Lessons Learned from Deploying Adaptive Machine Learning Agents with Limited Data for Real-time Cell Culture Process Monitoring

2025-08-29

AI Summary

Real-time prediction of glucose, lactate, and ammonium ion concentrations in cell culture remains challenging due to scarce labeled data and high process variability. Method: This paper proposes an adaptive Raman spectroscopy analytics framework integrating a pre-trained model, just-in-time learning (JITL), and online learning, coupled with a mode-aware mixture-of-experts (MoE) architecture for dynamic model selection and continuous adaptation. Contribution/Results: The work systematically characterizes the applicability boundaries of these three learning paradigms across distinct bioprocess phases and demonstrates that the MoE architecture significantly enhances model robustness. Evaluated on two industrial-scale bioprocess datasets, the framework achieves high-accuracy metabolite concentration prediction (RMSE < 0.2 g/L) with low latency (<30 s), validating its effectiveness and engineering practicality in dynamic biomanufacturing environments.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

This study explores the deployment of three machine learning (ML) approaches for real-time prediction of glucose, lactate, and ammonium concentrations in cell culture processes, using Raman spectroscopy as input features. The research addresses challenges associated with limited data availability and process variability, providing a comparative analysis of pretrained models, just-in-time learning (JITL), and online learning algorithms. Two industrial case studies are presented to evaluate the impact of varying bioprocess conditions on model performance. The findings highlight the specific conditions under which pretrained models demonstrate superior predictive accuracy and identify scenarios where JITL or online learning approaches are more effective for adaptive process monitoring. This study also highlights the critical importance of updating the deployed models/agents with the latest offline analytical measurements during bioreactor operations to maintain the model performance against the changes in cell growth behaviours and operating conditions throughout the bioreactor run. Additionally, the study confirms the usefulness of a simple mixture-of-experts framework in achieving enhanced accuracy and robustness for real-time predictions of metabolite concentrations based on Raman spectral data. These insights contribute to the development of robust strategies for the efficient deployment of ML models in dynamic and changing biomanufacturing environments.

Problem

• Real-time prediction of metabolite concentrations using Raman spectroscopy
• Addressing limited data and process variability in bioprocessing
• Comparative analysis of adaptive ML approaches for monitoring

Innovation

• Using pretrained, JITL, and online learning algorithms
• Employing mixture-of-experts framework for enhanced accuracy
• Updating models with latest offline measurements during operation