-called microbial factories. In this work we describe the design and implementation of a computational system, an Automated Scientist we call Lila, which handles all metabolic engineering design and optimization through the design-build-test-learn (DBTL) paradigm. Lila generates metabolic routes, identifies relevant genetic elements for perturbation, and specifies the design and re-design of microbial strains in a matter of seconds to minutes. Strains specified by Lila are then built and subsequently phenotyped as part of a largely automated in-house pipeline. Humans remain in-the-loop to curate choices made by the system, helping for example to refine the metabolic model or suggest custom protein modifications. Lila attempted to build strains that could produce 454 biochemically diverse molecules with precursors located broadly throughout the metabolism of two microbial hosts, Saccharomyces cerevisiae and Escherichia coli. Notably, we observed the highest published titers for the molecule naringenin, the metabolic precursor to flavonoids. In total we created hundreds of thousands of microbial strains capable of overproducing 242 molecules, of which 180 are not native to S. cerevisiae or E. coli.