Observability analysis of biochemical process models as a valuable tool for the development of mechanistic soft sensors

By enabling the estimation of difficult‐to‐measure target variables using available indirect measurements, mechanistic soft sensors have become important tools for various bioprocess monitoring and control scenarios. Despite promising higher process efficiencies and increased process understanding, widespread application of soft sensors has been stalled by uncertainty about the feasibility and reliability of their estimations given present process analytical constraints. Observability analysis can provide an indication of the possibility and reliability of soft sensor estimations by analyzing the structural properties of first‐principle (mechanistic) models. In addition, it can provide a criteria for selection of suitable measurement methods with respect to their information content; thereby leading to successful implementation of soft sensors in bioprocess development and manufacturing environments. We demonstrate the utility of observability analysis for two classes of upstream bioprocesses: the processes involving growth and ethanol formation by Saccharomyces cerevisiae and the process of penicillin production by Penicillium chrysogenum. Results obtained from laboratory‐scale cultivations in addition to in‐silico experiments enable a comparison of theoretical aspects of observability analysis and the real‐life performance of soft sensors. By taking the expected error of measurements provided to the soft sensor into account, an innovative scaling approach facilitates a higher degree of comparability of observability results among various measurement configurations and process conditions. © 2015 American Institute of Chemical Engineers Biotechnol. Prog., 31:1703–1715, 2015     

Comparison of Modeling Methods for DoE‐Based Holistic Upstream Process Characterization

Upstream bioprocess characterization and optimization are time and resource‐intensive tasks. Regularly in the biopharmaceutical industry, statistical design of experiments (DoE) in combination with response surface models (RSMs) are used, neglecting the process trajectories and dynamics. Generating process understanding with time‐resolved, dynamic process models allows to understand the impact of temporal deviations, production dynamics, and provides a better understanding of the process variations that stem from the biological subsystem. The authors propose to use DoE studies in combination with hybrid modeling for process characterization. This approach is showcased on Escherichia coli fed‐batch cultivations at the 20L scale, evaluating the impact of three critical process parameters. The performance of a hybrid model is compared to a pure data‐driven model and the widely adopted RSM of the process endpoints. Further, the performance of the time‐resolved models to simultaneously predict biomass and titer is evaluated. The superior behavior of the hybrid model compared to the pure black‐box approaches for process characterization is presented. The evaluation considers important criteria, such as the prediction accuracy of the biomass and titer endpoints as well as the time‐resolved trajectories. This showcases the high potential of hybrid models for soft‐sensing and model predictive control.     

Biomass soft sensor for a Pichia pastoris fed-batch process based on phase detection and hybrid modeling

A common control strategy for the production of recombinant proteins in Pichia pastoris using the alcohol oxidase 1 (AOX1) promotor is to separate the bioprocess into two main phases: biomass generation on glycerol and protein production via methanol induction. This study reports the establishment of a soft sensor for the prediction of biomass concentration that adapts automatically to these distinct phases. A hybrid approach combining mechanistic (carbon balance) and data-driven modeling (multiple linear regression) is used for this purpose. The model parameters are dynamically adapted according to the current process phase using a multilevel phase detection algorithm. This algorithm is based on the online data of CO₂ in the off-gas (absolute value and first derivative) and cumulative base feed. The evaluation of the model resulted in a mean relative prediction error of 5.52% and R² of .96 for the entire process. The resulting model was implemented as a soft sensor for the online monitoring of the P. pastoris bioprocess. The soft sensor can be used for quality control and as input to process control systems, for example, for methanol control.     

Process adapted calibration improves fluorometric pH sensor precision in sophisticated fermentation processes

Miniaturization and automation have become increasingly popular in bioprocess development in recent years, enabling rapid high‐throughput screening and optimization of process conditions. In addition, advances in the bioprocessing industry have led to increasingly complex process designs, such as pH and temperature shifts, in microbial fed‐batch fermentations for optimal soluble protein expression in a range of hosts. However, in order to develop an accurate scale‐down model for bioprocess screening and optimization, small‐scale bioreactors must be able to accurately reproduce these complex process designs. Monitoring methods, such as fluorometric‐based pH sensors, provide elegant solutions for the miniaturization of bioreactors, however, previous research suggests that the intrinsic fluorescence of biomass alters the sigmoidal calibration curve of fluorometric pH sensors, leading to inaccurate pH control. In this article, we present results investigating the impact of biomass on the accuracy of a commercially available fluorometric pH sensor. Subsequently, we present our calibration methodology for more precise online measurement and provide recommendations for improved pH control in sophisticated fermentation processes.     

Advanced near‐infrared monitor for stable real‐time measurement and control of Pichia pastoris bioprocesses

Near‐infrared spectroscopy is considered to be one of the most promising spectroscopic techniques for upstream bioprocess monitoring and control. Traditionally the nature of near‐infrared spectroscopy has demanded multivariate calibration models to relate spectral variance to analyte concentrations. The resulting analytical measurements have proven unreliable for the measurement of metabolic substrates for bioprocess batches performed outside the calibration process. This paper presents results of an innovative near‐infrared spectroscopic monitor designed to follow the concentrations of glycerol and methanol, as well as biomass, in real time and continuously during the production of a monoclonal antibody by a Pichia pastoris high cell density process. A solid state instrumental design overcomes the ruggedness limitations of conventional interferometer‐based spectrometers. Accurate monitoring of glycerol, methanol, and biomass is demonstrated over 274 days postcalibration. In addition, the first example of feedback control to maintain constant methanol concentrations, as low as 1 g/L, is presented. Postcalibration measurements over a 9‐month period illustrate a level of reliability and robustness that promises its adoption for online bioprocess monitoring throughout product development, from early laboratory research and development to pilot and manufacturing scale operation. © 2014 American Institute of Chemical Engineers Biotechnol. Prog., 30:749–759, 2014     

Estimation of biomass concentrations in fermentation processes for recombinant protein production

Online biomass estimation for bioprocess supervision and control purposes is addressed. As the biomass concentration cannot be measured online during the production to sufficient accuracy, indirect measurement techniques are required. Here we compare several possibilities for the concrete case of recombinant protein production with genetically modified Escherichia coli bacteria and perform a ranking. At normal process operation, the best estimates can be obtained with artificial neural networks (ANNs). When they cannot be employed, statistical correlation techniques can be used such as multivariate regression techniques. Simple model-based techniques, e.g., those based on the Luedeking/Piret-type are not as accurate as the ANN approach; however, they are very robust. Techniques based on principal component analysis can be used to recognize abnormal cultivation behavior. For the cases investigated, a complete ranking list of the methods is given in terms of the root-mean-square error of the estimates. All techniques examined are in line with the recommendations expressed in the process analytical technology (PAT)-initiative of the FDA.     

Two‐dimensional fluorescence as soft sensor in the monitoring of biotransformation performed by yeast

Soft sensors are powerful tools for bioprocess monitoring due to their ability to perform online, noninvasive measurement, and possibility of detection of multiple components in cultivation media, which in turn can provide tools for the quantification of more than one metabolite/substrate/product in real time. In this work, soft sensor based on excitation‐emission fluorescence is for the first time applied for the monitoring of biotransformation production of 2‐phenylethanol (2‐PE) by yeast strains. Main process parameters—such as optical density, glucose, and 2‐PE concentrations—were determined with high accuracy and precision by fluorescence fingerprinting coupled with partial least squares regression. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 33:299–307, 2017     

The potential of random forest and neural networks for biomass and recombinant protein modeling in Escherichia coli fed‐batch fermentations

Product quality assurance strategies in production of biopharmaceuticals currently undergo a transformation from empirical “quality by testing” to rational, knowledge‐based “quality by design” approaches. The major challenges in this context are the fragmentary understanding of bioprocesses and the severely limited real‐time access to process variables related to product quality and quantity. Data driven modeling of process variables in combination with model predictive process control concepts represent a potential solution to these problems. The selection of statistical techniques best qualified for bioprocess data analysis and modeling is a key criterion. In this work a series of recombinant Escherichia coli fed‐batch production processes with varying cultivation conditions employing a comprehensive on‐ and offline process monitoring platform was conducted. The applicability of two machine learning methods, random forest and neural networks, for the prediction of cell dry mass and recombinant protein based on online available process parameters and two‐dimensional multi‐wavelength fluorescence spectroscopy is investigated. Models solely based on routinely measured process variables give a satisfying prediction accuracy of about ± 4% for the cell dry mass, while additional spectroscopic information allows for an estimation of the protein concentration within ± 12%. The results clearly argue for a combined approach: neural networks as modeling technique and random forest as variable selection tool.     

Monitoring of fed-batch <Emphasis Type="Italic">E. coli</Emphasis> fermentations with software sensors

Accurate monitoring and control of industrial bioprocess requires the knowledge of a great number of variables, being some of them not measurable with standard devices. To overcome this difficulty, software sensors can be used for on-line estimation of those variables and, therefore, its development is of paramount importance. An Asymptotic Observer was used for monitoring Escherichia coli fed-batch fermentations. Its performance was evaluated using simulated and experimental data. The results obtained showed that the observer was able to predict the biomass concentration profiles showing, however, less satisfactory results regarding the estimation of glucose and acetate concentrations. In comparison with the results obtained with an Extended Kalman Observer, the performance of the Asymptotic Observer in the fermentation monitoring was slightly better.     

Real-time estimation of biomass and specific growth rate in physiologically variable recombinant fed-batch processes

The real-time measurement of biomass has been addressed since many years. The quantification of biomass in the induction phase of a recombinant bioprocess is not straight forward, since biological burden, caused by protein expression, can have a significant impact on the cell morphology and physiology. This variability potentially leads to poor generalization of the biomass estimation, hence is a very important issue in the dynamic field of process development with frequently changing processes and producer lines. We want to present a method to quantify “biomass” in real-time which avoids off-line sampling and the need for representative training data sets. This generally applicable soft-sensor, based on first principles, was used for the quantification of biomass in induced recombinant fed-batch processes. Results were compared with “state of the art” methods to estimate the biomass concentration and the specific growth rate µ. Gross errors such as wrong stoichiometric assumptions or sensor failure were detected automatically. This method allows for variable model coefficients such as yields in contrast to other process models, hence does not require prior experiments. It can be easily adapted to a different growth stoichiometry; hence the method provides good generalization, also for induced culture mode. This approach estimates the biomass (or anabolic bioconversion) in induced fed-batch cultures in real-time and provides this key variable for process development for control purposes.     

Estimating animal population density using passive acoustics

Reliable estimation of the size or density of wild animal populations is very important for effective wildlife management, conservation and ecology. Currently, the most widely used methods for obtaining such estimates involve either sighting animals from transect lines or some form of capture‐recapture on marked or uniquely identifiable individuals. However, many species are difficult to sight, and cannot be easily marked or recaptured. Some of these species produce readily identifiable sounds, providing an opportunity to use passive acoustic data to estimate animal density. In addition, even for species for which other visually based methods are feasible, passive acoustic methods offer the potential for greater detection ranges in some environments (e.g. underwater or in dense forest), and hence potentially better precision. Automated data collection means that surveys can take place at times and in places where it would be too expensive or dangerous to send human observers. Here, we present an overview of animal density estimation using passive acoustic data, a relatively new and fast‐developing field. We review the types of data and methodological approaches currently available to researchers and we provide a framework for acoustics‐based density estimation, illustrated with examples from real‐world case studies. We mention moving sensor platforms (e.g. towed acoustics), but then focus on methods involving sensors at fixed locations, particularly hydrophones to survey marine mammals, as acoustic‐based density estimation research to date has been concentrated in this area. Primary among these are methods based on distance sampling and spatially explicit capture‐recapture. The methods are also applicable to other aquatic and terrestrial sound‐producing taxa. We conclude that, despite being in its infancy, density estimation based on passive acoustic data likely will become an important method for surveying a number of diverse taxa, such as sea mammals, fish, birds, amphibians, and insects, especially in situations where inferences are required over long periods of time. There is considerable work ahead, with several potentially fruitful research areas, including the development of (i) hardware and software for data acquisition, (ii) efficient, calibrated, automated detection and classification systems, and (iii) statistical approaches optimized for this application. Further, survey design will need to be developed, and research is needed on the acoustic behaviour of target species. Fundamental research on vocalization rates and group sizes, and the relation between these and other factors such as season or behaviour state, is critical. Evaluation of the methods under known density scenarios will be important for empirically validating the approaches presented here.     

Optimizing the biosynthesis of oxygenated and acetylated Taxol precursors in Saccharomyces cerevisiae using advanced bioprocessing strategies

Taxadien‐5α‐hydroxylase and taxadien‐5α‐ol O‐acetyltransferase catalyze the oxidation of taxadiene to taxadien‐5α‐ol and subsequent acetylation to taxadien‐5α‐yl‐acetate in the biosynthesis of the blockbuster anticancer drug, paclitaxel (Taxol®). Despite decades of research, the promiscuous and multispecific CYP725A4 enzyme remains a major bottleneck in microbial biosynthetic pathway development. In this study, an interdisciplinary approach was applied for the construction and optimization of the early pathway in Saccharomyces cerevisiae, across a range of bioreactor scales. High‐throughput microscale optimization enhanced total oxygenated taxane titer to 39.0 ± 5.7 mg/L and total taxane product titers were comparable at micro and minibioreactor scale at 95.4 ± 18.0 and 98.9 mg/L, respectively. The introduction of pH control successfully mitigated a reduction of oxygenated taxane production, enhancing the potential taxadien‐5α‐ol isomer titer to 19.2 mg/L, comparable with the 23.8 ± 3.7 mg/L achieved at microscale. A combination of bioprocess optimization and increased gas chromatography‐mass spectrometry resolution at 1 L bioreactor scale facilitated taxadien‐5α‐yl‐acetate detection with a final titer of 3.7 mg/L. Total oxygenated taxane titers were improved 2.7‐fold at this scale to 78 mg/L, the highest reported titer in yeast. Critical parameters affecting the productivity of the engineered strain were identified across a range of scales, providing a foundation for the development of robust integrated bioprocess control systems.     

On-line biomass estimation in biosurfactant production process by <Emphasis Type="Italic">Candida lipolytica</Emphasis> UCP 988

Biomass is an important variable in biosurfactant production process. However, such bioprocess variable, usually, is collected by sampling and determined by off-line analysis, with significant time delay. Therefore, simple and reliable on-line biomass estimation procedures are highly desirable. An artificial neural network model (ANN) is presented for the on-line estimation of biomass concentration, in biosurfactant production by Candida lipolytica UCP 988, as a nonlinear function of pH and dissolved oxygen. Several configurations were evaluated while developing the optimal ANN model. The optimal ANN model consists of one hidden layer with four neurons. The performance of the ANN was checked using experimental data. The results obtained indicate a very good predictive capacity for the ANN-based software sensor with values of R2 of 0.969 and RMSE of 0.021 for biomass concentration. Estimated biomass using the ANN was proved to be a simple, robust and accurate method.     

Soft sensor based on 2D‐fluorescence and process data enabling real‐time estimation of biomass in Escherichia coli cultivations

In bioprocesses, specific process responses such as the biomass cannot typically be measured directly on‐line, since analytical sampling is associated with unavoidable time delays. Accessing those responses in real‐time is essential for Quality by Design and process analytical technology concepts. Soft sensors overcome these limitations by indirectly measuring the variables of interest using a previously derived model and actual process data in real time. In this study, a biomass soft sensor based on 2D‐fluorescence data and process data, was developed for a comprehensive study with a 20‐L experimental design, for Escherichia coli fed‐batch cultivations. A multivariate adaptive regression splines algorithm was applied to 2D‐fluorescence spectra and process data, to estimate the biomass concentration at any time during the process. Prediction errors of 4.9% (0.99 g/L) for validation and 3.8% (0.69 g/L) for new data (external validation), were obtained. Using principal component and parallel factor analyses on the 2D‐fluorescence data, two potential chemical compounds were identified and directly linked to cell metabolism. The same wavelength pairs were also important predictors for the regression‐model performance. Overall, the proposed soft sensor is a valuable tool for monitoring the process performance on‐line, enabling Quality by Design.     

Universal label‐free in‐process quantification of influenza virus‐like particles

Virus‐like particles (VLPs) are becoming established as vaccines, in particular for influenza pandemics, increasing the interest in the development of VLPs manufacturing bioprocess. However, for complex VLPs, the analytical tools used for quantification are not yet able to keep up with the bioprocess progress. Currently, quantification for Influenza relies on traditional methods: hemagglutination assay or Single Radial Immunodiffusion. These analytical technologies are time‐consuming, cumbersome, and not supportive of efficient downstream process development and monitoring. Hereby we report a label‐free tool that uses Biolayer interferometry (BLI) technology applied on an Octet platform to quantify Influenza VLPs at all stages of bioprocess. Human (α2,6‐linked sialic acid) and avian (α2,3‐linked sialic acid) biotinylated receptors associated with streptavidin biosensors were used, to quantify hemagglutinin content in several mono‐ and multivalent Influenza VLPs. The applied method was able to quantify hemagglutinin from crude samples up to final bioprocessing VLP product. BLI technology confirmed its value as a high throughput analytical tool with high sensitivity and improved detection limits compared to traditional methods. This simple and fast method allowed for real‐time results, which are crucial for in‐line monitoring of downstream processing, improving process development, control and optimization.     

Early‐stage sustainability assessment of biotechnological processes: A case study of citric acid production

Sustainability assessment using a life‐cycle approach is indispensable to contemporary bioprocess development. This assessment is particularly important for early‐stage bioprocess development. As early‐stage investigations of bioprocesses involve the evaluation of their ecological and socioeconomic effects, they can be adjusted more effectively and improved towards sustainability, thereby reducing environmental risk and production costs. Early‐stage sustainability assessment is an important precautionary practice and, despite limited data, a unique opportunity to determine the primary impacts of bioprocess development. To this end, a simple and robust method was applied based on the standardized life‐cycle sustainability assessment methodology and commercially available datasets. In our study, we elaborated on the yeast‐based citric acid production process with Yarrowia lipolytica assessing 11 different substrates in different process modes. The focus of our analysis comprised both cultivation and down‐stream processing. According to our results, the repeated batch raw glycerol based bioprocess alternative showed the best environmental performance. The second‐ and third‐best options were also glycerol‐based. The least sustainable processes were those using molasses, chemically produced ethanol, and soy bean oil. The aggregated results of environmental, economic, and social impacts display waste frying oil as the best‐ranked alternative. The bioprocess with sunflower oil in the batch mode ranked second. The least favorable alternatives were the chemically produced ethanol‐, soy oil‐, refined glycerol‐, and molasses‐based citric acid production processes. The scenario analysis demonstrated that the environmental impact of nutrients and wastewater treatment is negligible, but energy demand of cultivation and down‐stream processing dominated the production process. However, without energy demand the omission of neutralizers almost halves the total impact, and neglecting pasteurization also considerably decreases the environmental impact.     

Optimal Combinations of Diagnostic Tests Based on AUC

Summary When several diagnostic tests are available, one can combine them to achieve better diagnostic accuracy. This article considers the optimal linear combination that maximizes the area under the receiver operating characteristic curve (AUC); the estimates of the combination's coefficients can be obtained via a nonparametric procedure. However, for estimating the AUC associated with the estimated coefficients, the apparent estimation by re‐substitution is too optimistic. To adjust for the upward bias, several methods are proposed. Among them the cross‐validation approach is especially advocated, and an approximated cross‐validation is developed to reduce the computational cost. Furthermore, these proposed methods can be applied for variable selection to select important diagnostic tests. The proposed methods are examined through simulation studies and applications to three real examples.     

Use of uniform designs in combination with neural networks for viral infection process development

This work aimed to compare the predictive capacity of empirical models, based on the uniform design utilization combined to artificial neural networks with respect to classical factorial designs in bioprocess, using as example the rabies virus replication in BHK‐21 cells. The viral infection process parameters under study were temperature (34°C, 37°C), multiplicity of infection (0.04, 0.07, 0.1), times of infection, and harvest (24, 48, 72 hours) and the monitored output parameter was viral production. A multilevel factorial experimental design was performed for the study of this system. Fractions of this experimental approach (18, 24, 30, 36 and 42 runs), defined according uniform designs, were used as alternative for modelling through artificial neural network and thereafter an output variable optimization was carried out by means of genetic algorithm methodology. Model prediction capacities for all uniform design approaches under study were better than that found for classical factorial design approach. It was demonstrated that uniform design in combination with artificial neural network could be an efficient experimental approach for modelling complex bioprocess like viral production. For the present study case, 67% of experimental resources were saved when compared to a classical factorial design approach. In the near future, this strategy could replace the established factorial designs used in the bioprocess development activities performed within biopharmaceutical organizations because of the improvements gained in the economics of experimentation that do not sacrifice the quality of decisions. © 2015 American Institute of Chemical Engineers Biotechnol. Prog., 31:532–540, 2015     

Optimization of norovirus virus‐like particle production in Pichia pastoris using a real‐time near‐infrared bioprocess monitor

The production of norovirus virus‐like particles (NoV VLPs) displaying NY‐ESO‐1 cancer testis antigen in Pichia pastoris BG11 Mut⁺ has been enhanced through feed‐strategy optimization using a near‐infrared bioprocess monitor (RTBio^® Bioprocess Monitor, ASL Analytical, Inc.), capable of monitoring and controlling the concentrations of glycerol and methanol in real‐time. The production of NoV VLPs displaying NY‐ESO‐1 in P. pastoris has potential as a novel cancer vaccine platform. Optimization of the growth conditions resulted in an almost two‐fold increase in the expression levels in the fermentation supernatant of P. pastoris as compared to the starting conditions. We investigated the effect of methanol concentration, batch phase time, and batch to induction transition on NoV VLP‐NY‐ESO‐1 production. The optimized process included a glycerol transition phase during the first 2 h of induction and a methanol concentration set point of 4 g L^?1 during induction. Utilizing the bioprocess monitor to control the glycerol and methanol concentrations during induction resulted in a maximum NoV VP1‐NY‐ESO‐1 yield of 0.85 g L^?1. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 32:518–526, 2016