Simulated Design–Build–Test–Learn Cycles for Consistent Comparison of Machine Learning Methods in Metabolic Engineering

Abstract

Combinatorial pathway optimization is an important tool in metabolic flux optimization. Simultaneous optimization of a large number of pathway genes often leads to combinatorial explosions. Strain optimization is therefore often performed using iterative design–build–test–learn (DBTL) cycles. The aim of these cycles is to develop a product strain iteratively, every time incorporating learning from the previous cycle. Machine learning methods provide a potentially powerful tool to learn from data and propose new designs for the next DBTL cycle. However, due to the lack of a framework for consistently testing the performance of machine learning methods over multiple DBTL cycles, evaluating the effectiveness of these methods remains a challenge. In this work, we propose a mechanistic kinetic model-based framework to test and optimize machine learning for iterative combinatorial pathway optimization. Using this framework, we show that gradient boosting and random forest models outperform the other tested methods in the low-data regime. We demonstrate that these methods are robust for training set biases and experimental noise. Finally, we introduce an algorithm for recommending new designs using machine learning model predictions. We show that when the number of strains to be built is limited, starting with a large initial DBTL cycle is favorable over building the same number of strains for every cycle.

Machine Learning-Guided Optimization of p-Coumaric Acid Production in Yeast

“…Another challenge to combinatorial pathway optimization is the need for the characterization of genetic parts that ensure that the solution space is sufficiently explored. This is especially important when the aim is to fine-tune the expression levels of pathway genes. ,,, In principle, the optimization of gene expression would benefit from the use of quantitative variables as factors (e.g., GFP fluorescence, protein levels) as they would allow the identification of an optimal expression level . However, although effort is taken to appropriately characterize how regulatory elements affect gene expression, this is seldom achieved as in vivo expression depends on factors such as the downstream gene or the gene order in an operon and cannot be accurately predicted.…”

“… 5 , 8 , 10 , 12 In principle, the optimization of gene expression would benefit from the use of quantitative variables as factors (e.g., GFP fluorescence, protein levels) as they would allow the identification of an optimal expression level. 25 However, although effort is taken to appropriately characterize how regulatory elements affect gene expression, this is seldom achieved as in vivo expression depends on factors such as the downstream gene 26 or the gene order in an operon 3 and cannot be accurately predicted. Alternatively, regulatory elements can be treated as categorical variables reducing the impact of the characterization data.…”

Mid–Long Chain Dicarboxylic Acid Production via Systems Metabolic Engineering: Progress and Prospects

“…Changes in these characteristics may alter the speed of microbial reproduction and shorten the fermentation cycle. , Specific to DCA synthesis, by using a rational design synthetic biology strategy called “Design–Build–Test–Learn (DBTL),” the unnecessary coding regions and noncoding regions in the genome of C. tropicalis can be deleted on a large scale. The energy metabolism required for growth and the ω-oxidation conducive to DCA formation can be retained to obtain chassis cells with a “minimum genome”. − Of course, before the genome is simplified, bioinformatics must be applied to analyze the metabolic network model and maximize the biosynthesis of DCAs by satisfying the most basic growth metabolism requirements …”

teemi: An open-source literate programming approach for iterative design-build-test-learn cycles in bioengineering

“…Complementary to this, DeepLearning algorithms have been used in a plethora of bioengineering applications notably to predict guide-RNA activity for CRISPR/Cas-based genome engineering [40] and gene regulation [41]. Additionally, stacked ensembles, distributed random forest (DRF), general linear models (GLMs), and gradient booster models (GBM) models have also found applications in various biological areas [42,43].…”