Evaluating the Contribution of Model Complexity in\nPredicting Robustness in Synthetic Genetic Circuits

The\ndesign–build–test–learn workflow is pivotal\nin synthetic biology as it seeks to broaden access to diverse levels\nof expertise and enhance circuit complexity through recent advancements\nin automation. The design of complex circuits depends on developing\nprecise models and parameter values for predicting the circuit performance\nand noise resilience. However, obtaining characterized parameters\nunder diverse experimental conditions is a significant challenge,\noften requiring substantial time, funding, and expertise. This work\ncompares five computational models of three different genetic circuit\nimplementations of the same logic function to evaluate their relative\npredictive capabilities. The primary focus is on determining whether\nsimpler models can yield conclusions similar to those of more complex\nones and whether certain models offer greater analytical benefits.\nThese models explore the influence of noise, parametrization, and\nmodel complexity on predictions of synthetic circuit performance through\nsimulation. The findings suggest that when developing a new circuit\nwithout characterized parts or an existing design, any model can effectively\npredict the optimal implementation by facilitating qualitative comparison\nof designs’ failure probabilities (e.g., higher or lower).\nHowever, when characterized parts are available and accurate quantitative\ndifferences in failure probabilities are desired, employing a more\nprecise model with characterized parts becomes necessary, albeit requiring\nadditional effort.