Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Learning From Limited Data and Feedback for Cell Culture Process Monitoring: A Comparative Study

Johnny Peng, Thanh Tung Khuat, Ellen Otte, Katarzyna Musial, Bogdan Gabrys

TL;DR

This paper tackles the problem of building accurate soft sensors for real-time bioprocess batch monitoring under limited data and infrequent feedback. It benchmarks dimensionality reduction, just-in-time learning, and online learning across an in silico IndPenSim dataset and two real-world datasets (CSL and AstraZeneca), using a standardized pipeline with leave-one-batch-out validation and Friedman/Nemenyi testing. Key findings show that adaptivity through JITL and OL generally outperforms static pre-training in cold-start scenarios, that transferability is strongly influenced by meta-features like feed media and control strategy, and that hybrid fusion of Raman spectroscopy with lagged offline measurements improves monitoring accuracy. The results provide practical guidance for designing robust bioprocess soft sensors and highlight future directions in meta-learning and multi-source data fusion to handle heterogeneity and sparsity in upstream bioprocess monitoring.

Abstract

In cell culture bioprocessing, real-time batch process monitoring (BPM) refers to the continuous tracking and analysis of key process variables such as viable cell density, nutrient levels, metabolite concentrations, and product titer throughout the duration of a batch run. This enables early detection of deviations and supports timely control actions to ensure optimal cell growth and product quality. BPM plays a critical role in ensuring the quality and regulatory compliance of biopharmaceutical manufacturing processes. However, the development of accurate soft sensors for BPM is hindered by key challenges, including limited historical data, infrequent feedback, heterogeneous process conditions, and high-dimensional sensory inputs. This study presents a comprehensive benchmarking analysis of machine learning (ML) methods designed to address these challenges, with a focus on learning from historical data with limited volume and relevance in the context of bioprocess monitoring. We evaluate multiple ML approaches including feature dimensionality reduction, online learning, and just-in-time learning across three datasets, one in silico dataset and two real-world experimental datasets. Our findings highlight the importance of training strategies in handling limited data and feedback, with batch learning proving effective in homogeneous settings, while just-in-time learning and online learning demonstrate superior adaptability in cold-start scenarios. Additionally, we identify key meta-features, such as feed media composition and process control strategies, that significantly impact model transferability. The results also suggest that integrating Raman-based predictions with lagged offline measurements enhances monitoring accuracy, offering a promising direction for future bioprocess soft sensor development.

Learning From Limited Data and Feedback for Cell Culture Process Monitoring: A Comparative Study

TL;DR

This paper tackles the problem of building accurate soft sensors for real-time bioprocess batch monitoring under limited data and infrequent feedback. It benchmarks dimensionality reduction, just-in-time learning, and online learning across an in silico IndPenSim dataset and two real-world datasets (CSL and AstraZeneca), using a standardized pipeline with leave-one-batch-out validation and Friedman/Nemenyi testing. Key findings show that adaptivity through JITL and OL generally outperforms static pre-training in cold-start scenarios, that transferability is strongly influenced by meta-features like feed media and control strategy, and that hybrid fusion of Raman spectroscopy with lagged offline measurements improves monitoring accuracy. The results provide practical guidance for designing robust bioprocess soft sensors and highlight future directions in meta-learning and multi-source data fusion to handle heterogeneity and sparsity in upstream bioprocess monitoring.

Abstract

In cell culture bioprocessing, real-time batch process monitoring (BPM) refers to the continuous tracking and analysis of key process variables such as viable cell density, nutrient levels, metabolite concentrations, and product titer throughout the duration of a batch run. This enables early detection of deviations and supports timely control actions to ensure optimal cell growth and product quality. BPM plays a critical role in ensuring the quality and regulatory compliance of biopharmaceutical manufacturing processes. However, the development of accurate soft sensors for BPM is hindered by key challenges, including limited historical data, infrequent feedback, heterogeneous process conditions, and high-dimensional sensory inputs. This study presents a comprehensive benchmarking analysis of machine learning (ML) methods designed to address these challenges, with a focus on learning from historical data with limited volume and relevance in the context of bioprocess monitoring. We evaluate multiple ML approaches including feature dimensionality reduction, online learning, and just-in-time learning across three datasets, one in silico dataset and two real-world experimental datasets. Our findings highlight the importance of training strategies in handling limited data and feedback, with batch learning proving effective in homogeneous settings, while just-in-time learning and online learning demonstrate superior adaptability in cold-start scenarios. Additionally, we identify key meta-features, such as feed media composition and process control strategies, that significantly impact model transferability. The results also suggest that integrating Raman-based predictions with lagged offline measurements enhances monitoring accuracy, offering a promising direction for future bioprocess soft sensor development.

Paper Structure

This paper contains 34 sections, 15 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Research Problem and Aims
  3. Benchmark Experiment Design
  4. Datasets
  5. IndPenSim In Silico Dataset
  6. CSL Experimental Dataset
  7. AstraZeneca Experimental Dataset
  8. Experimentation Pipeline and Methodology
  9. Data Preprocessing
  10. Machine Learning Models
  11. Dimensionality Reduction Method
  12. Hyperparameter Tuning and Testing Strategy
  13. Training Strategies
  14. Training Strategy 0 – No Training (Baseline Models)
  15. Training Strategy 1 – Pre-training
  16. ...and 19 more sections

Figures (15)

  • Figure 1: Experiment Pipeline Design
  • Figure 2: Heatmap of NMAE for predicting Lactate
  • Figure 3: Critical Difference Plot for Offline Measurements
  • Figure 4: NMAE Heatmap for Peniciling Concentration Prediction - Training Batch on X-axis, Target Batch on Y-axis
  • Figure 5: Critical Difference Plot for comparing the performance of ML models for Penicillin Concentration Prediction
  • ...and 10 more figures