Implementation of machine learning tool for continued process verification of process chromatography unit operation.

Abstract

Recent advancements in technology, such as the emergence of artificial intelligence (AI) and machine learning (ML), have facilitated the progression of the biopharmaceutical industry toward the implementation of Industry 4.0. As per the guidelines set by the USFDA, process validation for biopharmaceutical production consists of three stages: process design, process qualification, and continuous process verification (CPV). This paper proposes a strategy for achieving CPV for a cation exchange chromatography unit operation, emphasizing the urgent need for such strategies in the biopharmaceutical industry. Statistical process control (SPC) charts were generated based on real-time measurement of the various critical process parameters (CPPs) measured via in-built sensors (pH, conductivity, UV, and delta column pressure) as well as of critical quality attributes (CQAs) like charge variant composition (Raman spectroscopy) and concentration (Near infrared spectroscopy). A Python-based program was created to read these SPC charts and respond to any deviation. The developed models for NIR coupled DNN PAT tool and Raman coupled DNN PAT tool exhibited satisfactory R² values (> 0.90), highlighting the statistical significance of the proposed model. Further, the control strategy designed based on Raman spectroscopy for charge variant composition in CEX eluate has been demonstrated by intentional perturbations in the CEX load. The resulting CEX eluate output showed a consistent charge variant composition as that of control runs (acidic ∼20 ± 2 %, main ∼62 ± 2 % and basic ∼18 ± 2 %). It has been demonstrated how an appropriate selection of analyzers, soft sensors, and advanced data analytics can be used to execute CPV and enable the biopharmaceutical industry to implement Industry 4.0.