(516b) Machine Learning-Driven Pathway Optimization: Application of Automation in Industrial Biotechnology

Microbial biosynthesis of fuels and chemicals often requires heterologous expression of metabolic pathways. To achieve the highest efficiency, the pathway flux must be carefully tuned by changing the expression level of each enzyme in the pathway coupled with a high-throughput screening. However, the high-throughput screening is unavailable for most targets. Here, we present a versatile platform for pathway optimization where a machine learning algorithm directs the experimentation to find the highest producing combination with a minimum number of experiments. Compared with other methods like Exterior Derivative Estimation (EDE) used by Lee et al., [1], our machine learning algorithm was able to find the maximum of the test data with less than half of evaluations required by EDE model. As proof of concept, we sought to optimize the lycopene biosynthesis pathway with 24 different expression levels for each gene resulting in more than 13,000 different combinations. The expression level gradient was achieved by mutating both the T7 promoter and ribosome binding sites (RBS). Although we have developed the iBioFAB platform [2], assembly and characterization of all these combinations is very difficult and expensive, if not impossible. By adding the machine learning component and developing an automation friendly lycopene extraction assay, we can achieve an unprecedented level of automation in pathway optimization. This machine learning driven pathway optimization strategy can be applied to optimize any pathway of interest, especially when a high throughput screening method is not available.

1. Lee, M.E., et al., Expression-level optimization of a multi-enzyme pathway in the absence of a high-throughput assay. Nucleic Acids Res, 2013. 41(22): p. 10668-78.

2. Chao, R., Y. Yuan, and H. Zhao, Building biological foundries for next-generation synthetic biology. Sci China Life Sci, 2015. 58(7): p. 658-65.