Machine learning modeling aids intelligent process analysis for high-performance virus filtration

A research team led by Professor Wang Inhua of the Institute for Process Engineering (IPE) at the Chinese Academy of Sciences has developed a machine learning (ML) framework for analyzing the virus filtration process in therapeutic protein purification. The new method allows for intelligent identification of key parameters that affect virus retention efficiency and provides predictive guidance for process optimization. The survey results have been published on Journal of Membrane Science.

Virus filtration is an important step in ensuring the safety of biopharmaceuticals, but this process is complicated by the interaction of membrane properties, operational settings, and solution environments. Traditional approaches struggle to capture nonlinear relationships and synergistic effects between factors such as membrane type, flux, protein concentration, and volume throughput, limiting efficient process development.

To address these challenges, researchers collected over 900 datasets from peer-reviewed publications, trained ML models, constructed a database of virus clearance processes to analyze the complex binding effects of key parameters, and established a data-driven approach for efficient development of virus removal processes.

Characteristic importance assessments prioritized key viral filtration parameters and provided guidance on process optimization. Univariate partial-dependent plot (PDP) analysis revealed the independent influence of each variable on viral maintenance, whereas bivariate PDP analysis provided increased fluxes to provide data-driven mechanical insights, reducing the effect of negative variable interactions with bundles and providing data-driven mechanical insights.

Validation experiments confirmed a strong agreement between model prediction and experimentally measured LRV values, and the model maintained accuracy beyond the original dataset range, showing robust extrapolation capabilities and important engineering values.

“Our framework reduces the reliance on trial and error experiments, allowing for rapid process prediction, and addresses the standardization of data protocols essential to enhancing integrated processes,” said Professor Hwang Long, author of the study.

This study establishes a data-driven optimization paradigm of viral filtration and provides a scalable ML solution to intelligently enhance downstream biopharmaceutical purification.