A reinforcement learning-based hybrid modeling framework for bioprocess kinetics identification.

Constructing predictive models to simulate complex bioprocess dynamics, particularly time-varying (i.e., parameters varying over time) and history-dependent (i.e., current kinetics dependent on historical culture conditions) behavior, has been a longstanding research challenge. Current advances in hybrid modeling offer a solution to this by integrating kinetic models with data-driven techniques. This article proposes a novel two-step framework: first (i) speculate and combine several possible kinetic model structures sourced from process and phenomenological knowledge, then (ii) identify the most likely kinetic model structure and its parameter values using model-free Reinforcement Learning (RL). Specifically, Step 1 collates feasible history-dependent model structures, then Step 2 uses RL to simultaneously identify the correct model structure and the time-varying parameter trajectories. To demonstrate the performance of this framework, a range of in-silico case studies were carried out. The results show that the proposed framework can efficiently construct high-fidelity models to quantify both time-varying and history-dependent kinetic behaviors while minimizing the risks of over-parametrization and over-fitting. Finally, the primary advantages of the proposed framework and its limitation were thoroughly discussed in comparison to other existing hybrid modeling and model structure identification techniques, highlighting the potential of this framework for general bioprocess modeling.

A transfer learning approach for predictive modeling of bioprocesses using small data.

Predictive modeling of new biochemical systems with small data is a great challenge. To fill this gap, transfer learning, a subdomain of machine learning that serves to transfer knowledge from a generalized model to a more domain-specific model, provides a promising solution. While transfer learning has been used in natural language processing, image analysis, and chemical engineering fault detection, its application within biochemical engineering has not been systematically explored. In this study, we demonstrated the benefits of transfer learning when applied to predict dynamic behaviors of new biochemical processes. Two different case studies were presented to investigate the accuracy, reliability, and advantage of this innovative modeling approach. We thoroughly discussed the different transfer learning strategies and the effects of topology on transfer learning, comparing the performance of the transfer learning models against benchmark kinetic and data-driven models. Furthermore, strong connections between the underlying process mechanism and the transfer learning model's optimal structure were highlighted, suggesting the interpretability of transfer learning to enable more accurate prediction than a naive data-driven modeling approach. Therefore, this study shows a novel approach to effectively combining data from different resources for bioprocess simulation.

Kinetic and hybrid modeling for yeast astaxanthin production under uncertainty.

Astaxanthin is a high-value compound commercially synthesized through Xanthophyllomyces dendrorhous fermentation. Using mixed sugars decomposed from biowastes for yeast fermentation provides a promising option to improve process sustainability. However, little effort has been made to investigate the effects of multiple sugars on X. dendrorhous biomass growth and astaxanthin production. Furthermore, the construction of a high-fidelity model is challenging due to the system's variability, also known as batch-to-batch variation. Two innovations are proposed in this study to address these challenges. First, a kinetic model was developed to compare process kinetics between the single sugar (glucose) based and the mixed sugar (glucose and sucrose) based fermentation methods. Then, the kinetic model parameters were modeled themselves as Gaussian processes, a probabilistic machine learning technique, to improve the accuracy and robustness of model predictions. We conclude that although the presence of sucrose does not affect the biomass growth kinetics, it introduces a competitive inhibitory mechanism that enhances astaxanthin accumulation by inducing adverse environmental conditions such as osmotic gradients. Moreover, the hybrid model was able to greatly reduce model simulation error and was particularly robust to uncertainty propagation. This study suggests the advantage of mixed sugar-based fermentation and provides a novel approach for bioprocess dynamic modeling.

Synergising biomass growth kinetics and transport mechanisms to simulate light/dark cycle effects on photo-production systems.

Light attenuation is a primary challenge limiting the upscaling of photobioreactors for sustainable bio-production. One key to this challenge, is to model and optimise the light/dark cycles so that cells within the dark region can be frequently transferred to the light region for photosynthesis. Therefore, this study proposes the first mechanistic model to integrate the light/dark cycle effects into biomass growth kinetics. This model was initially constructed through theoretical derivation based on the intracellular reaction kinetics, and was subsequently modified by embedding a new parameter, effective light coefficient, to account for the effects of culture mixing. To generate in silico process data, a new multiscale reactive transport modelling strategy was developed to couple fluid dynamics with biomass growth kinetics and light transmission. By comparing against previous experimental and computational studies, the multiscale model shows to be of high accuracy. Based on its simulation result, an original correlation was proposed to link effective light coefficient with photobioreactor gas inflow rate; this has not been done before. The impact of this study is that by using the proposed mechanistic model and correlation, we can easily control and optimise photobioreactor gas inflow rates to alleviate light attenuation and maintain a high biomass growth rate.

Combining model structure identification and hybrid modelling for photo-production process predictive simulation and optimisation.

Integrating physical knowledge and machine learning is a critical aspect of developing industrially focused digital twins for monitoring, optimisation, and design of microalgal and cyanobacterial photo-production processes. However, identifying the correct model structure to quantify the complex biological mechanism poses a severe challenge for the construction of kinetic models, while the lack of data due to the time-consuming experiments greatly impedes applications of most data-driven models. This study proposes the use of an innovative hybrid modelling approach that consists of a simple kinetic model to govern the overall process dynamic trajectory and a data-driven model to estimate mismatch between the kinetic equations and the real process. An advanced automatic model structure identification strategy is adopted to simultaneously identify the most physically probable kinetic model structure and minimum number of data-driven model parameters that can accurately represent multiple data sets over a broad spectrum of process operating conditions. Through this hybrid modelling and automatic structure identification framework, a highly accurate mathematical model was constructed to simulate and optimise an algal lutein production process. Performance of this hybrid model for long-term predictive modelling, optimisation, and online self-calibration is demonstrated and thoroughly discussed, indicating its significant potential for future industrial application.

Hybrid physics-based and data-driven modeling for bioprocess online simulation and optimization.

Model-based online optimization has not been widely applied to bioprocesses due to the challenges of modeling complex biological behaviors, low-quality industrial measurements, and lack of visualization techniques for ongoing processes. This study proposes an innovative hybrid modeling framework which takes advantages of both physics-based and data-driven modeling for bioprocess online monitoring, prediction, and optimization. The framework initially generates high-quality data by correcting raw process measurements via a physics-based noise filter (a generally available simple kinetic model with high fitting but low predictive performance); then constructs a predictive data-driven model to identify optimal control actions and predict discrete future bioprocess behaviors. Continuous future process trajectories are subsequently visualized by re-fitting the simple kinetic model (soft sensor) using the data-driven model predicted discrete future data points, enabling the accurate monitoring of ongoing processes at any operating time. This framework was tested to maximize fed-batch microalgal lutein production by combining with different online optimization schemes and compared against the conventional open-loop optimization technique. The optimal results using the proposed framework were found to be comparable to the theoretically best production, demonstrating its high predictive and flexible capabilities as well as its potential for industrial application.

CFD and kinetic-based modeling to optimize the sparger design of a large-scale photobioreactor for scaling up of biofuel production.

Microalgal biofuels have not yet achieved wide-spread commercialization, partially as a result of the complexities involved with designing and scaling up of their biosystems. The sparger design of a pilot-scale photobioreactor (120 L) was optimized to enable the scale-up of biofuel production. An integrated model coupling computational fluid dynamics and microalgal biofuel synthesis kinetics was used to simulate the biomass growth and novel biofuel production (i.e., bisabolene) in the photobioreactor. Bisabolene production from Chlamydomonas reinhardtii mutant was used as an example to test the proposed model. To select the optimal sparger configuration, a rigorous procedure was followed by examining the effects of sparger design parameters (number and diameter of sparger holes and gas flow rates) on spatially averaged bubble volume fraction, light intensity, friction velocity, power input, biomass concentration, and bisabolene production. The optimized sparger design increases the final biomass concentration by 18%, thereby facilitating the scaling up of biofuel production.

Review of advanced physical and data-driven models for dynamic bioprocess simulation: Case study of algae-bacteria consortium wastewater treatment.

Microorganism production and remediation processes are of critical importance to the next generation of sustainable industries. Undertaking mathematical treatment of dynamic biosystems operating at any spatial or temporal scale is essential to guarantee their performance and safety. However, constructing physical models remains a challenge due to the extreme complexity of process biological mechanisms. Data-driven models also encounter severe limitations because datasets from large-scale bioprocesses are often scarce without complete information and on a restricted operational space. To fill this gap, the current research compares the performance of advanced physical and data-driven models for dynamic bioprocess simulations subject to incomplete and scarce datasets, which to the best of our knowledge has never been addressed before. In specific, kinetic models were constructed by integrating different classic models, and state-of-the-art hyperparameter selection frameworks were developed to design artificial neural networks and Gaussian process regression models. An algae-bacteria consortium wastewater treatment process was selected to test the accuracy of these modeling strategies, as it is one of the most sophisticated biosystems due to the intricate mutualistic and competitive interactions. Based on the current results and available data, a heuristic model selection procedure is provided. This study paves the way to facilitate future bioprocess modeling.

Comparison of physics-based and data-driven modelling techniques for dynamic optimisation of fed-batch bioprocesses.

The development of digital bioprocessing technologies is critical to operate modern industrial bioprocesses. This study conducted the first investigation on the efficiency of using physics-based and data-driven models for the dynamic optimisation of long-term bioprocess. More specifically, this study exploits a predictive kinetic model and a cutting-edge data-driven model to compute open-loop optimisation strategies for the production of microalgal lutein during a fed-batch operation. Light intensity and nitrate inflow rate are used as control variables given their key impact on biomass growth and lutein synthesis. By employing different optimisation algorithms, several optimal control sequences were computed. Due to the distinct model construction principles and sophisticated process mechanisms, the physics-based and the data-driven models yielded contradictory optimisation strategies. The experimental verification confirms that the data-driven model predicted a closer result to the experiments than the physics-based model. Both models succeeded in improving lutein intracellular content by over 40% compared to the highest previous record; however, the data-driven model outperformed the kinetic model when optimising total lutein production and achieved an increase of 40-50%. This indicates the possible advantages of using data-driven modelling for optimisation and prediction of complex dynamic bioprocesses, and its potential in industrial bio-manufacturing systems.

Overproduction of L-tryptophan via simultaneous feed of glucose and anthranilic acid from recombinant Escherichia coli W3110: Kinetic modeling and process scale-up.

L-tryptophan is an essential amino acid widely used in food and pharmaceutical industries. However, its production via Escherichia coli fermentation suffers severely from both low glucose conversion efficiency and acetic acid inhibition, and to date effective process control methods have rarely been explored to facilitate its industrial scale production. To resolve these challenges, in the current research an engineered strain of E. coli was used to overproduce L-tryptophan. To achieve this, a novel dynamic control strategy which incorporates an optimized anthranilic acid feeding into a dissolved oxygen-stat (DO-stat) glucose feeding framework was proposed for the first time. Three original contributions were observed. Firstly, compared to previous DO control methods, the current strategy was able to inhibit completely the production of acetic acid, and its glucose to L-tryptophan yield reached 0.211 g/g, 62.3% higher than the previously reported. Secondly, a rigorous kinetic model was constructed to simulate the underlying biochemical process and identify the effect of anthranilic acid on both glucose conversion and L-tryptophan synthesis. Finally, a thorough investigation was conducted to testify the capability of both the kinetic model and the novel control strategy for process scale-up. It was found that the model possesses great predictive power, and the presented strategy achieved the highest glucose to L-tryptophan yield (0.224 g/g) ever reported in large scale processes, which approaches the theoretical maximum yield of 0.227 g/g. This research, therefore, paves the way to significantly enhance the profitability of the investigated bioprocess.

Dynamic modeling of green algae cultivation in a photobioreactor for sustainable biodiesel production.

Biodiesel produced from microalgae has been extensively studied due to its potentially outstanding advantages over traditional transportation fuels. In order to facilitate its industrialization and improve the process profitability, it is vital to construct highly accurate models capable of predicting the complex behavior of the investigated biosystem for process optimization and control, which forms the current research goal. Three original contributions are described in this paper. Firstly, a dynamic model is constructed to simulate the complicated effect of light intensity, nutrient supply and light attenuation on both biomass growth and biolipid production. Secondly, chlorophyll fluorescence, an instantly measurable variable and indicator of photosynthetic activity, is embedded into the model to monitor and update model accuracy especially for the purpose of future process optimal control, and its correlation between intracellular nitrogen content is quantified, which to the best of our knowledge has never been addressed so far. Thirdly, a thorough experimental verification is conducted under different scenarios including both continuous illumination and light/dark cycle conditions to testify the model predictive capability particularly for long-term operation, and it is concluded that the current model is characterized by a high level of predictive capability. Based on the model, the optimal light intensity for algal biomass growth and lipid synthesis is estimated. This work, therefore, paves the way to forward future process design and real-time optimization.

An efficient model construction strategy to simulate microalgal lutein photo-production dynamic process.

Lutein is a high-value bioproduct synthesized by microalga Desmodesmus sp. It has great potential for the food, cosmetics, and pharmaceutical industries. However, in order to enhance its productivity and to fulfil its ever-increasing global market demand, it is vital to construct accurate models capable of simulating the entire behavior of the complicated dynamics of the underlying biosystem. To this aim, in this study two highly robust artificial neural networks (ANNs) are designed for the first time. Contrary to conventional ANNs, these networks model the rate of change of the dynamic system, which makes them highly relevant in practice. Different strategies are incorporated into the current research to guarantee the accuracy of the constructed models, which include determining the optimal network structure through a hyper-parameter selection framework, generating significant amounts of artificial data sets by embedding random noise of appropriate size, and rescaling model inputs through standardization. Based on experimental verification, the high accuracy and great predictive power of the current models for long-term dynamic bioprocess simulation in both real-time and offline frameworks are thoroughly demonstrated. This research, therefore, paves the way to significantly facilitate the future investigation of lutein bioproduction process control and optimization. In addition, the model construction strategy developed in this research has great potential to be directly applied to other bioprocesses. Biotechnol. Bioeng. 2017;114: 2518-2527. © 2017 Wiley Periodicals, Inc.

Analysis of green algal growth via dynamic model simulation and process optimization.

Chlamydomonas reinhardtii is a green microalga with the potential to generate sustainable biofuels for the future. Process simulation models are required to predict the impact of laboratory-scale growth experiments on future scaled-up system operation. Two dynamic models were constructed to simulate C. reinhardtii photo-autotrophic and photo-mixotrophic growth. A novel parameter estimation methodology was applied to determine the values of key parameters in both models, which were then verified using experimental results. The photo-mixotrophic model was used to accurately predict C. reinhardtii growth under different light intensities and in different photobioreactor configurations. The optimal dissolved CO2 concentration for C. reinhardtii photo-autotrophic growth was determined to be 0.0643 g·L(-1) , and the optimal light intensity for algal growth was 47 W·m(-2) . Sensitivity analysis revealed that the primary factor limiting C. reinhardtii growth was its intrinsic cell decay rate rather than light attenuation, regardless of the growth mode. The photo-mixotrophic growth model was also applied to predict the maximum biomass concentration at different flat-plate photobioreactors scales. A double-exposure-surface photobioreactor with a lower light intensity (less than 50 W·m(-2) ) was the best configuration for scaled-up C. reinhardtii cultivation. Three different short-term (30-day) C. reinhardtii photo-mixotrophic cultivation processes were simulated and optimised. The maximum biomass productivity was 0.053 g·L(-1) ·hr(-1) , achieved under continuous photobioreactor operation. The continuous stirred-tank reactor was the best operating mode, as it provides both the highest biomass productivity and lowest electricity cost of pump operation.

Dynamic modelling of high biomass density cultivation and biohydrogen production in different scales of flat plate photobioreactors.

This paper investigates the scaling-up of cyanobacterial biomass cultivation and biohydrogen production from laboratory to industrial scale. Two main aspects are investigated and presented, which to the best of our knowledge have never been addressed, namely the construction of an accurate dynamic model to simulate cyanobacterial photo-heterotrophic growth and biohydrogen production and the prediction of the maximum biomass and hydrogen production in different scales of photobioreactors. To achieve the current goals, experimental data obtained from a laboratory experimental setup are fitted by a dynamic model. Based on the current model, two key original findings are made in this work. First, it is found that selecting low-chlorophyll mutants is an efficient way to increase both biomass concentration and hydrogen production particularly in a large scale photobioreactor. Second, the current work proposes that the width of industrial scale photobioreactors should not exceed 0.20 m for biomass cultivation and 0.05 m for biohydrogen production, as severe light attenuation can be induced in the reactor beyond this threshold.

Integrating Autoencoder and Heteroscedastic Noise Neural Networks for the Batch Process Soft-Sensor Design.

Viscosity represents a key product quality indicator but has been difficult to measure in-process in real-time. This is particularly true if the process involves complex mixing phenomena operated at dynamic conditions. To address this challenge, in this study, we developed an innovative soft sensor by integrating advanced artificial neural networks. The soft sensor first employs a deep learning autoencoder to extract information-rich process features by compressing high-dimensional industrial data and then adopts a heteroscedastic noise neural network to simultaneously predict product viscosity and associated uncertainty. To evaluate its performance, predictions of product viscosity were made for a number of industrial batches operated over different seasons. Furthermore, probabilistic machine learning techniques, including the Gaussian process and the Bayesian neural network, were selected to benchmark against the heteroscedastic noise neural network. Through comparison, it is found that the proposed soft-sensor has both high accuracy and high reliability, indicating its potential for process monitoring and quality control.