Near infrared (NIR) spectroscopy has the capability of providing real-time, multi-analyte monitoring of the complex reaction mixture associated with cell culture processes. However, the development of robust models to predict the concentration of key analytes has proven difficult. In this study, a modeling methodology using semisynthetic process samples was used to predict glucose concentrations in Chinese Hamster Ovary (CHO) cell culture processes. Partial Least Squares (PLS) regression models were built from in situ NIR spectra, and glucose levels between 4.0 and 14.0 g/L. Two models were constructed. The "standard model" used data provided by cell culture production process samples. The "full model" included the data provided from both cell culture production process samples and semisynthetic samples. The semisynthetic samples were generated by titrating cell culture samples with target viable cell density (VCD) and lactate levels to defined glucose concentrations. The robustness of each model was gauged by predicting glucose in a subsequent cell culture process utilizing a media formulation and cell line not contained in the calibration data sets. The "full model" generated glucose predictions with a root mean square error of prediction (RMSEP) of 0.99 g/L while the "standard model" provided glucose predictions with a RMSEP of 2.26 g/L. The modeling approach utilizing semisynthetic samples proved to be faster development and more effective than using just standard cell culture processes.
Many industrial pharmaceutical manufacturing processes are composed of multiple-step batch operations. However, uncontrolled variations often occur during operations that affect the cell growth performance. Because of the complexity of biological processes, one leading challenge in process operation is the identification of potential causes of undesirable process variabilities. In this paper, we propose a classification and diagnosis strategy to analyze cell culture manufacturing variability in bioprocesses with the objective of unveiling hidden factors affecting process yield and performance. The proposed strategy includes two parts: (i) a clustering method performed in the principal component residuals that effectively separates the low lactate batches into different clusters and (ii) a fault diagnosis method based on regularized LDA contribution analysis for exploring the leading contributors to each of the low performance classes. The proposed strategy is applied to industrial production data collected over 8 years from a batch process. The effectiveness of the proposed approach is demonstrated on this data set, allowing the classification and diagnosis of the sources of the low performance situations.
In the pharmaceutical industry, the goal of a supply planner is to make efficient capacity allocation decisions that ensure an uninterrupted supply of drug products to patients and to maintain product inventory levels close to the target stock. This task can be challenging due to the limited availability of manufacturing assets, uncertainties in product demand, fluctuations in production yields, and unplanned site downtimes. It is not uncommon to observe uneven distribution of product inventories with some products carrying excess inventories, while other products may be close to a stockout. Maintaining high stock levels can have economic repercussions due to the risk of expiration of unused products (whereas products facing a stockout can adversely affect the treatment regimen of patients). The network complexity of pharmaceutical supply‐chains coupled with regulatory constraints and siloed planning systems force supply planners to rely on manual (error‐prone) decision‐making processes. Such an approach results in suboptimal capacity allocation and inventory management decisions. In this work, we propose a stochastic optimization methodology for the production scheduling of multiple drug products in lyophilization units across multiple sites. The framework leverages information obtained from historical and forecast data to generate scenarios of uncertain parameters (e.g., yield, demand, and downtimes) that can realize in the future. The optimization model determines a product filling schedule that maintains product stock levels close to targets under diverse scenarios. We show that this approach helps in avoiding reactive scheduling and in maintaining a more robust production plan than deterministic procedures (which ignore uncertainty). Specifically, planning under a stochastic optimization approach reduces the number of scenarios under which backlogs are observed and also reduces the magnitude of the backlogs.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.