Process Analytical Technology Beyond Real-Time Analyzers: The Role of Multivariate Analysis

Identification of Reaction Network Hypotheses for Complex Feedstocks from Spectroscopic Measurements with Minimal Human Intervention

In this work, we detail an automated reaction network hypothesis generation protocol for processes involving complex feedstocks where information about the species and reactions involved is unknown. Our methodology is process agnostic and can be utilized in any reactive process with spectroscopic measurements that provide information on the evolution of the components in the mixture. We decompose the mixture spectra to obtain spectroscopic signatures of the individual components and use a 1-D convolutional neural network to automatically identify functional groups indicated by them. We employ atom-atom mapping to automatically recover reaction rules that are applied on candidate molecules identified from chemistry databases through fingerprint similarity. The method is tested on synthetic data and on spectroscopic measurements of lab-scale batch hydrothermal liquefaction (HTL) of biomass to determine the accuracy of prediction across datasets of varying complexities. Our methodology is able to identify reaction network hypotheses containing reaction networks close to the ground truth in the case of synthetic data, and we are also able to recover candidate molecules and reaction networks close to the ones reported in the previous literature studies for biomass pyrolysis.

Application of Multivariate Data Analysis on Historical Recombinant Adenovirus Zoster Vaccine Production Data for Upstream Process Improvements

In recent years, multivariate data analysis (MVDA) has been widely used for process characterization and fault diagnosis in the biopharmaceutical industry. This study aims to investigate the feasibility of using MVDA for the development and scale-up of a perfusion process for HEK293 cell-based recombinant adenovirus zoster vaccine (Ad-HER) production. The Principal Component Analysis (PCA) results suggested comparable performance among the ATF, PATFP, and BFP perfusion systems in benchtop-scale stirred-tank bioreactor (STR). Then a Batch Evolution Model (BEM) was built using representative data from 10 L STR with a BFP system to assess the Ad-HER perfusion process performance at pilot-scale bioreactor (50 L STR and 50 L wave bioreactor). Furthermore, another BEM model and Batch Level Model (BLM) were built to monitor process parameters over time and predict the final adenovirus titer in 50 L wave bioreactor. The loading plot revealed that lactate dehydrogenase activity, viable cell diameter, and base-added during the virus production phase could be used as preliminary indicators of adenovirus yield. Finally, an adenovirus titer of 2.0±0.3×1010 IFU/mL was achieved in the 50 L wave bioreactor with BFP system, highlighting the robustness of the Ad-HER perfusion process at pilot-scale. Overall, this study emphasizes the effectiveness of MVDA as a tool for advancing the understanding of recombinant adenovirus vaccine perfusion production process development and scale-up.

Protocol development, validation, and troubleshooting of in-situ fiber optic bathless dissolution system (FODS) for a pharmaceutical drug testing

Currently, there is no systematic approach available for the validation, quantitative assessment, and troubleshooting for the in-situ fiber optic/bathless dissolution system (FODS). In this report, a dissolution protocol was developed and validated for a model product, chlorpheniramine maleate (CPM) 4 mg IR tablets. Dissolution runs were conducted at 37 ± 0.2 °C using a USP apparatus II, at 50 rpm in 500 mL of 0.01 N hydrochloric acid. The dissolution system was validated for linearity, accuracy, precision, specificity, and robustness analogously to an HPLC method validation. The linearity determination method was developed using five concentration levels between 25-125 % of the expected concentration, while for accuracy, 80 %, 100 %, and 120 % levels were used, and precision was determined using six runs at the 100 % level. Probe sampling depth, orientation, analytical wavelength, and paddle speed were varied to evaluate the robustness of the system tested. Method equivalence was established by comparing the dissolution results from FODS and the traditional dissolution method using UV spectrophotometry. Based on the statistics generated using the dissolution tests, the results are linear, accurate, precise, and specific. Robustness testing demonstrates that small changes in operating conditions did not significantly change the result. No significant difference in the amount dissolved at Q-timepoint was observed between FODS and traditional testing. Therefore, the FODS is a suitable alternative to traditional dissolution for CPM immediate-release tablets (many other drug products have been tested in the laboratory, and reports are in preparation). Additionally, the current work discusses problems related to media preparation, probe sensitivity, and excipient effects on data collected using FODS. The instrument-specific artifacts and data analysis problems are addressed and troubleshooting with possible solutions to eliminate or mitigate the errors. Although the FODS method was developed and evaluated using CPM in 500 mL dissolution volume, the dissolution method using a more common pharmacopoeial dissolution volume, i.e., 900 mL, was used to demonstrate the troubleshooting experiments for the drug products requiring 900 mL dissolution media.

Monitoring Parallel Robotic Cultivations with Online Multivariate Analysis

In conditional microbial screening, a limited number of candidate strains are tested at different conditions searching for the optimal operation strategy in production (e.g., temperature and pH shifts, media composition as well as feeding and induction strategies). To achieve this, cultivation volumes of >10 mL and advanced control schemes are required to allow appropriate sampling and analyses. Operations become even more complex when the analytical methods are integrated into the robot facility. Among other multivariate data analysis methods, principal component analysis (PCA) techniques have especially gained popularity in high throughput screening. However, an important issue specific to high throughput bioprocess development is the lack of so-called golden batches that could be used as a basis for multivariate analysis. In this study, we establish and present a program to monitor dynamic parallel cultivations in a high throughput facility. PCA was used for process monitoring and automated fault detection of 24 parallel running experiments using recombinant E. coli cells expressing three different fluorescence proteins as the model organism. This approach allowed for capturing events like stirrer failures and blockage of the aeration system and provided a good signal to noise ratio. The developed application can be easily integrated in existing data- and device-infrastructures, allowing automated and remote monitoring of parallel bioreactor systems.

Data fusion by joint non-negative matrix factorization for hypothesizing pseudo-chemistry using Bayesian networks

We infer reaction networks and chemistry using data fusion of spectroscopic sensors.

Decision Tree-PLS (DT-PLS) algorithm for the development of process: Specific local prediction models

This work presents a novel multivariate statistical algorithm, Decision Tree-PLS (DT-PLS), to improve the prediction and understanding of dynamic processes based on local partial least square regression (PLSR) models for characteristic process groups defined based on Decision Tree (DT) analysis. The DT-PLS algorithm is successfully applied to two different cell culture data sets, one obtained from bioreactors of 3.5 L lab scale and the other obtained from the 15 ml ambr microbioreactor system. Substantial improvement in the predictive capabilities of the model can be achieved based on the localization compared to the classical PLSR approach, which is implemented in the commercially available packages. Additionally, the differences in the model parameters of the local models suggest that the governing process variables vary for the different process regimes indicating the different states of the cell under different process conditions.

Raman Spectroscopy for Process Analytical Technologies of Pharmaceutical Secondary Manufacturing

As the process analytical technology (PAT) mindset is progressively introduced and adopted by the pharmaceutical companies, there is an increasing demand for effective and versatile real-time analyzers to address the quality assurance challenges of drug manufacturing. In the last decades, Raman spectroscopy has emerged as one of the most promising tools for non-destructive and fast characterization of the pharmaceutical processes. This review summarizes the achieved results of the real-time application of Raman spectroscopy in the field of the secondary manufacturing of pharmaceutical solid dosage forms, covering the most common secondary process steps of a tablet production line. In addition, the feasibility of Raman spectroscopy for real-time control is critically reviewed, and challenges and possible approaches to moving from real-time monitoring to process analytically controlled technologies (PACT) are discussed.

Advanced multivariate data analysis to determine the root cause of trisulfide bond formation in a novel antibody-peptide fusion

Product quality heterogeneities, such as a trisulfide bond (TSB) formation, can be influenced by multiple interacting process parameters. Identifying their root cause is a major challenge in biopharmaceutical production. To address this issue, this paper describes the novel application of advanced multivariate data analysis (MVDA) techniques to identify the process parameters influencing TSB formation in a novel recombinant antibody-peptide fusion expressed in mammalian cell culture. The screening dataset was generated with a high-throughput (HT) micro-bioreactor system (AmbrTM 15) using a design of experiments (DoE) approach. The complex dataset was firstly analyzed through the development of a multiple linear regression model focusing solely on the DoE inputs and identified the temperature, pH and initial nutrient feed day as important process parameters influencing this quality attribute. To further scrutinize the dataset, a partial least squares model was subsequently built incorporating both on-line and off-line process parameters and enabled accurate predictions of the TSB concentration at harvest. Process parameters identified by the models to promote and suppress TSB formation were implemented on five 7 L bioreactors and the resultant TSB concentrations were comparable to the model predictions. This study demonstrates the ability of MVDA to enable predictions of the key performance drivers influencing TSB formation that are valid also upon scale-up. Biotechnol. Bioeng. 2017;114: 2222-2234. © 2017 The Authors. Biotechnology and Bioengineering Published by Wiley Periodicals, Inc.

Multivariate statistical monitoring as applied to clean-in-place (CIP) and steam-in-place (SIP) operations in biopharmaceutical manufacturing

Multivariate statistical process monitoring (MSPM) is becoming increasingly utilized to further enhance process monitoring in the biopharmaceutical industry. MSPM can play a critical role when there are many measurements and these measurements are highly correlated, as is typical for many biopharmaceutical operations. Specifically, for processes such as cleaning-in-place (CIP) and steaming-in-place (SIP, also known as sterilization-in-place), control systems typically oversee the execution of the cycles, and verification of the outcome is based on offline assays. These offline assays add to delays and corrective actions may require additional setup times. Moreover, this conventional approach does not take interactive effects of process variables into account and cycle optimization opportunities as well as salient trends in the process may be missed. Therefore, more proactive and holistic online continued verification approaches are desirable. This article demonstrates the application of real-time MSPM to processes such as CIP and SIP with industrial examples. The proposed approach has significant potential for facilitating enhanced continuous verification, improved process understanding, abnormal situation detection, and predictive monitoring, as applied to CIP and SIP operations.

Batch-to-batch quality consistency evaluation of botanical drug products using multivariate statistical analysis of the chromatographic fingerprint

Botanical drug products have batch-to-batch quality variability due to botanical raw materials and the current manufacturing process. The rational evaluation and control of product quality consistency are essential to ensure the efficacy and safety. Chromatographic fingerprinting is an important and widely used tool to characterize the chemical composition of botanical drug products. Multivariate statistical analysis has showed its efficacy and applicability in the quality evaluation of many kinds of industrial products. In this paper, the combined use of multivariate statistical analysis and chromatographic fingerprinting is presented here to evaluate batch-to-batch quality consistency of botanical drug products. A typical botanical drug product in China, Shenmai injection, was selected as the example to demonstrate the feasibility of this approach. The high-performance liquid chromatographic fingerprint data of historical batches were collected from a traditional Chinese medicine manufacturing factory. Characteristic peaks were weighted by their variability among production batches. A principal component analysis model was established after outliers were modified or removed. Multivariate (Hotelling T(2) and DModX) control charts were finally successfully applied to evaluate the quality consistency. The results suggest useful applications for a combination of multivariate statistical analysis with chromatographic fingerprinting in batch-to-batch quality consistency evaluation for the manufacture of botanical drug products.

Towards better process understanding: chemometrics and multivariate measurements in manufacturing of solid dosage forms

The manufacturing of tablets involves many unit operations that possess multivariate and complex characteristics. The interactions between the material characteristics and process related variation are presently not comprehensively analyzed due to univariate detection methods. As a consequence, current best practice to control a typical process is to not allow process-related factors to vary i.e. lock the production parameters. The problem related to the lack of sufficient process understanding is still there: the variation within process and material properties is an intrinsic feature and cannot be compensated for with constant process parameters. Instead, a more comprehensive approach based on the use of multivariate tools for investigating processes should be applied. In the pharmaceutical field these methods are referred to as Process Analytical Technology (PAT) tools that aim to achieve a thorough understanding and control over the production process. PAT includes the frames for measurement as well as data analyzes and controlling for in-depth understanding, leading to more consistent and safer drug products with less batch rejections. In the optimal situation, by applying these techniques, destructive end-product testing could be avoided. In this paper the most prominent multivariate data analysis measuring tools within tablet manufacturing and basic research on operations are reviewed.

Combining microwave resonance technology to multivariate data analysis as a novel PAT tool to improve process understanding in fluid bed granulation

A set of 192 fluid bed granulation batches at industrial scale were in-line monitored using microwave resonance technology (MRT) to determine moisture, temperature and density of the granules. Multivariate data analysis techniques such as multiway partial least squares (PLS), multiway principal component analysis (PCA) and multivariate batch control charts were applied onto collected batch data sets. The combination of all these techniques, along with off-line particle size measurements, led to significantly increased process understanding. A seasonality effect could be put into evidence that impacted further processing through its influence on the final granule size. Moreover, it was demonstrated by means of a PLS that a relation between the particle size and the MRT measurements can be quantitatively defined, highlighting a potential ability of the MRT sensor to predict information about the final granule size. This study has contributed to improve a fluid bed granulation process, and the process knowledge obtained shows that the product quality can be built in process design, following Quality by Design (QbD) and Process Analytical Technology (PAT) principles.