gcFront: a tool for determining a Pareto front of growth-coupled cell factory designs

Abstract

Motivation: A widely applicable strategy to create cell factories is to knockout (KO) genes or reactions to redirect cell metabolism so that chemical synthesis is made obligatory when the cell grows at its maximum rate. Synthesis is thus growth-coupled, and the stronger the coupling the more deleterious any impediments in synthesis are to cell growth, making high producer phenotypes evolutionarily robust. Additionally, we desire that these strains grow and synthesize at high rates. Genome-scale metabolic models can be used to explore and identify KOs that growth-couple synthesis, but these are rare in an immense design space, making the search difficult and slow.

Results: To address this multi-objective optimization problem, we developed a software tool named gcFront-using a genetic algorithm it explores KOs that maximize cell growth, product synthesis and coupling strength. Moreover, our measure of coupling strength facilitates the search so that gcFront not only finds a growth-coupled design in minutes but also outputs many alternative Pareto optimal designs from a single run-granting users flexibility in selecting designs to take to the lab.

Availability and implementation: gcFront, with documentation and a workable tutorial, is freely available at GitHub: https://github.com/lLegon/gcFront and archived at Zenodo, DOI: 10.5281/zenodo.5557755.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

gcFront finds many Pareto optimal growth-coupled designs, faster and with superior performance versus other algorithms. The speed and designs found from 6-h runs of gcFront were compared to those of RobustKnock, gcOpt, FastPros and OptGene (see Supplementary Note S4) on a MacBook Pro (2.3 GHz Quad-Core Intel core i5 processor, 8 GB 2133 MHz LPDDR3 RAM). Designs were based on KOs of only non-essential, gene-associated reactions, for the synthesis of three example products: succinate, tyrosine and pyruvate from the E.coli iML1515 GSM model, in aerobic, minimal media with glucose. (a) Time to identify the first gc-design from each procedure. Due to the stochastic nature of searching using the genetic algorithm in OptGene and gcFront, the average (bars) and standard deviation (error bars) of times are reported from three runs (N = 3, ±SD). (b) Pareto fronts of all gc-designs found from three 6-h runs