Statistical Optimisation of Escherichia coli cell-Free Protein Synthesis Reaction Composition and Genetic Template Topology

Statistical engineering - the application of Design of Experiments and statistical modelling - is being used increasingly to understand and optimise biological systems. It has been applied successfully to metabolic pathways, codon use optimisation and enzyme functionality. Cell-free protein synthesis (CFPS) is also gaining popularity within the synthetic biology community as it holds advantages over cell-based systems as a means to rapidly prototype genetic circuits, synthesise high-value products, and provide point-of-care diagnostics. We have performed simultaneous investigations into the chemical composition of CFPS reactions and the gene construct topology of template DNA to identify factors impacting CFPS reaction performance. Specifically, we challenged our learning algorithm to optimise multiple objectives with the overall aim of developing rapid CFPS. Here we describe the implementation of a rapid experimental design-build-test cycle exploiting small-scale automated liquid handling, and standardised modular cloning coupled with machine learning, to explore a complex, multifactorial design space. The performance of a subset of experimental combinations was used to prime our learning algorithm, this directed further experimental design and characterisation. During each iteration the algorithm balanced exploration and validation of its understanding against the optimisation task. Its challenge is therefore to minimise the number of data-points required to inform the relative influence of factors on CFPS performance, whilst developing models capable of predicting effective combinations. Here we discuss the opportunities and challenges of embedding machine learning into experimental workflows.