Deep Learning for Fast Inference of Mechanistic Models' Parameters
Authors:
Maxim Borisyak,
Stefan Born,
Peter Neubauer,
Mariano Nicolas Cruz-Bournazou
Abstract:
Inferring parameters of macro-kinetic growth models, typically represented by Ordinary Differential Equations (ODE), from the experimental data is a crucial step in bioprocess engineering. Conventionally, estimates of the parameters are obtained by fitting the mechanistic model to observations. Fitting, however, requires a significant computational power. Specifically, during the development of ne…
▽ More
Inferring parameters of macro-kinetic growth models, typically represented by Ordinary Differential Equations (ODE), from the experimental data is a crucial step in bioprocess engineering. Conventionally, estimates of the parameters are obtained by fitting the mechanistic model to observations. Fitting, however, requires a significant computational power. Specifically, during the development of new bioprocesses that use previously unknown organisms or strains, efficient, robust, and computationally cheap methods for parameter estimation are of great value. In this work, we propose using Deep Neural Networks (NN) for directly predicting parameters of mechanistic models given observations. The approach requires spending computational resources for training a NN, nonetheless, once trained, such a network can provide parameter estimates orders of magnitude faster than conventional methods. We consider a training procedure that combines Neural Networks and mechanistic models. We demonstrate the performance of the proposed algorithms on data sampled from several mechanistic models used in bioengineering describing a typical industrial batch process and compare the proposed method, a typical gradient-based fitting procedure, and the combination of the two. We find that, while Neural Network estimates are slightly improved by further fitting, these estimates are measurably better than the fitting procedure alone.
△ Less
Submitted 5 December, 2023;
originally announced December 2023.
Fitting nonlinear models to continuous oxygen data with oscillatory signal variations via a loss based on DynamicTime War**
Authors:
Judit Aizpuru,
Annina Karolin Kemmer,
Jong Woo Kim,
Stefan Born,
Peter Neubauer,
Mariano N. Cruz Bournazou,
Tilman Barz
Abstract:
High throughput experimental systems play an important role in bioprocess development, as they provide an efficient way of analysing different experimental conditions and perform strain discrimination in previous phases to the industrial scale production. In the millilitre scale, these systems are combinations of parallel mini-bioreactors, liquid handling robots and automated workflows for data ha…
▽ More
High throughput experimental systems play an important role in bioprocess development, as they provide an efficient way of analysing different experimental conditions and perform strain discrimination in previous phases to the industrial scale production. In the millilitre scale, these systems are combinations of parallel mini-bioreactors, liquid handling robots and automated workflows for data handling and model based operation. For successfully monitoring cultivation conditions and improving the overall process quality by model-based approaches, a proper model identification is crucial. However, the quality and amount of measurements makes this task challenging considering the complexity of the bio-processes. TheDissolved Oxygen Tension is often the only measurement which is available online, and therefore, a good understanding of the errors in this signal is important for performing a robust estimation.Some of the expected errors will provoke uncertainties in the time-domain of the measurement, and in those cases, the common Weighted Least Squares estimation procedure can fail providing good results. Moreover, these errors will have even a larger effect in the fed-batch phase where bolus feeding is applied, as this generates fast dynamic responses in the signal. In the present work, an insilico study of the performance of Weighted Least Squares estimator is analysed when the expected time-uncertainties are present in the oxygen signal. As an alternative, a loss based on the Dynamic Time War** measure is proposed. The results show how this latter procedure outperforms the former reconstructing the oxygen signal, and in addition, returns less biased parameter estimates.
△ Less
Submitted 25 December, 2021;
originally announced December 2021.
