Neural network modeling in optimisation of continuous fermentation processes

The capability of self-recurrent neural networks in dynamic modeling of continuous fermentation is investigated in this simulation study. In the past, feedforward neural networks have been successfully used as one-step-ahead predictors. However, in steady-state optimisation of continuous fermentations the neural network model has to be iterated to predict many time steps ahead into the future in order to get steady-state values of the variables involved in objective cost function, and this iteration may result in increasing errors. Therefore, as an alternative to classical feedforward neural network trained by using backpropagation method, self-recurrent multilayer neural net trained by backpropagation through time method was chosen in order to improve accuracy of long-term predictions. Prediction capabilities of the resulting neural network model is tested by implementing this into the Integrated System Optimisation and Parameter Estimation (ISOPE) optimisation algorithm. Maximisation of cellular productivity of the baker's yeast continuous fermentation was used as the goal of the proposed optimising control problem. The training and prediction results of proposed neural network and performances of resulting optimisation structure are demonstrated.     

A hybrid representation approach for modelling complex dynamic bioprocesses

This paper considers the use of hybrid models to represent the dynamic behaviour of biotechnological processes. Each hybrid model consists of a set of non linear differential equations and a neural model. The set of differential equations attempts to describe as much as possible the phenomenology of the process whereas neural networks model predict some key parameters that are an essential part of the phenomenological model. The neural model is obtained indirectly, that is, using the prediction errors of one or more state variables to adjust its weights instead of successive presentations of input-output data of the neural network. This approach allows to use actual measurements to derive a suitable neural model that not only represents the variation of some key parameters but it is also able to partly include dynamic behaviour unaccounted for by the phenomenological model. The approach is described in detail using three test cases: (1) the fermentation of glucose to gluconic acid by the micro-organism Pseudomonas ovalis, (2) the growth of filamentous fungi in a solid state fermenter, and (3) the propagation of filamentous fungi growing on a 2-D solid substrate. Results for the three applications clearly demon- strate that using a hybrid model is a viable alternative for modelling complex biotechnological bioprocesses.     

Yeast concentration estimation and prediction with static and dynamic neural network models in batch cultures

The second fermentation is one of the most important steps in Champagne production. For this purpose, yeasts are grown on a wine based medium to adapt their metabolism to ethanol. Several models built with various static and dynamic neural network configurations were investigated. The main objective was to achieve real-time estimation and prediction of yeast concentration during growth. The model selected, based on recurrent neural networks, was first order with respect to the yeast concentration and to the volume of CO₂ released. Temperature and pH were included as model parameters as well. Yeast concentration during growth could thus be estimated with an error lower than 3% (±1.7×10⁶ yeasts/ml). From the measurement of initial yeast population and temperature, it was possible to predict the final yeast concentration (after 21 hours of growth) from the beginning of the growth, with about ±3×10⁶ yeasts/ml accuracy. So a predictive control strategy of this process could be investigated.     

Using historical data for bioprocess optimization: modeling wine characteristics using artificial neural networks and archived process information

Optimization of fermentation processes is a difficult task that relies on an understanding of the complex effects of processing inputs on productivity and quality outputs. Because of the complexity of these biological systems, traditional optimization methods utilizing mathematical models and statistically designed experiments are less effective, especially on a production scale. At the same time, information is being collected on a regular basis during the course of normal manufacturing and process development that is rarely fully utilized. We are developing an optimization method in which historical process data is used to train an artificial neural network for correlation of processing inputs and outputs. Subsequently, an optimization routine is used in conjunction with the trained neural network to find optimal processing conditions given the desired product characteristics and any constraints on inputs. Wine processing is being used as a case study for this work. Using data from wine produced in our pilot winery over the past 3 years, we have demonstrated that trained neural networks can be used successfully to predict the yeast-fermentation kinetics, as well as chemical and sensory properties of the finished wine, based solely on the properties of the grapes and the intended processing. To accomplish this, a hybrid neural network training method, Stop Training with Validation (STV), has been developed to find the most desirable neural network architecture and training level. As industrial historical data will not be evenly spaced over the entire possible search space, we have also investigated the ability of the trained neural networks to interpolate and extrapolate with data not used during training. Because a company will utilize its own existing process data for this method, the result of this work will be a general fermentation optimization method that can be applied to fermentation processes to improve quality and productivity.     

Retrospective optimization of time-dependent fermentation control strategies using time-independent historical data

We have previously shown the usefulness of historical data for fermentation process optimization. The methodology developed includes identification of important process inputs, training of an artificial neural network (ANN) process model, and ultimately use of the ANN model with a genetic algorithm to find the optimal values of each critical process input. However, this approach ignores the time-dependent nature of the system, and therefore, does not fully utilize the available information within a database. In this work, we propose a method for incorporating time-dependent optimization into our previously developed three-step optimization routine. This is achieved by an additional step that uses a fermentation model (consisting of coupled ordinary differential equations (ODE)) to interpret important time-course features of the collected data through adjustments in model parameters. Important process variables not explicitly included in the model were then identified for each model parameter using automatic relevance determination (ARD) with Gaussian process (GP) models. The developed GP models were then combined with the fermentation model to form a hybrid neural network model that predicted the time-course activity of the cell and protein concentrations of novel fermentation conditions. A hybrid-genetic algorithm was then used in conjunction with the hybrid model to suggest optimal time-dependent control strategies. The presented method was implemented upon an E. coli fermentation database generated in our laboratory. Optimization of two different criteria (final protein yield and a simplified economic criteria) was attempted. While the overall protein yield was not increased using this methodology, we were successful in increasing a simplified economic criterion by 15% compared to what had been previously observed. These process conditions included using 35% less arabinose (the inducer) and 33% less typtone in the media and reducing the time required to reach the maximum protein concentration by 10% while producing approximately the same level of protein as the previous optimum.     

Application of neural networks for controlling and predicting quality parameters in beer fermentation

The biochemical pathways involved in the production of ethyl caproate, a secondary product of the beer fermentation process, are not well established. Hence, there are no phenomenological models available to control and predict the production of this particular compound as with other related products. In this work, neural networks have been used to fit experimental results with constant and variable pH, giving a good fit of laboratory and industrial scale data. The results at constant pH were also used to predict results at variable pH. Finally, the application of neural networks obtained from laboratory experiments gave excellent predictions of results in industrial breweries and so could be used in the control of industrial operations. The input pattern to the neural network included the accumulated fermentation time, cell dry weight, consumption of sugars and aminoacids and, in some cases, the pH. The output from the neural network was an estimation of quantity of the ethyl caproate ester.     

Toward intensifying design of experiments in upstream bioprocess development: An industrial Escherichia coli feasibility study

In this study, step variations in temperature, pH, and carbon substrate feeding rate were performed within five high cell density Escherichia coli fermentations to assess whether intraexperiment step changes, can principally be used to exploit the process operation space in a design of experiment manner. A dynamic process modeling approach was adopted to determine parameter interactions. A bioreactor model was integrated with an artificial neural network that describes biomass and product formation rates as function of varied fed‐batch fermentation conditions for heterologous protein production. A model reliability measure was introduced to assess in which process region the model can be expected to predict process states accurately. It was found that the model could accurately predict process states of multiple fermentations performed at fixed conditions within the determined validity domain. The results suggest that intraexperimental variations of process conditions could be used to reduce the number of experiments by a factor, which in limit would be equivalent to the number of intraexperimental variations per experiment. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:1343–1352, 2016     

A neural network study on the dynamic identification of a fermentation system

The objective of this paper is to propose neural networks for the study of dynamic identification and prediction of a fermentation system which produces mainly 2,3-butanediol (2,3-BDL). The metabolic products of the fermentation, acetic acid, acetoin, ethanol, and 2,3-BDL were measured on-line via a mass spectrometer modified by the insertion of a dimethylvinylsilicone membrane probe. The measured data at different sampling times were included as the input and output nodes, at different learning batches, of the network. A fermentation system is usually nonlinear and dynamic in nature. Measured fermentation data obtained from the complex metabolic pathways are often difficult to be entirely included in a static process model, therefore, a dynamic model was suggested instead. In this work, neural networks were provided by a dynamic learning and prediction process that moved along the time sequence batchwise. In other words, a scheme of two-dimensional moving window (number of input nodes by the number of training data) was proposed for reading in new data while forgetting part of the old data. Proper size of the network including proper number of input/output nodes were determined by trained with the real-time fermentation data. Different number of hidden nodes under the consideration of both learning performance and computation efficiency were tested. The data size for each learning batch was determined. The performance of the learning factors such as the learning coefficient η and the momentum term coefficient α were also discussed. The effect of different dynamic learning intervals, with different starting points and the same ending point, both on the learning and prediction performance were studied. On the other hand, the effect of different dynamic learning intervals, with the same starting point and different ending points, was also investigated. The size of data sampling interval was also discussed. The performance from four different types of transfer functions, x/(1+|x|), sgn(x)·x ²/(1+x ²), 2/(1+e ^{? x} )?1, and 1/(1+e ^{? x} ) was compared. A scaling factor b was added to the transfer function and the effect of this factor on the learning was also evaluated. The prediction results from the time-delayed neural networks were also studied.     

Modeling enzyme production with Aspergillus oryzae in pilot scale vessels with different agitation, aeration, and agitator types

The purpose of this article is to demonstrate how a model can be constructed such that the progress of a submerged fed-batch fermentation of a filamentous fungus can be predicted with acceptable accuracy. The studied process was enzyme production with Aspergillus oryzae in 550 L pilot plant stirred tank reactors. Different conditions of agitation and aeration were employed as well as two different impeller geometries. The limiting factor for the productivity was oxygen supply to the fermentation broth, and the carbon substrate feed flow rate was controlled by the dissolved oxygen tension. In order to predict the available oxygen transfer in the system, the stoichiometry of the reaction equation including maintenance substrate consumption was first determined. Mainly based on the biomass concentration a viscosity prediction model was constructed, because rising viscosity of the fermentation broth due to hyphal growth of the fungus leads to significant lower mass transfer towards the end of the fermentation process. Each compartment of the model was shown to predict the experimental results well. The overall model can be used to predict key process parameters at varying fermentation conditions.     

模拟青霉素发酵过程中菌体生长动态的细胞自动机模型

在青霉素发酵生产机理及其动力学微分方程模型的基础上,建立了模拟青霉素分批发酵过程中菌体生长动态的细胞自动机模型(CABGM)。CABGM采用三维细胞自动机作为菌体生长空间,采用Moore型邻域作为细胞邻域,其演化规则根据青霉素分批发酵过程中菌体生长机理和动力学微分方程模型设计。CABGM中的每一个细胞既可代表单个的青霉素产生菌,又可代表特定数量的青霉素产生菌,它具有不同的状态。对CABGM进行了统计特性的理论分析和仿真实验,理论分析和仿真实验结果均证明了CABGM能一致地复现动力学微分方程模型所描述的青霉素分批发酵菌体生长过程。最后,对所建模型在实际生产过程中的应用问题进行了分析,指出了需要进一步研究的问题。     

A robust soft sensor to monitor 1,3-propanediol fermentation process by Clostridium butyricum based on artificial neural network

With the aggravation of environmental pollution and energy crisis, the sustainable microbial fermentation process of converting glycerol to 1,3-propanediol (1,3-PDO) has become an attractive alternative. However, the difficulty in the online measurement of glycerol and 1,3-PDO creates a barrier to the fermentation process and then leads to the residual glycerol and therefore, its wastage. Thus, in the present study, the four-input artificial neural network (ANN) model was developed successfully to predict the concentration of glycerol, 1,3-PDO, and biomass with high accuracy. Moreover, an ANN model combined with a kinetic model was also successfully developed to simulate the fed-batch fermentation process accurately. Hence, a soft sensor from the ANN model based on NaOH-related parameters has been successfully developed which cannot only be applied in software to solve the difficulty of glycerol and 1,3-PDO online measurement during the industrialization process, but also offer insight and reference for similar fermentation processes.     

Neural network predictive control by MIMS monitored 2,3-butanediol fermentation by Klebsiella oxytoca

With the aid of a membrane introduction mass spectrometer (MIMS), the major product 2,3-butanediol (2,3-BDL) as well as the other metabolites from the fermentation carried by Klebsiella oxytoca can be measured on-line simultaneously. A backpropagation neural network (BPN) being recognized with superior mapping ability was applied to this control study. This neural network adaptive control differs from those conventional controls for fermentation systems in which the measurements of cell mass and glucose are not included in the network model. It is only the measured product concentrations from the MIMS that are involved. Oxygen composition was chosen to be the control variable for this fermentation system. Oxygen composition was directly correlated to the measured product concentrations in the controller model. A two-dimensional (number of input nodes by number of data sets) moving window for on-line, dynamic learning of this fermentation system was applied. The input nodes of the network were also properly selected. Number of the training data sets for obtaining better control results was also determined empirically. Two control structures for this 2,3-BDL fermentation are discussed and compared in this work. The effect from adding time delay element to the network controller was also investigated.     

Hybrid modeling as a QbD/PAT tool in process development: an industrial <Emphasis Type="Italic">E. coli</Emphasis> case study

Process understanding is emphasized in the process analytical technology initiative and the quality by design paradigm to be essential for manufacturing of biopharmaceutical products with consistent high quality. A typical approach to developing a process understanding is applying a combination of design of experiments with statistical data analysis. Hybrid semi-parametric modeling is investigated as an alternative method to pure statistical data analysis. The hybrid model framework provides flexibility to select model complexity based on available data and knowledge. Here, a parametric dynamic bioreactor model is integrated with a nonparametric artificial neural network that describes biomass and product formation rates as function of varied fed-batch fermentation conditions for high cell density heterologous protein production with E. coli. Our model can accurately describe biomass growth and product formation across variations in induction temperature, pH and feed rates. The model indicates that while product expression rate is a function of early induction phase conditions, it is negatively impacted as productivity increases. This could correspond with physiological changes due to cytoplasmic product accumulation. Due to the dynamic nature of the model, rational process timing decisions can be made and the impact of temporal variations in process parameters on product formation and process performance can be assessed, which is central for process understanding.     

Backpropagation neural network predictive control and control scheme comparison of 2,3-butanediol fermentation by Klebsiella oxytoca

A backpropagation neural network (BPN) was applied for the control study of 2,3-butanediol fermentation (2,3-BDL) carried by Klebsiella oxytoca. The measurements of cell mass and glucose were not included in the network models, instead, only the on-line measured product concentrations from the MIMS (membrane introduction mass spectrometer) were involved. Oxygen composition was chosen to be the control variable for this fermentation system for the formation of 2,3-BDL is regulated by oxygen. Oxygen composition was directly correlated to the measured product concentrations. A two-dimensional (number of input nodes by number of data sets) moving window to supply data for on-line, dynamic learning of this fermentation system was applied. The input nodes of the networks were also properly selected. Two neural network control schemes for this 2,3-BDL fermentation were discussed and compared in this work. Fermentations often exist time delay due to the measurement and their slow reaction nature. Hence, the order of time delay for the network controller was also investigated.     

A comprehensive model of simultaneous denitrification and methanogenic fermentation processes

The denitrification process was incorporated into the IWA Anaerobic Digestion Model No. 1 (ADM1) in order to account for the effect of denitrification on the methanogenic fermentation process. The model was calibrated and optimized using previously published experimental data and kinetic parameter values obtained with a mixed, mesophilic (35°C) methanogenic culture. Model simulations were used to predict the effect of nitrate reduction on the methanogenic fermentation process in batch, semi‐continuous, and continuous flow reactors experiencing operational changes and/or system disturbances. The extended model clearly revealed the importance of substrate competition between denitrifiers and non‐denitrifiers as well as the impact of N‐oxide inhibition on process interactions between fermentation, methanogenesis, and denitrification. Biotechnol. Bioeng. 2010;105: 98–108. © 2009 Wiley Periodicals, Inc.     

拉曼光谱分析技术在资源微生物发酵性能评价中的应用

微生物发酵过程是细胞新陈代谢进行物质转化的过程,为了提高目标产物的转化率,需要对微生物发酵动态特性进行实时分析,以便实时优化发酵过程。拉曼光谱(Raman spectroscopy)量化测试作为一种有应用前景的在线过程分析技术,可以在避免微生物污染的条件下,实现精准监测,进而用于优化控制微生物发酵过程。【目的】以运动发酵单胞菌(Zymomonas mobilis)为例,建立微生物发酵过程中葡萄糖、木糖、乙醇和乳酸浓度拉曼光谱预测模型,并进行准确性验证。【方法】采用浸入式在线拉曼探头,收集运动发酵单胞菌发酵过程中多个组分的拉曼光谱,采用偏最小二乘法对光谱信号进行预处理和多元数据分析,结合离线色谱分析数据,对拉曼光谱进行建模分析和浓度预测。【结果】针对运动发酵单胞菌,首先实现拉曼分析仪对单一产品乙醇发酵过程的精准检测,其次基于多元变量分析,建立葡萄糖、乙醇和乳酸浓度变化的预测模型,实现对发酵过程中各成分浓度变化的准确有效分析。【结论】成功建立了一种评价资源微生物尤其是工业菌株发酵液多种组分的拉曼光谱分析方法。该方法为运动发酵单胞菌等工业菌株利用多组分底物工业化生产不同产物的实时检测,以及其他微生物尤其工业菌株的选育和过程优化提供了新方法。     

谷氨酸发酵过程的神经网络模拟预测模型

人工神经网络是八十年代迅速兴起的一门非线性科学．它力图模拟人脑的一些基本特性。如自组织性、自适应性和容错性能等,已在模式识别、数据处理和自动化控制等方面得到了初步应用,取得了很好的效果〔1〕。 本文根据上海某味精厂某发酵罐的一批批报数据,利用人工神经网络的一典型模型－“反向传播”模型,初步尝试了神经网络模拟预测方法的效果,有关这方面的研究工作尚未见报道．     

Dynamic metagenome-scale metabolic modeling of a yogurt bacterial community

Genome-scale metabolic models and flux balance analysis (FBA) have been extensively used for modeling and designing bacterial fermentation. However, FBA-based metabolic models that accurately simulate the dynamics of coculture are still rare, especially for lactic acid bacteria used in yogurt fermentation. To investigate metabolic interactions in yogurt starter culture of Streptococcus thermophilus and Lactobacillus delbrueckii subsp. bulgaricus, this study built a dynamic metagenome-scale metabolic model which integrated constrained proteome allocation. The accuracy of the model was evaluated by comparing predicted bacterial growth, consumption of lactose and production of lactic acid with reference experimental data. The model was then used to predict the impact of different initial bacterial inoculation ratios on acidification. The dynamic simulation demonstrated the mutual dependence of S. thermophilus and L. d. bulgaricus during the yogurt fermentation process. As the first dynamic metabolic model of the yogurt bacterial community, it provided a foundation for the computer-aided process design and control of the production of fermented dairy products.     

Bioprocess model building using artificial neural networks

Artificial neural networks are made upon of highly interconnected layers of simple neuron-like nodes. The neurons act as non-linear processing elements within the network. An attractive property of artificial neural networks is that given the appropriate network topology, they are capable of learning and characterising non-linear functional relationships. Furthermore, the structure of the resulting neural network based process model may be considered generic, in the sense that little prior process knowledge is required in its determination. The methodology therefore provides a cost efficient and reliable process modelling technique. One area where such a technique could be useful is biotechnological systems. Here, for example, the use of a process model within an estimation scheme has long been considered an effective means of overcoming inherent on-line measurement problems. However, the development of an accurate process model is extremely time consuming and often results in a model of limited applicability. Artificial neural networks could therefore prove to be a useful model building tool when striving to improve bioprocess operability. Two large scale industrial fermentation systems have been considered as test cases; a fed-batch penicillin fermentation and a continuous mycelial fermentation. Both systems serve to demonstrate the utility, flexibility and potential of the artificial neural network approach to process modelling.     

A robust soft sensor based on artificial neural network for monitoring microbial lipid fermentation processes using Yarrowia lipolytica

Microbial oils produced by Yarrowia lipolytica offer an environmentally friendly and sustainable alternative to petroleum as well as traditional lipids from animals and plants. The accurate measurement of fermentation parameters, including the substrate concentration, dry cell weight, and lipid accumulation, is the foundation of process control, which is indispensable for industrial lipid production. However, it remains a great challenge to measure the complex parameters online during the lipid fermentation process, which is nonlinear, multivariate, and characterized by strong coupling. As a type of AI technology, the artificial neural network model is a powerful tool for handling extremely complex problems, and it can be employed to develop a soft sensor to monitor the microbial lipid fermentation process of Y. lipolytica. In this study, we first analyzed and emphasized the volume of sodium hydroxide and dissolved oxygen concentration as central parameters of the fermentation process. Then, a soft sensor based on a four-input artificial neural network model was developed, in which the input variables were fermentation time, dissolved oxygen concentration, initial glucose concentration, and additional volume of sodium hydroxide. This provides the possibility of online monitoring of dry cell weight, glucose concentration, and lipid production with high accuracy, which can be extended to similar fermentation processes characterized by the addition of bases or acids, as well as changes of the dissolved oxygen concentration.