A machine-learning approach to calibrate generic Raman models for real-time monitoring of cell culture processes

The manufacture of biotherapeutic proteins consists of complex upstream unit operations requiring multiple raw materials, analytical techniques, and control strategies to produce safe and consistent products for patients. Raman spectroscopy is a ubiquitous multipurpose analytical technique in biopharmaceutical manufacturing for real-time predictions of critical parameters in cell culture processes. The accuracy of Raman spectroscopy relies on chemometric models that need to be carefully calibrated. The existing calibration procedure is nontrivial to implement as it necessitates executing multiple carefully designed experiments for generating relevant calibration sets. Further, existing procedure yields calibration models that are reliable only in operating conditions they were calibrated in. This creates a unique challenge in clinical manufacturing where products have limited production history. In this paper, a novel machine-learning procedure based on just-in-time learning (JITL) is proposed to calibrate Raman models. Unlike traditional techniques, JITL-based generic Raman models can be reliably used for different modalities, cell lines, culture media, and operating conditions. The accuracy of JITL-based generic models is demonstrated on several validation studies involving real-time predictions of critical cell culture performance parameters, such as glucose, glutamate, glutamine, ammonium, lactate, sodium, calcium, viability, and viable cell density. The proposed JITL framework introduces a paradigm shift in the way industrial Raman models are calibrated, which to the best of authors’ knowledge have not been done before.     

Bioprocess in-line monitoring using Raman spectroscopy and Indirect Hard Modeling (IHM): A simple calibration yields a robust model

To increase the process productivity and product quality of bioprocesses, the in-line monitoring of critical process parameters is highly important. For monitoring substrate, metabolite, and product concentrations, Raman spectroscopy is a commonly used Process Analytical Technology (PAT) tool that can be applied in-situ and non-invasively. However, evaluating bioprocess Raman spectra with a robust state-of-the-art statistical model requires effortful model calibration. In the present study, we in-line monitored a glucose to ethanol fermentation by Saccharomyces cerevisiae (S. cerevisiae) using Raman spectroscopy in combination with the physics-based Indirect Hard Modeling (IHM) and showed successfully that IHM is an alternative to statistical models with significantly lower calibration effort. The IHM prediction model was developed and calibrated with only 16 Raman spectra in total, which did not include any process spectra. Nevertheless, IHM's root mean square errors of prediction (RMSEPs) for glucose (3.68 g/L) and ethanol (1.69 g/L) were comparable to the prediction quality of similar studies that used statistical models calibrated with several calibration batches. Despite our simple calibration, we succeeded in developing a robust model for evaluating bioprocess Raman spectra.     

In‐line monitoring of amino acids in mammalian cell cultures using raman spectroscopy and multivariate chemometrics models

The application of PAT for in‐line monitoring of biopharmaceutical manufacturing operations has a central role in developing more robust and consistent processes. Various spectroscopic techniques have been applied for collecting real‐time data from cell culture processes. Among these, Raman spectroscopy has been shown to have advantages over other spectroscopic techniques, especially in aqueous culture solutions. Measurements of several process parameters such as glucose, lactate, glutamine, glutamate, ammonium, osmolality and VCD using Raman‐based chemometrics models have been reported in literature. The application of Raman spectroscopy, coupled with calibration models for amino acid measurement in cell cultures, has been assessed. The developed models cover four amino acids important for cell growth and production: tyrosine, tryptophan, phenylalanine and methionine. The chemometrics models based on Raman spectroscopy data demonstrate the significant potential for the quantification of tyrosine, tryptophan and phenylalanine. The model for methionine would have to be further refined to improve quantification.     

Automated calibration and in-line measurement of product quality during therapeutic monoclonal antibody purification using Raman spectroscopy

Current manufacturing and development processes for therapeutic monoclonal antibodies demand increasing volumes of analytical testing for both real-time process controls and high-throughput process development. The feasibility of using Raman spectroscopy as an in-line product quality measuring tool has been recently demonstrated and promises to relieve this analytical bottleneck. Here, we resolve time-consuming calibration process that requires fractionation and preparative experiments covering variations of product quality attributes (PQAs) by engineering an automation system capable of collecting Raman spectra on the order of hundreds of calibration points from two to three stock seed solutions differing in protein concentration and aggregate level using controlled mixing. We used this automated system to calibrate multi-PQA models that accurately measured product concentration and aggregation every 9.3 s using an in-line flow-cell. We demonstrate the application of a nonlinear calibration model for monitoring product quality in real-time during a biopharmaceutical purification process intended for clinical and commercial manufacturing. These results demonstrate potential feasibility to implement quality monitoring during GGMP manufacturing as well as to increase chemistry, manufacturing, and controls understanding during process development, ultimately leading to more robust and controlled manufacturing processes.     

Nondestructive near-infrared spectroscopic measurement of multiple analytes in undiluted samples of serum-based cell culture media.

An adaptive calibration procedure is used to build selective multivariate calibration models for the measurement of glucose, lactate, glutamine, and ammonia in undiluted serum-based cell culture media. This adaptive procedure removes metabolism-induced covariance between these analytes in a series of calibration samples collected during the cultivation of PC-3 human prostate cancer cells. Partial least-squares calibration models are generated from single-beam near-infrared (NIR) spectra collected over the 4800- to 4200-cm(-1) combination spectral range. Calibration models were generated with both the full spectral range and optimized spectral ranges. In both cases, the number of model factors was optimized and model validity was determined by comparing analyte concentrations predicted from a series of independent and unaltered samples that were obtained during a subsequent cultivation of the PC-3 cells. Similar analytical performance was achieved with fewer model factors when the optimized spectral range was used. The lowest standard errors of prediction were 0.82, 0.94, 0.55, and 0.76 mM for glucose, lactate, glutamine, and ammonia, respectively. Different spectral ranges were optimal for each analyte and the optimized spectral range coincided with the distinguishing spectral features of the analyte. The results of this study demonstrate that NIR spectroscopy can be used effectively in the off-line measurement of important nutrients (glucose and glutamine) and byproducts (lactate and ammonia) in a serum-based animal cell culture medium.     

Real‐time monitoring of high‐gravity corn mash fermentation using in situ raman spectroscopy

In situ Raman spectroscopy was employed for real‐time monitoring of simultaneous saccharification and fermentation (SSF) of corn mash by an industrial strain of Saccharomyces cerevisiae. An accurate univariate calibration model for ethanol was developed based on the very strong 883 cm^?1 C–C stretching band. Multivariate partial least squares (PLS) calibration models for total starch, dextrins, maltotriose, maltose, glucose, and ethanol were developed using data from eight batch fermentations and validated using predictions for a separate batch. The starch, ethanol, and dextrins models showed significant prediction improvement when the calibration data were divided into separate high‐ and low‐concentration sets. Collinearity between the ethanol and starch models was avoided by excluding regions containing strong ethanol peaks from the starch model and, conversely, excluding regions containing strong saccharide peaks from the ethanol model. The two‐set calibration models for starch (R² = 0.998, percent error = 2.5%) and ethanol (R² = 0.999, percent error = 2.1%) provide more accurate predictions than any previously published spectroscopic models. Glucose, maltose, and maltotriose are modeled to accuracy comparable to previous work on less complex fermentation processes. Our results demonstrate that Raman spectroscopy is capable of real time in situ monitoring of a complex industrial biomass fermentation. To our knowledge, this is the first PLS‐based chemometric modeling of corn mash fermentation under typical industrial conditions, and the first Raman‐based monitoring of a fermentation process with glucose, oligosaccharides and polysaccharides present. Biotechnol. Bioeng. 2013; 110: 1654–1662. © 2013 Wiley Periodicals, Inc.     

Monitoring a bioprocess for ethanol production using FT-MIR and FT-Raman spectroscopy

The application of Fourier transform mid-infrared (FT-MIR) spectroscopy and Fourier transform Raman (FT-Raman) spectroscopy for process and quality control of fermentative production of ethanol was investigated. FT-MIR and FT-Raman spectroscopy along with multivariate techniques were used to determine simultaneously glucose, ethanol, and optical cell density of Saccharomyces cerevisiae during ethanol fermentation. Spectroscopic measurement of glucose and ethanol were compared and validated with the high-performance liquid chromatography (HPLC) method. Spectral wave number regions were selected for partial least-squares (PLS) regression and principal component regression (PCR) and calibration models for glucose, ethanol, and optical cell density were developed for culture samples. Correlation coefficient (R ²) value for the prediction for glucose and ethanol was more than 0.9 using various calibration methods. The standard error of prediction for the PLS first-derivative calibration models for glucose, ethanol, and optical cell density were 1.938 g/l, 1.150 g/l, and 0.507, respectively. Prediction errors were high with FT-Raman because the Raman scattering of the cultures was weak. Results indicated that FT-MIR spectroscopy could be used for rapid detection of glucose, ethanol, and optical cell density in S. cerevisiae culture during ethanol fermentation. Journal of Industrial Microbiology & Biotechnology (2001) 26, 185–190. Received 16 November 2000/ Accepted in revised form 12 January 2001     

Analysis of chemometric models applied to Raman spectroscopy for monitoring key metabolites of cell culture

The Food and Drug Administration (FDA) initiative of Process Analytical Technology (PAT) encourages the monitoring of biopharmaceutical manufacturing processes by innovative solutions. Raman spectroscopy and the chemometric modeling tool partial least squares (PLS) have been applied to this aim for monitoring cell culture process variables. This study compares the chemometric modeling methods of Support Vector Machine radial (SVMr), Random Forests (RF), and Cubist to the commonly used linear PLS model for predicting cell culture components—glucose, lactate, and ammonia. This research is performed to assess whether the use of PLS as standard practice is justified for chemometric modeling of Raman spectroscopy and cell culture data. Model development data from five small-scale bioreactors (2 × 1 L and 3 × 5 L) using two Chinese hamster ovary (CHO) cell lines were used to predict against a manufacturing scale bioreactor (2,000 L). Analysis demonstrated that Cubist predictive models were better for average performance over PLS, SVMr, and RF for glucose, lactate, and ammonia. The root mean square error of prediction (RMSEP) of Cubist modeling was acceptable for the process concentration ranges of glucose (1.437 mM), lactate (2.0 mM), and ammonia (0.819 mM). Interpretation of variable importance (VI) results theorizes the potential advantages of Cubist modeling in avoiding interference of Raman spectral peaks. Predictors/Raman wavenumbers (cm⁻¹) of interest for individual variables are X1139–X1141 for glucose, X846–X849 for lactate, and X2941–X2943 for ammonia. These results demonstrate that other beneficial chemometric models are available for use in monitoring cell culture with Raman spectroscopy.     

Generic Raman‐based calibration models enabling real‐time monitoring of cell culture bioreactors

Raman‐based multivariate calibration models have been developed for real‐time in situ monitoring of multiple process parameters within cell culture bioreactors. Developed models are generic, in the sense that they are applicable to various products, media, and cell lines based on Chinese Hamster Ovarian (CHO) host cells, and are scalable to large pilot and manufacturing scales. Several batches using different CHO‐based cell lines and corresponding proprietary media and process conditions have been used to generate calibration datasets, and models have been validated using independent datasets from separate batch runs. All models have been validated to be generic and capable of predicting process parameters with acceptable accuracy. The developed models allow monitoring multiple key bioprocess metabolic variables, and hence can be utilized as an important enabling tool for Quality by Design approaches which are strongly supported by the U.S. Food and Drug Administration. © 2015 American Institute of Chemical Engineers Biotechnol. Prog., 31:1004–1013, 2015     

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。     

Raman spectroscopy as a method to replace off-line pH during mammalian cell culture processes

Raman spectroscopy is a robust, well-established tool utilized for measuring important cell culture process variables for example, feed, metabolites, and biomass in real-time. This study further expands the functionality of in-line Raman spectroscopy coupled with partial least squares (PLS) regression modelling to develop a pH measurement tool. Cell line specific models were developed to enhance the robustness for processes with different pH setpoints, deadbands, and cellular metabolism. The modelling strategy further improved robustness by reducing the temporal complexity of pH shifts by splitting data sets into two time zones reflective of major changes in pH. In addition, models were developed to assess if lactate and partial pressure of carbon dioxide (pCO₂) could be used in a PLS model for pH. Splitting the data sets into early and late for the process resulted in errors of 0.035 pH and 0.034 pH for the two respective Raman cell lines models which was within acceptance criteria. The lactate and pCO₂ PLS model with values provided by Raman models had a further 0.001 pH error reduction. This study illustrates the potential to eliminate off-line samples to correct for in-line measurements of pH and further illustrates the capabilities of Raman to measure additional process variables.     

Enabling PAT in insect cell bioprocesses: In situ monitoring of recombinant adeno-associated virus production by fluorescence spectroscopy

The process analytical technology (PAT) initiative shifted the bioprocess development mindset towards real-time monitoring and control tools to measure relevant process variables online, and acting accordingly when undesirable deviations occur. Online monitoring is especially important in lytic production systems in which released proteases and changes in cell physiology are likely to affect product quality attributes, as is the case of the insect cell-baculovirus expression vector system (IC-BEVS), a well-established system for production of viral vectors and vaccines. Here, we applied fluorescence spectroscopy as a real-time monitoring tool for recombinant adeno-associated virus (rAAV) production in the IC-BEVS. Fluorescence spectroscopy is simple, yet sensitive and informative. To overcome the strong fluorescence background of the culture medium and improve predictive ability, we combined artificial neural network models with a genetic algorithm-based approach to optimize spectra preprocessing. We obtained predictive models for rAAV titer, cell viability and cell concentration with normalized root mean squared errors of 7%, 4%, and 7%, respectively, for leave-one-batch-out cross-validation. Our approach shows fluorescence spectroscopy allows real-time determination of the best time of harvest to maintain rAAV infectivity, an important quality attribute, and detection of deviations from the golden batch profile. This methodology can be applied to other biopharmaceuticals produced in the IC-BEVS, supporting the use of fluorescence spectroscopy as a versatile PAT tool.     

Process control in cell culture technology using dielectric spectroscopy

In the biopharmaceutical industry, mammalian and insect cells as well as plant cell cultures are gaining worldwide importance to produce biopharmaceuticals and as products themselves, for example in stem cell therapy. These highly sophisticated cell-based production processes need to be monitored and controlled to guarantee product quality and to satisfy GMP requirements. With the process analytical technology (PAT) initiative, requirements regarding process monitoring and control have changed and real-time in-line monitoring tools are now recommended. Dielectric spectroscopy (DS) can serve as a tool to satisfy some PAT requirements. DS has been used in the medical field for quite some time and it may allow real-time process monitoring of biological cell culture parameters. DS has the potential to enable process optimization, automation, cost reduction, and a more consistent product quality. Dielectric spectroscopy is reviewed here as a tool to monitor biochemical processes. Commercially available dielectric sensing systems are discussed. The potential of this technology is demonstrated through examples of current and potential future applications in research and industry for mammalian and insect cell culture.     

Improving predictive mapping in Swiss mire ecosystems through re‐calibration of indicator values

Question: How may Landolt indicator values be re‐calibrated to improve the performance of predictive models? Location: Mires Gross Moos Schwändital (1250 m a.s.l.) in the Prealps, Burgmoos (465 m. a.s.l.) on the Central Plateau and La Burtignière (1000 m a.s.l.) in the Jura, Switzerland. Methods: Habitat distribution models based on high resolution remotely sensed data and vegetation field data are applied to monitor 130 mires. Instead of plant species or communities we used mean indicator values of vegetation records as response variables. To improve the differential power of indicator values for wetland habitat conditions, we calibrated these values using field data. Different methods were tested with our predictive models in three mires to see which calibration method is best in enhancing model performance. To assess the effect of the uneven distribution of vegetation records along environmental gradients, calibrations based on random and evenly distributed samples were compared. As a test of the predictive power of the models we used r² between ground truth and model prediction. This approach is illustrated through an application with nutrient indicator values in the mire La Burtignière. Results: Model performances were not the same for the three mires. The predictive power was better for the nutrient values, soil reaction and humus values than for light and moisture values. 2000 records were sufficient as basis for re‐calibration. Models based on original Landolt indicator values were overall the weakest compared with re‐calibrated values. By comparing the predictive power of Models based on randomly or evenly selected records were about equally predictive. Conclusions: 1. Ahabitat‐specific re‐calibration of the Landolt indicator values enhances the predictive mapping of the Swiss mire ecosystems. 2. The re‐calibration based on weighted averaging gives a better performance than the one based on Gaussian logistic regression. 3. The uneven distribution of indicator values due to the over‐representation of mire habitats does not hamper model performance. 4. 2000 vegetation records are a sufficient basis for an optimal re‐calibration of the vegetation types. An illustration of the method is given by using the soil fertility pattern of the mire La Burtignière.     

Straightforward method for calibration of mechanistic cation exchange chromatography models for industrial applications

Mechanistic modeling of chromatography processes is one of the most promising techniques for the digitalization of biopharmaceutical process development. Possible applications of chromatography models range from in silico process optimization in early phase development to in silico root cause investigation during manufacturing. Nonetheless, the cumbersome and complex model calibration still decelerates the implementation of mechanistic modeling in industry. Therefore, the industry demands model calibration strategies that ensure adequate model certainty in a limited amount of time. This study introduces a directed and straightforward approach for the calibration of pH-dependent, multicomponent steric mass action (SMA) isotherm models for industrial applications. In the case investigated, the method was applied to a monoclonal antibody (mAb) polishing step including four protein species. The developed strategy combined well-established theories of preparative chromatography (e.g. Yamamoto method) and allowed a systematic reduction of unknown model parameters to 7 from initially 32. Model uncertainty was reduced by designing two representative calibration experiments for the inverse estimation of remaining model parameters. Dedicated experiments with aggregate-enriched load material led to a significant reduction of model uncertainty for the estimates of this low-concentrated product-related impurity. The model was validated beyond the operating ranges of the final unit operation, enabling its application to late-stage downstream process development. With the proposed model calibration strategy, a systematic experimental design is provided, calibration effort is strongly reduced, and local minima are avoided.     

Real-time monitoring of adherent Vero cell density and apoptosis in bioreactor processes

This study proposes an easy to use in situ device, based on multi-frequency permittivity measurements, to monitor the growth and death of attached Vero cells cultivated on microporous microcarriers, without any cell sampling. Vero cell densities were on-line quantified up to 10⁶ cell mL⁻¹. Some parameters which could potentially impact Vero cell morphological and physiological states were assessed through different culture operating conditions, such as media formulation or medium feed-harvest during cell growth phase. A new method of in situ cell death detection with dielectric spectroscopy was also successfully implemented. Thus, through permittivity frequency scanning, major rises of the apoptotic cell population in bioreactor cultures were detected by monitoring the characteristic frequency of the cell population, f_c, which is one of the culture dielectric parameters. Both cell density quantification and cell apoptosis detection are strategic information in cell-based production processes as they are involved in major events of the process, such as scale-up or choice of the viral infection conditions. This new application of dielectric spectroscopy to adherent cell culture processes makes it a very promising tool for risk-mitigation strategy in industrial processes. Therefore, our results contribute to the development of Process Analytical Technology in cell-based industrial processes.     

Monitoring complex media fermentations with near-infrared spectroscopy: comparison of different variable selection methods

Near-infrared spectroscopy (NIRS) is known to be a suitable technique for rapid fermentation monitoring. Industrial fermentation media are complex, both chemically (ill-defined composition) and physically (multiphase sample matrix), which poses an additional challenge to the development of robust NIRS calibration models. We investigated the use of NIRS for at-line monitoring of the concentration of clavulanic acid during an industrial fermentation. An industrial strain of Streptomyces clavuligerus was cultivated at 200-L scale for the production of clavulanic acid. Partial least squares (PLS) regression was used to develop calibration models between spectral and analytical data. In this work, two different variable selection methods, genetic algorithms (GA) and PLS-bootstrap, were studied and compared with models built using all the spectral variables. Calibration models for clavulanic acid concentration performed well both on internal and external validation. The two variable selection methods improved the predictive ability of the models up to 20%, relative to the calibration model built using the whole spectra.     

Rapid characterization and quality control of complex cell culture media solutions using raman spectroscopy and chemometrics

The use of Raman spectroscopy coupled with chemometrics for the rapid identification, characterization, and quality assessment of complex cell culture media components used for industrial mammalian cell culture was investigated. Raman spectroscopy offers significant advantages for the analysis of complex, aqueous‐based materials used in biotechnology because there is no need for sample preparation and water is a weak Raman scatterer. We demonstrate the efficacy of the method for the routine analysis of dilute aqueous solution of five different chemically defined (CD) commercial media components used in a Chinese Hamster Ovary (CHO) cell manufacturing process for recombinant proteins.The chemometric processing of the Raman spectral data is the key factor in developing robust methods. Here, we discuss the optimum methods for eliminating baseline drift, background fluctuations, and other instrumentation artifacts to generate reproducible spectral data. Principal component analysis (PCA) and soft independent modeling of class analogy (SIMCA) were then employed in the development of a robust routine for both identification and quality evaluation of the five different media components. These methods have the potential to be extremely useful in an industrial context for “in‐house” sample handling, tracking, and quality control. Biotechnol. Bioeng. 2010;107: 290–301. © 2010 Wiley Periodicals, Inc.     

A mixture theory model of fluid and solute transport in the microvasculature of normal and malignant tissues. II: Factor sensitivity analysis, calibration, and validation

The treatment of cancerous tumors is dependent upon the delivery of therapeutics through the blood by means of the microcirculation. Differences in the vasculature of normal and malignant tissues have been recognized, but it is not fully understood how these differences affect transport and the applicability of existing mathematical models has been questioned at the microscale due to the complex rheology of blood and fluid exchange with the tissue. In addition to determining an appropriate set of governing equations it is necessary to specify appropriate model parameters based on physiological data. To this end, a two stage sensitivity analysis is described which makes it possible to determine the set of parameters most important to the model’s calibration. In the first stage, the fluid flow equations are examined and a sensitivity analysis is used to evaluate the importance of 11 different model parameters. Of these, only four substantially influence the intravascular axial flow providing a tractable set that could be calibrated using red blood cell velocity data from the literature. The second stage also utilizes a sensitivity analysis to evaluate the importance of 14 model parameters on extravascular flux. Of these, six exhibit high sensitivity and are integrated into the model calibration using a response surface methodology and experimental intra- and extravascular accumulation data from the literature (Dreher et al. in J Natl Cancer Inst 98(5):335–344, 2006). The model exhibits good agreement with the experimental results for both the mean extravascular concentration and the penetration depth as a function of time for inert dextran over a wide range of molecular weights.