Data integration to model-based description of biological systems incorporating gene dynamics improves the performance of microbial systems. Bioprocess performance, typically predicted using empirical Monod-type models, is essential for a sustainable bioeconomy. To replace empirical models, we updated a hybrid gene regulatory network-growth kinetic model, predicting aromatic pollutants degradation and biomass growth in Pseudomonas putida mt-2. We modeled a complex biological system including extensive information to understand the role of the regulatory elements in toluene biodegradation and biomass growth. The updated model exhibited extra complications such as the existence of oscillations and discontinuities. As parameter estimation of complex biological models remains a key challenge, we used the updated model to present a dual-parameter identification approach (the 'dual approach') combining two independent methodologies. Approach I handled the complexity by incorporation of demonstrated biological knowledge in the model-development process and combination of global sensitivity analysis and optimisation. Approach II complemented Approach I handling multimodality, ill-conditioning and overfitting through regularisation estimation, global optimisation, and identifiability analysis. To systematically quantify the biological system, we used a vast amount of high-quality time-course data. The dual approach resulted in an accurately calibrated kinetic model (NRMSE: 0.17055) efficiently handling the additional model complexity. We tested model validation using three independent experimental data sets, achieving greater predictive power (NRMSE: 0.18776) than the individual approaches (NRMSE I: 0.25322, II: 0.25227) and increasing model robustness. These results demonstrated data-driven predictive modeling potentially leading to bioprocess' model-based control and optimisation.

中文翻译：

---

一种双参数识别方法，用于基于数据的恶臭假单胞菌mt-2杂种基因调控网络-生长动力学的预测建模。

数据整合到结合了基因动力学的生物系统的基于模型的描述中，可以改善微生物系统的性能。通常使用经验Monod型模型预测的生物过程性能对于可持续的生物经济至关重要。为了替换经验模型，我们更新了杂种基因调控网络-生长动力学模型，预测恶臭假单胞菌mt-2的芳香族污染物降解和生物量增长。我们对包括大量信息的复杂生物系统进行了建模，以了解调节元素在甲苯生物降解和生物量增长中的作用。更新的模型显示出额外的复杂性，例如振荡和不连续性的存在。由于复杂生物模型的参数估算仍然是一项关键挑战，我们使用更新后的模型来提出结合了两种独立方法的双参数识别方法（“双重方法”）。方法I通过将已证明的生物学知识纳入模型开发过程中，并结合了全局敏感性分析和优化来处理了复杂性。方法II是对方法I的补充，它通过正则化估计，全局优化和可识别性分析来处理多模态，病态和过度拟合。为了系统地量化生物系统，我们使用了大量高质量的时程数据。双重方法产生了精确校准的动力学模型（NRMSE：0.17055），可有效处理附加的模型复杂性。我们使用三个独立的实验数据集测试了模型验证，与单个方法（NRMSE I：0.25322，II：0.25227）相比，可实现更大的预测能力（NRMSE：0.18776），并提高了模型的鲁棒性。这些结果证明了数据驱动的预测模型可能导致生物过程基于模型的控制和优化。