Search

• Typically, hybrid models use conventional machine learning approaches together with mechanistic modeling. • The so obtained hybrid models are not completely interpretable by domain experts. • The Functional-Hybrid model instead use library of domain-specific knowledge blocks. • Functional-Hybrid models are interpretable. • Compared to conventional hybrid model, they require less data and are superior in extrapolation. Mathematical models used for the representation of (bio)-chemical processes can be grouped into two broad paradigms: white-box or mechanistic models, completely based on knowledgeor black-box data-driven models based on patterns observed in data. However, in the past two-decade, hybrid modeling that explores the synergy between the two paradigms has emerged as a pragmatic compromise. The data-driven part of these has been largely based on conventional machine learning algorithms (e.g., artificial neural network, support vector regression), which prevents interpretability of the finally learnt model by the domain experts. In this work, we present a novel hybrid modeling framework, the Functional-Hybrid model, that uses the ranked domain-specific functional beliefs together with symbolic regression to develop dynamic models. We demonstrate the successful implementation of the Functional-Hybrid model and its interpretability, focusing on applying chemical reaction kinetic principles to classical chemical reactions, biochemistry, ecology, physiology, and a bioreactor. Furthermore, we demonstrate that during interpolation, the Functional-Hybrid model performs similarly to a Hybrid-ANN hybrid model implementing a conventional ANN. However, it provides the advantage of being –to some extent– interpretable, unlike the conventional Hybrid-ANN model. Additionally, it is shown that the Functional-Hybrid model outperforms the Hybrid-ANN model for a very low number of experiments, making it more suitable when data is scarce. Finally, the Functional-Hybrid models show superior extrapolation capabilities compared to the Hybrid-ANN model. This improved performance can be attributed to the structure imposed by the functional transformations introduced in the Functional-Hybrid model. [ABSTRACT FROM AUTHOR]

Pyrimidine-nucleoside phosphorylases (Py-NPases) have a significant potential to contribute to the economic and ecological production of modified nucleosides. These can be produced via pentose-1-phosphates, an interesting but mostly labile and expensive precursor. Thus far, no dynamic model exists for the production process of pentose-1-phosphates, which involves the equilibrium state of the Py-NPase catalyzed reversible reaction. Previously developed enzymological models are based on the understanding of the structural principles of the enzyme and focus on the description of initial rates only. The model generation is further complicated, as Py-NPases accept two substrates which they convert to two products. To create a well-balanced model from accurate experimental data, we utilized an improved high-throughput spectroscopic assay to monitor reactions over the whole time course until equilibrium was reached. We examined the conversion of deoxythymidine and phosphate to deoxyribose-1-phosphate and thymine by a thermophilic Py-NPase from Geobacillus thermoglucosidasius. The developed process model described the reactant concentrations in excellent agreement with the experimental data. Our model is built from ordinary differential equations and structured in such a way that integration with other models is possible in the future. These could be the kinetics of other enzymes for enzymatic cascade reactions or reactor descriptions to generate integrated process models. [ABSTRACT FROM AUTHOR]

• A framework for uncertainty analysis of biokinetic models is presented. • Parameter identifiability is deduced from the dynamic sensitivity matrix. • A mechanistic model of Escherichia coli was analyzed for parameter identifiability and uncertainty of model outputs. • Model prediction accuracy is improved up to 15 times, after regularization. Mechanistic models are simplifications of bio-physical systems, for which the true values of the model parameters are sometimes unknown. Therefore, before using model-based predictions to study or improve a process, it is essential to ensure that the outputs of the model are reliable. This paper covers the development and application of a framework for practical identifiability and uncertainty analyses of dynamic growth models for bioprocesses. By exploring the numerical properties of the sensitivity matrix, a simple algorithm to determine the presence of non-identifiable parameters in models with high output uncertainty is presented. The framework detects the existence of non-identifiable parameters within the model and proposes a regularisation technique, in conjunction with Monte Carlo Analysis. As an example, the framework was used to analyse a macro-kinetic growth model of Escherichia coli describing a fed-batch process. The results show a reduction in the uncertainty of model outputs from a maximum coefficient of variation of 748% to 5% after regularization, and a 15-fold improvement in the accuracy of model predictions for two independent validation datasets. The presented framework aims to improve the reliability of model predictions and promote a more thorough handling of dynamical models to extend their use in biotechnology. [ABSTRACT FROM AUTHOR]