Using Machine Learning During Process Chromatography Operations

Scientists from the Indian Institute of Technology Delhi in New Delhi, India, created a new machine learning (ML)-based tool for process verification during process chromatography unit operations. Their processes were published in the Journal of Chromatography A (1).

Artificial intelligence (AI), machine learning and modern computer technologies concepts. Business, Technology, Internet and network concept. | Image Credit: © photon_photo - stock.adobe.com

In 2011, the United States Food and Drug Administration (FDA) created new process validation guidelines, suggesting three discrete stages: process design (stage 1), process qualification (stage 2), and continued process verification (CPV) (stage 3). This process has significantly influenced the biopharmaceutical industry, especially regarding process control and quality assurance. With CPV, the goal is to assess and assure that critical quality attributes (CQAs) and critical process parameters (CPPs) stay within the design space, ensuring that a process is robustly controlled. If deviations occur, the control strategy should effectively deal with it and ensure consistent product quality.

Artificial intelligence-machine learning (AI-ML) methodologies are being used to design automation and control platforms that require minimal manual intervention. This facilitates migration from quality-by-testing (QbT) to quality-by-design (QbD), as recommended by the FDA. One type of AI-ML methodology is deep neural networks (DNNs), which can easily adapt and effectively represent intricate, non-linear connections that commonly occur in real-world datasets. Unlike other AI-ML models that may need human feature engineering or encounter difficulties with multidimensional data, DNNs can automatically identify complex patterns and relationships among features. When the quantity and complexity of datasets increase, DNNs are believed to adapt and perform effectively, all while maintaining or enhancing their performance.

In this study, a new CPV strategy was proposed, specifically regarding a cation exchange chromatography (CEX) unit operation. Statistical process control (SPC) charts were created using real-time measurements of various CPPs measured via in-built sensors (pH, conductivity, UV, and delta column pressure) and CQAs, such as charge variant composition (Raman spectroscopy) and concentration (near-infrared spectroscopy [NIR]).

With CPV highlighted in the FDA’s process validation guidance, the scientists realized that performing three or five batches during traditional process validation campaigns may be insufficient to ensure that a manufacturing process has the desired robustness and consistently produces products that meet preset specifications. With this study, the scientists showed how CPV for CEX process chromatography can be effectively used to facilitate real-time process monitoring and process control. Data from in-built sensors, such as pH, conductivity, and UV, in addition to more sophisticated analytical tools, such as Raman spectroscopy and NIR spectroscopy, were used to develop SPC charts that offer real-time visualization of process performance. Using Python script, deviations triggered alarms, letting end users know when to take appropriate action. The approach was then demonstrated by incorporating deviations into the load. Alert in the Raman based SPC chart of charge variants triggered model-based control, resulting in consistent charge variant composition in the CEX eluate as that of control runs (acidic ∼20 ± 2%, main ∼62 ± 2%, and basic ∼18 ± 2%).

Though this study focuses on CPV of CEX process chromatography, the scientists believe their approach can extend to any unit operation in bioprocessing where data signals can be captured to monitor CQAs. Further, AI-ML tools show great potential for the biopharmaceutical sector.

Reference

(1) Anupa, A.; Jesubalan, N. G.; Trivedi, R.; et al. Implementation of Machine Learning Tool for Continued Process Verification of Process Chromatography Unit Operation. J. Chromatogr. A 2025, 1742, 465642. DOI: 10.1016/j.chroma.2024.465642