(567e) Smart Process Analytics for the Prediction of Critical Quality Attributes in End-to-End Batch Manufacturing of Monoclonal Antibodies

Biopharmaceuticals are continuously growing in terms of both global sales and their percentage of the overall pharmaceutical pipeline [1]. Especially, monoclonal antibodies are number two in the number of drugs in the current pipeline, which is after the general class of small molecules produced via traditional synthetic chemistry techniques [2].

Depending on the degree of process understanding, the process models range from data-driven to mechanistic [3]. For many processes for manufacturing of biopharmaceuticals, mechanistic models are not available. This situation has required manufacturers to develop data-driven models using data analytics (DA)/machine learning (ML) methods [4]. Past studies have already demonstrated that DA/ML methods can construct accurate and reliable models for industrial biomanufacturing processes by identifying and building relationships between critical process parameters (CPPs) and CQAs [5,6]. Since the available DA/ML tools and software packages are rapidly increasing today, the challenge is how to select the best methods and tools for a specific biomanufacturing dataset to ensure that the most accurate and reliable model is constructed. The optimal selection of DA/ML tools requires a substantial level of expertise due to diverse nature of biomanufacturing data in terms of quantity and quality.

This challenge motivated the development of a smart process analytics (SPA) software, which automates the robust selection of methods and construction of models [5,6,7]. With the given datasets, the software first assesses the specific data characteristics such as nonlinearity, multicollinearity, and dynamics. Based on the data characteristics and the user objective, the software selects the most suitable DA/ML tool by following a decision tree which can be represented in the form of a triangle. The SPA software constructs the final model using a rigorous cross-validation procedure for optimal tradeoff between robustness (variance) and accuracy (bias). This presentation describes learnings obtained by the application of smart process data analytics to the prediction of critical quality attributes from biomanufacturing datasets. The automated determination of the best DA/ML tools for model construction and process understanding in SPA is validated by comparison to alternative DA/ML methods – including partial least squares, ridge regression, elastic net, and sparse partial least squares – and we describe the results of a detailed residual analysis that identifies biases and other deviations from method assumptions that occur in applications to such datasets. Motivated by these residual analyses, we describe extensions to SPA that improve the predictive accuracy while accounting for variations that occur on longer time scales than considered in past models.

[1] M.S. Hong, K.A. Severson, M. Jiang, A.E. Lu, J.C. Love, and R.D. Braatz. Challenges and opportunities of biopharmaceutical manufacturing control. Computers & Chemical Engineering, 110:106-114, 2018.

[2] I. Lloyd. Pharma R&D Annual Review 2021. London, UK: Informa, 2021.

[3] Basic Principles of GMP. Geneva, Switzerland: World Health Organization, 2008.

[4] M.S. Hong, W. Sun, A.E. Lu, and R.D. Braatz. Process analytical technology and digital biomanufacturing of monoclonal antibodies. American Pharmaceutical Review, 23(6):122-125, 2020.

[5] K. Severson, J.G. VanAntwerp, V. Natarajan, C. Antoniou, J. Thömmes, and R.D. Braatz. A systematic approach to process data analytics in pharmaceutical manufacturing: The data analytics triangle and its application to the manufacturing of a monoclonal antibody. In: Multivariate Analysis in the Pharmaceutical Industry, London, UK: Academic Press, 295-312, 2018.

[6] K. Severson, J.G. VanAntwerp, V. Natarajan, C. Antoniou, J. Thömmes, and R.D. Braatz. Elastic net with Monte Carlo sampling for data-based modeling in biopharmaceutical manufacturing facilities. Computers & Chemical Engineering, 80:30-36, 2015.

[7] W. Sun and R.D. Braatz. Smart process analytics for predictive modeling. Computers & Chemical Engineering, 144:107134, 2021.