Location: Genomics and Bioinformatics Research
Project Number: 6066-21310-006-015-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 20, 2020
End Date: Sep 19, 2025
Objective:
Many agricultural commodities are used in bio-refining to create commercially important chemicals. The process often involves fermentation with genetically modified microorganisms and a lot of methods have been successfully applied to optimizing the enzymes themselves. Often these engineered enzymes are put into a microbe under the control of a single promoter. This does not allow for the fine-tuning of enzyme expression that can increase overall metabolic conversion in a multi-enzyme pathway. The Collaborator has developed a molecular technique to put multiple genes under the control of individual promoters. The challenge now is to determine how to induce those promoters to optimize product yields. In complex pathways systematic grid search rapidly becomes intractable. The project proposes to use an efficient learning method called Bayesian optimization to identify the optimal inducer concentrations in the fewest number of experimental steps. To speed up optimization further, each round of testing will be set up using an Opentrons robotic liquid handler and software will be written using integer linear programming optimization set up all inducer reactions and product concentration will be measured by plate reader. This will create a benchtop scale, promoter optimization system suitable to application in agro-industrial refining.
Approach:
The cooperator will design a prototype experimental system using the lycopene biosynthesis pathway from tomato placed in E. coli. The system will use five separate chemical inducers to induce 5 enzymes under the control of 5 promoters. Concentration-response curves are known for each promoter/inducer. These curves are logistic functions spanning several orders of magnitude in inducer concentration.
The cooperator will culture cells iteratively in 96 well plate format with varying concentrations of the 5 inducers and the lycopene produced will be assessed by absorbance at 471/502 nm on a plate reader, iteratively using software developed by ARS.
ARS will develop software for optimization and robotic control. Initially, 96 random sets of inducer concentrations will be selected. Custom software developed for the project will use integer linear programming (Gurobi solver) to determine the optimal pipetting scheme for each inducer set and pipette the 5 inducers into each well in a microtiter plate containing an inoculum of bacteria and medium. The software will control the Opentrons robot through the python API so that each well will be standardized to have 200 ul of 1x culture medium and uniform inoculum at the start of the experiment.
At the end of the growth period, lycopene in the supernatant will be optically measured by plate reader and results will be loaded into the custom software. From those values, a new batch of 96 inducer sets will be selected and pipetted out for the next round of experimentation. The software will use the package Scikit-optimize to implement a type of Bayesian optimization known as gaussian process regression. Once the system has converged on an optimized yield the experiment will be complete.
