Comparing metabolic engineering scenarios using simulated design-build-test-learn-cycles

Introduction: Design-Build-Test-Learn (DBTL) cycles are a widely employed engineering framework in metabolic engineering. Nonetheless, their performance depends on a wide range of experimental and algorithmic design choices, whose combined effects on the successful optimization of microbial strains remain an open question. Methods: In this study, we performed in silico DBTL cycles based on metabolic kinetic models to quantitatively assess how key process parameters affect strain optimization outcomes across four distinct metabolic pathway models. This includes parameters governing DNA library design, experimental budget limitations, and machine learning configuration. Results: The results show that screening capacity is a dominant driver of optimization success, whereas DNA sequencing capacity has surprisingly little impact, despite its importance for model training. Selecting top-producing strains for sequencing consistently outperforms stratified sampling, highlighting a trade-off between predictive accuracy and optimization efficiency. DNA library structure strongly affects performance: increasing the number of editable positions generally improves outcomes, while expanding the set of gene targets can hinder optimization due to increased dimensionality or sparse sampling. Discussion: Together, these findings offer actionable guidance for designing more effective DBTL workflows and underscore the value of simulation frameworks for exploring metabolic engineering strategies prior to experimental implementation.