Zymergen develops genetically engineered microbes that can produce molecules used in industries like electronics, agriculture, and pharmaceuticals, among others. To do so, Zymergenc must identify strains (genetically distinct microbes) that are most likely to produce high concentrations of the desired molecule. To estimate the true performance of a strain from high-throughput screening data (e.g., where Zymergen tests hundreds to thousands of strains per experiment), their team uses a Bayesian modeling framework. This technique is designed to minimize process-driven noise. In this post, Aisha Ellahi explains why Bayesian modeling is well suited for this use case (e.g., to enable hierarchical modeling, incorporation of prior knowledge, and uncertainty estimation). In addition, she clarifies how Zymergen validates its Bayesian models through simulation and comparison with observations from strains selected for large-scale testing in fermentation tanks.