Application of ensemble learning to augment fluorescence-based PAT and enable real-time monitoring of protein refolding

Abstract

Recently, there has been a noticeable increase in the attention given to Process Analytical Technology (PAT) within biopharmaceutical manufacturing. Specifically, there has been a focus on using spectroscopic tools in conjunction with chemometrics analysis. The emergence of computational advancements, such as artificial intelligence and machine learning, has resulted in a shift away from traditional regression models towards more contemporary learning algorithms. Examples of these algorithms include artificial neural networks and ensemble learning. This transition has significantly enhanced the effectiveness of analytical tools within the field. This investigation aims to showcase the development and implementation of an ensemble-boosting regression model. This model relies on fluorescence spectroscopy to create a PAT tool based on XGBoost. This tool’s effectiveness is demonstrated by examining the refolding process of ranibizumab, a Fab fragment biotherapeutic. This case study highlights the model’s utility in real-time monitoring of protein refolding. The accuracy and efficacy of the proposed PAT tool are substantiated through an analysis of various statistical indicators. These include the coefficient of determination and its variants, which consistently fall within the range of 0.940–0.990. Additionally, the RMSE is 0.196, the NRMSE is 0.040, and the RMSE-CV is 0.637. These results provide evidence of the reliability and precision of the PAT tool. The overall framework has been implemented in real-time to validate the developed framework's robustness. The ability of the framework to accurately predict protein refolding has been examined, with the error percentage consistently below 10 percent. In conclusion, the proposed framework represents a comprehensive approach for integrating process understanding, advanced monitoring sensors, and control strategies to consistently achieve refolding targets over extended periods of continuous protein refolding.