The Strain Optimization API: A Flexible Strain Design Formalism for an Automated High-Throughput Industrial Pipeline

As a pioneer in making molecules in a sustainable way by strain engineering and industrial fermentation, Amyris has invested heavily in technology to develop strains and bring valuable molecules to the market at unprecedented speed. In the past three years, we have developed the Automated Scientist, a suite of mathematical models and machine learning algorithms that can iteratively perform the Design-Build-Test-Learn cycle for automated strain design and improvement. Thus far, the Automated Scientist has successfully created > 50,000 strains producing over 190 molecules. The design module of the Automated Scientist incorporates genome-scale metabolic models, an in-house knowledge store of design elements annotated with public data, and proprietary design algorithms to create pathway and strain designs. The pathway design to a target molecule can be visualized in the open-source tool Escher (first extended for industrial application at Amyris in 2014 by the author of the code) for route curation and additional flux balance simulation by biologists. To facilitate the exchange of knowledge between human scientists and the Automated Scientist, we have recently developed the Strain Optimization API. The API allows human scientists to specify the intended strain designs at a high level of abstraction, to validate their designs in silico, and to visualize the genotype-phenotype relationships. The Automated Scientist can parse the high-level input and automatically design the genetic architecture of the strains. In addition, the API can be accessed by Amyris’ high-throughput omics pipeline to identify strain-specific gene/protein/metabolite targets for testing up to 5,000 samples/day, and thus seamlessly accelerate the strain re-design process.