Comparison of data science workflows for root cause analysis of bioprocesses.

Root cause analysis (RCA) is one of the most prominent tools used to comprehensively evaluate a biopharmaceutical production process. Despite of its widespread use in industry, the Food and Drug Administration has observed a lot of unsuitable approaches for RCAs within the last years. The reasons for those unsuitable approaches are the use of incorrect variables during the analysis and the lack in process understanding, which impede correct model interpretation. Two major approaches to perform RCAs are currently dominating the chemical and pharmaceutical industry: raw data analysis and feature-based approach. Both techniques are shown to be able to identify the significant variables causing the variance of the response. Although they are different in data unfolding, the same tools as principal component analysis and partial least square regression are used in both concepts. Within this article we demonstrate the strength and weaknesses of both approaches. We proved that a fusion of both results in a comprehensive and effective workflow, which not only increases better process understanding. We demonstrate this workflow along with an example. Hence, the presented workflow allows to save analysis time and to reduce the effort of data mining by easy detection of the most important variables within the given dataset. Subsequently, the final obtained process knowledge can be translated into new hypotheses, which can be tested experimentally and thereby lead to effectively improving process robustness.

Multivariate analytics of chromatographic data: Visual computing based on moving window factor models.

Chromatography is one of the most versatile unit operations in the biotechnological industry. Regulatory initiatives like Process Analytical Technology and Quality by Design led to the implementation of new chromatographic devices. Those represent an almost inexhaustible source of data. However, the analysis of large datasets is complicated, and significant amounts of information stay hidden in big data. Here we present a new, top-down approach for the systematic analysis of chromatographic datasets. It is the goal of this approach to analyze the dataset as a whole, starting with the most important, global information. The workflow should highlight interesting regions (outliers, drifts, data inconsistencies), and help to localize those regions within a multi-dimensional space in a straightforward way. Moving window factor models were used to extract the most important information, focusing on the differences between samples. The prototype was implemented as an interactive visualization tool for the explorative analysis of complex datasets. We found that the tool makes it convenient to localize variances in a multidimensional dataset and allows to differentiate between explainable and unexplainable variance. Starting with one global difference descriptor per sample, the analysis ends up with highly resolute temporally dependent difference descriptor values, thought as a starting point for the detailed analysis of the underlying raw data.

Integrated Process Modeling-A Process Validation Life Cycle Companion.

During the regulatory requested process validation of pharmaceutical manufacturing processes, companies aim to identify, control, and continuously monitor process variation and its impact on critical quality attributes (CQAs) of the final product. It is difficult to directly connect the impact of single process parameters (PPs) to final product CQAs, especially in biopharmaceutical process development and production, where multiple unit operations are stacked together and interact with each other. Therefore, we want to present the application of Monte Carlo (MC) simulation using an integrated process model (IPM) that enables estimation of process capability even in early stages of process validation. Once the IPM is established, its capability in risk and criticality assessment is furthermore demonstrated. IPMs can be used to enable holistic production control strategies that take interactions of process parameters of multiple unit operations into account. Moreover, IPMs can be trained with development data, refined with qualification runs, and maintained with routine manufacturing data which underlines the lifecycle concept. These applications will be shown by means of a process characterization study recently conducted at a world-leading contract manufacturing organization (CMO). The new IPM methodology therefore allows anticipation of out of specification (OOS) events, identify critical process parameters, and take risk-based decisions on counteractions that increase process robustness and decrease the likelihood of OOS events.

Workflow for Criticality Assessment Applied in Biopharmaceutical Process Validation Stage 1.

Identification of critical process parameters that impact product quality is a central task during regulatory requested process validation. Commonly, this is done via design of experiments and identification of parameters significantly impacting product quality (rejection of the null hypothesis that the effect equals 0). However, parameters which show a large uncertainty and might result in an undesirable product quality limit critical to the product, may be missed. This might occur during the evaluation of experiments since residual/un-modelled variance in the experiments is larger than expected a priori. Estimation of such a risk is the task of the presented novel retrospective power analysis permutation test. This is evaluated using a data set for two unit operations established during characterization of a biopharmaceutical process in industry. The results show that, for one unit operation, the observed variance in the experiments is much larger than expected a priori, resulting in low power levels for all non-significant parameters. Moreover, we present a workflow of how to mitigate the risk associated with overlooked parameter effects. This enables a statistically sound identification of critical process parameters. The developed workflow will substantially support industry in delivering constant product quality, reduce process variance and increase patient safety.

Automatic feed phase identification in multivariate bioprocess profiles by sequential binary classification.

In this paper, we propose a new strategy for retrospective identification of feed phases from online sensor-data enriched feed profiles of an Escherichia Coli (E. coli) fed-batch fermentation process. In contrast to conventional (static), data-driven multi-class machine learning (ML), we exploit process knowledge in order to constrain our classification system yielding more parsimonious models compared to static ML approaches. In particular, we enforce unidirectionality on a set of binary, multivariate classifiers trained to discriminate between adjacent feed phases by linking the classifiers through a one-way switch. The switch is activated when the actual classifier output changes. As a consequence, the next binary classifier in the classifier chain is used for the discrimination between the next feed phase pair etc. We allow activation of the switch only after a predefined number of consecutive predictions of a transition event in order to prevent premature activation of the switch and undertake a sensitivity analysis regarding the optimal choice of the (time) lag parameter. From a complexity/parsimony perspective the benefit of our approach is three-fold: i) The multi-class learning task is broken down into binary subproblems which usually have simpler decision surfaces and tend to be less susceptible to the class-imbalance problem. ii) We exploit the fact that the process follows a rigid feed cycle structure (i.e. batch-feed-batch-feed) which allows us to focus on the subproblems involving phase transitions as they occur during the process while discarding off-transition classifiers and iii) only one binary classifier is active at the time which keeps effective model complexity low. We further use a combination of logistic regression and Lasso (i.e. regularized logistic regression, RLR) as a wrapper to extract the most relevant features for individual subproblems from the whole set of high-dimensional sensor data. We train different soft computing classifiers, including decision trees (DT), k-nearest neighbors (k-NN), support vector machines (SVM) and an own developed fuzzy classifier and compare our method with conventional multi-class ML. Our results show a remarkable out-performance of the here proposed method over static ML approaches in terms of accuracy and robustness. We achieved close to error free feed phase classification while reducing the misclassification rates in 17 out of 20 investigated test cases in the range between 39% and 98.2% depending on feature set and classifier architecture. Models trained on features based on selection by RLR significantly outperformed those trained on features suggested by experts and their predictive performance was considerably less affected by the choice of the lag parameter.

A novel method to recover inclusion body protein from recombinant E. coli fed-batch processes based on phage ΦX174-derived lysis protein E.

Production of recombinant proteins as inclusion bodies is an important strategy in the production of technical enzymes and biopharmaceutical products. So far, protein from inclusion bodies has been recovered from the cell factory through mechanical or chemical disruption methods, requiring additional cost-intensive unit operations. We describe a novel method that is using a bacteriophage-derived lysis protein to directly recover inclusion body protein from Escherichia coli from high cell density fermentation process: The recombinant inclusion body product is expressed by using a mixed feed fed-batch process which allows expression tuning via adjusting the specific uptake rate of the inducing substrate. Then, bacteriophage ΦX174-derived lysis protein E is expressed to induce cell lysis. Inclusion bodies in empty cell envelopes are harvested via centrifugation of the fermentation broth. A subsequent solubilization step reveals the recombinant protein. The process was investigated by analyzing the impact of fermentation conditions on protein E-mediated cell lysis as well as cell lysis kinetics. Optimal cell lysis efficiencies of 99% were obtained with inclusion body titers of >2.0 g/l at specific growth rates higher 0.12 h-1 and inducer uptake rates below 0.125 g/(g × h). Protein E-mediated cell disruption showed a first-order kinetics with a kinetic constant of -0.8 ± 0.3 h-1. This alternative inclusion body protein isolation technique was compared to the one via high-pressure homogenization. SDS gel analysis showed 10% less protein impurities when cells had been disrupted via high-pressure homogenization, than when empty cell envelopes including inclusion bodies were investigated. Within this contribution, an innovative technology, tuning recombinant protein production and substituting cost-intensive mechanical cell disruption, is presented. We anticipate that the presented method will simplify and reduce the production costs of inclusion body processes to produce technical enzymes and biopharmaceutical products.

Propagation of measurement accuracy to biomass soft-sensor estimation and control quality.

In biopharmaceutical process development and manufacturing, the online measurement of biomass and derived specific turnover rates is a central task to physiologically monitor and control the process. However, hard-type sensors such as dielectric spectroscopy, broth fluorescence, or permittivity measurement harbor various disadvantages. Therefore, soft-sensors, which use measurements of the off-gas stream and substrate feed to reconcile turnover rates and provide an online estimate of the biomass formation, are smart alternatives. For the reconciliation procedure, mass and energy balances are used together with accuracy estimations of measured conversion rates, which were so far arbitrarily chosen and static over the entire process. In this contribution, we present a novel strategy within the soft-sensor framework (named adaptive soft-sensor) to propagate uncertainties from measurements to conversion rates and demonstrate the benefits: For industrially relevant conditions, hereby the error of the resulting estimated biomass formation rate and specific substrate consumption rate could be decreased by 43 and 64 %, respectively, compared to traditional soft-sensor approaches. Moreover, we present a generic workflow to determine the required raw signal accuracy to obtain predefined accuracies of soft-sensor estimations. Thereby, appropriate measurement devices and maintenance intervals can be selected. Furthermore, using this workflow, we demonstrate that the estimation accuracy of the soft-sensor can be additionally and substantially increased.

Bioprocess development workflow: Transferable physiological knowledge instead of technological correlations.

Microbial bioprocesses need to be designed to be transferable from lab scale to production scale as well as between setups. Although substantial effort is invested to control technological parameters, usually the only true constant parameter is the actual producer of the product: the cell. Hence, instead of solely controlling technological process parameters, the focus should be increasingly laid on physiological parameters. This contribution aims at illustrating a workflow of data life cycle management with special focus on physiology. Information processing condenses the data into physiological variables, while information mining condenses the variables further into physiological descriptors. This basis facilitates data analysis for a physiological explanation for observed phenomena in productivity. Targeting transferability, we demonstrate this workflow using an industrially relevant Escherichia coli process for recombinant protein production and substantiate the following three points: (1) The postinduction phase is independent in terms of productivity and physiology from the preinduction variables specific growth rate and biomass at induction. (2) The specific substrate uptake rate during induction phase was found to significantly impact the maximum specific product titer. (3) The time point of maximum specific titer can be predicted by an easy accessible physiological variable: while the maximum specific titers were reached at different time points (19.8 ± 7.6 h), those maxima were reached all within a very narrow window of cumulatively consumed substrate dSn (3.1 ± 0.3 g/g). Concluding, this contribution provides a workflow on how to gain a physiological view on the process and illustrates potential benefits. © 2016 American Institute of Chemical Engineers Biotechnol. Prog., 33:261-270, 2017.

Simple monitoring of cell leakiness and viability in Escherichia coli bioprocesses-A case study.

In a recently published study, we developed a simple methodology to monitor Escherichia coli cell integrity and lysis during bioreactor cultivations, where we intentionally triggered leakiness. In this follow-up study, we used this methodology, comprising the measurement of extracellular alkaline phosphatase to monitor leakiness and flow cytometry to follow viability, to investigate the effect of process parameters on a recombinant E. coli strain producing the highly valuable vascular endothelial growth factor A165 (VEGF-A165) in the periplasm. Since the amount of soluble product was very little (<500 µg/g dry cell weight), we directly linked the effect of the three process parameters temperature, specific uptake rate of the inducer arabinose and specific growth rate (µ) to cell integrity and viability. We found that a low temperature and a high µ were beneficial for cell integrity and that an elevated temperature resulted in reduced viability. We concluded that the recombinant E. coli cells producing VEGF-A165 in the periplasm should be cultivated at low temperature and high µ to reduce leakiness and guarantee high viability. Summarizing, in this follow-up study we demonstrate the usefulness of our simple methodology to monitor leakiness and viability of recombinant E. coli cells during bioreactor cultivations.

Tunable recombinant protein expression in E. coli: promoter systems and genetic constraints.

Tuning of transcription is a promising strategy to overcome challenges associated with a non-suitable expression rate like outgrowth of segregants, inclusion body formation, metabolic burden and inefficient translocation. By adjusting the expression rate-even on line-to purposeful levels higher product titres and more cost-efficient production processes can be achieved by enabling culture long-term stability and constant product quality. Some tunable systems are registered for patents or already commercially available. Within this contribution, we discuss the induction mechanisms of various Escherichia coli inherent promoter systems with respect to their tunability and review studies using these systems for expression tuning. According to the current level of knowledge, some promoter systems were successfully used for expression tuning, and in some cases, analytical evidence on single-cell level is still pending. However, only a few studies using tunable strains apply a suitable process control strategy. So far, expression tuning has only gathered little attention, but we anticipate that expression tuning harbours great potential for enabling and optimizing the production of a broad spectrum of products in E. coli.

Fed-Batch Production of Bacterial Ghosts Using Dielectric Spectroscopy for Dynamic Process Control.

The Bacterial Ghost (BG) platform technology evolved from a microbiological expression system incorporating the ϕX174 lysis gene E. E-lysis generates empty but structurally intact cell envelopes (BGs) from Gram-negative bacteria which have been suggested as candidate vaccines, immunotherapeutic agents or drug delivery vehicles. E-lysis is a highly dynamic and complex biological process that puts exceptional demands towards process understanding and control. The development of a both economic and robust fed-batch production process for BGs required a toolset capable of dealing with rapidly changing concentrations of viable biomass during the E-lysis phase. This challenge was addressed using a transfer function combining dielectric spectroscopy and soft-sensor based biomass estimation for monitoring the rapid decline of viable biomass during the E-lysis phase. The transfer function was implemented to a feed-controller, which followed the permittivity signal closely and was capable of maintaining a constant specific substrate uptake rate during lysis phase. With the described toolset, we were able to increase the yield of BG production processes by a factor of 8-10 when compared to currently used batch procedures reaching lysis efficiencies >98%. This provides elevated potentials for commercial application of the Bacterial Ghost platform technology.

Tunable recombinant protein expression in E. coli: enabler for continuous processing?

Tuning of transcription is a powerful process technological tool for efficient recombinant protein production in Escherichia coli. Many challenges such as product toxicity, formation of inclusion bodies, cell death, and metabolic burden are associated with non-suitable (too high or too low) levels of recombinant protein expression. Tunable expression systems allow adjusting the recombinant protein expression using process technological means. This enables to exploit the cell's metabolic capacities to a maximum. Within this article, we review genetic and process technological aspects of tunable expression systems in E. coli, providing a roadmap for the industrial exploitation of the reviewed technologies. We attempt to differentiate the term "expression tuning" from its inflationary use by providing a concise definition and highlight interesting fields of application for this versatile new technology. Dependent on the type of inducer (metabolizable or non-metabolizable), different process strategies are required in order to achieve tuning. To fully profit from the benefits of tunable systems, an independent control of growth rate and expression rate is indispensable. Being able to tackle problems such as long-term culture stability and constant product quality expression tuning is a promising enabler for continuous processing in biopharmaceutical production.

Tunable recombinant protein expression with E. coli in a mixed-feed environment.

Controlling the recombinant protein production rate in Escherichia coli is of utmost importance to ensure product quality and quantity. Up to now, only the genetic construct, introduced into E. coli, and the specific growth rate of the culture were used to influence and stir the productivity. However, bioprocess technological means to control or even tune the productivity of E. coli are scarce. Here, we present a novel method for the process-technological control over the recombinant protein expression rate in E. coli. A mixed-feed fed-batch bioprocess based on the araBAD promoter expression system using both D-glucose and L-arabinose as assimilable C-sources was designed. Using the model product green fluorescent protein, we show that the specific product formation rate can be efficiently tuned even on the cellular level only via the uptake rate of L-arabinose. This novel approach introduces an additional degree of freedom for the design of recombinant bioprocesses with E. coli. We anticipate that the presented method will result in significant quality and robustness improvement as well as cost and process time reduction for recombinant bacterial bioprocesses in the future.

An Integrated Downstream Process Development Strategy along QbD Principles.

The development, optimization, and analysis of downstream processes are challenged by a high number of potentially critical process parameters that need to be investigated using lab-scale experiments. These process parameters are spread across multiple unit operations and potentially show interactions across unit operations. In this contribution, we present a novel strategy for bioprocess development that considers the risk of parameter interactions across unit operations for efficient experimental design. A novel risk assessment tool (interaction matrix) is introduced to the Quality by Design (QbD) workflow. Using this tool, the risk of interaction across unit operations is rated. Subsequently, a design of experiments (DoE) across unit operations is conducted that has the power to reveal multivariate interdependencies. The power of the presented strategy is demonstrated for protein isolation steps of an inclusion body process, focusing on the quality attribute inclusion body purity. The concentration of Triton X-100 in the course of inclusion body (IB) purification was shown to interact with the g-number of the subsequent centrifugation step. The presented strategy targets a holistic view on the process and allows handling of a high number of experimental parameters across unit operations using minimal experimental effort. It is generically applicable for process development along QbD principles.

A dynamic method for the investigation of induced state metabolic capacities as a function of temperature.

BACKGROUND: Science-based recombinant bioprocess designs as well as the design of statistical experimental plans for process optimization (Design of Experiments, DoE) demand information on physiological bioprocess boundaries, such as the onset of acetate production, adaptation times, mixed feed metabolic capabilities or induced state maximum metabolic rates as at the desired cultivation temperature. Dynamic methods provide experimental alternatives to determine this information in a fast and efficient way. Information on maximum metabolic capabilities as a function of temperature is needed in case a reduced cultivation temperature is desirable (e.g. to avoid inclusion body formation) and an appropriate feeding profile is to be designed. RESULTS: Here, we present a novel dynamic method for the determination of the specific growth rate as a function of temperature for induced recombinant bacterial bioprocesses. The method is based on the control of the residual substrate concentration at non-limiting conditions with dynamic changes in cultivation temperature. The presented method was automated in respect to information extraction and closed loop control by means of in-line Fourier Transformation Infrared Spectroscopy (FTIR) residual substrate measurements and on-line first principle rate-based soft-sensors. Maximum induced state metabolic capabilities as a function of temperature were successfully extracted for a recombinant E. coli C41 fed-batch bioprocess without the need for sampling in a time frame of 20 hours. CONCLUSIONS: The presented method was concluded to allow the fast and automated extraction of maximum metabolic capabilities (specific growth rate) as a function of temperature. This complements the dynamic toolset necessary for science-based recombinant bacterial bioprocess design and DoE design.

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.

Model-based analysis on the extractability of information from data in dynamic fed-batch experiments.

Dynamic changes of physiological bioprocess parameters, e.g. a change in the specific growth rate µ, are frequently observed during industrial manufacturing as well as bioprocess development. A quantitative description of these variations is of great interest, since it can bring elucidation to the physiological state of the culture. The goal of this contribution was to show limitations and issues for the calculation of rates with regard to temporal resolution for dynamic fed-batch experiments. The impact of measurement errors, temporal resolution and the physiological activity on the signal to noise ratio (SNR) of the calculated rates was evaluated using an in-silico approach. To make use of that in practice, a generally applicable rule of thumb equation for the estimation of the SNR of specific rates was presented. The SNR calculated by this rule of thumb equation helps with definition of sampling intervals and making a decision whether an observed change is statistically significant or should be attributed to random error. Furthermore, a generic reconciliation approach to remove random as well as systematic error from data was presented. This reconciliation technique requires only little prior knowledge. The validity of the proposed tools was checked with real data from a fed-batch culture of E. coli with dynamic variations due to feed profile.

Information Processing: Rate-Based Investigation of Cell Physiological Changes along Design Space Development.

Recent initiatives summarized under the term quality by design (QbD) urge for science and risk-based pharmaceutical bioprocess development strategies. One of the most accepted concepts communicated by the regulatory authorities is the concept of design space-a multidimensional combination of critical process parameter (CPP) ranges where the quality acceptance criteria (critical quality attributes, CQAs) are fulfilled. Current design space development along QbD principles focuses on the investigation of statistical CPP/CQA interactions, while the biological mechanistic of this interaction is hardly considered. Furthermore, the plethora of available online and offline data gathered within design space development is typically not used for the demonstration of process understanding. Here we present a methodology about how typical recorded process data can be processed and used to gather mechanistic process knowledge within upstream design space development, without the need for further experiments or additional analytical procedures. Data derived from online and offline measurements (off gas quantification, air flows, substrate flows, biomass dry cell weight measurements) were processed into scale-independent information in the form of specific rates and yield coefficients (data processing). Subsequently, the obtained information was regressed with the investigated process parameters aiming at the investigation of mechanistic interactions (information processing). The power of the presented approach was demonstrated on a multivariate study involving two process parameters (induction phase temperature and induction phase feeding strategy) aiming at the production of recombinant product in an Escherichia coli K12 strain. The knowledge successfully extracted indicated a time dependency of the metabolic load posed on the system, a possible down regulation of the promoter at reduced temperatures, and reduced cell lysis at higher specific feeding regimes. The presented data and information processing methodology for mechanistic process knowledge extraction is fully complementary to the task of design space development for QbD submissions and can serve as the basis of mechanistic modeling. LAY ABSTRACT: Manufacturing of pharmaceuticals intended for human use is under tight control of government authorities. To further improve product quality and allow more manufacturing flexibility, government agencies started to encourage manufactures to investigate and understand their manufacturing processes scientifically. This should lead to quality by design (QbD), hence a manufacturing that is so well understood that final product quality can be guaranteed by the manufacturing process itself.

Efficient feeding profile optimization for recombinant protein production using physiological information.

A multivariate study was performed aiming at the optimization of a recombinant rhamnose inducible E. coli induction system with alkaline phosphatase as target product. The effects of typical factors with impact on post- as well as pre-induction feeding rates were investigated with respect to the space-time yield of the target product. The goal was increased understanding as well as quantitative characterization of these factors with respect to their physiological impact on the model system. The optical density (OD) at which the culture was induced had a strong positive effect on the space-time yield. Pre-induction growth rate (k) had a second-order effect, while induction feed rate drop (J), a factor defining the linear post-induction feed rate, was interacting with (k). However, explanation of the observed effects to acquire more understanding regarding their effect on cell metabolism was not straight forward. Hence, the original process parameters were transformed into physiological more meaningful parameters and served as the basis for a multivariate data analysis. The observed variance with respect to observed volumetric activity was fully explained by the specific substrate uptake rate (q (s)) and induction OD, merging the process parameters pre-induction growth rate (k) and feed rate drop (J) into the physiological parameter specific substrate uptake rate (q (s)). After transformation of the response volumetric activity (U/ml) into the biomass specific activity (U/g(biomass)), the observed variance was fully explained solely by the specific substrate uptake rate (q (s)). Due to physiological multivariate data analysis, the interpretation of the results was facilitated and factors were reduced. On the basis of the obtained results, it was concluded that the physiological parameter q (s) rather than process parameters (k, J, induction OD) should be used for process optimization with respect to the feeding profile.