Less Is More: Practical Insights Into Multivariate Regression Models for Raman Spectroscopy in Bioprocess Monitoring

Abstract

Process Analytical Technology (PAT) has been encouraged in bioprocess industries as a transformative approach for real-time process monitoring. However, two decades after its introduction, PAT implementation remains limited, particularly in quantitative online monitoring, partly due to the perceived complexity of chemometrics and contradictory communication from PAT suppliers. This review aims to demystify chemometrics in the context of multivariate regression models based on Raman spectroscopy, a versatile tool for bioprocess monitoring. By providing a critically evaluated workflow for model development—that is, Pretreatment, pre-processing, modelling, and evaluation—this review advocates for practical, simplified methodologies emphasizing robustness, transparency, and maintainability over convoluted processes. Two applications of Raman-based models, glucose monitoring in fermentation and lactate monitoring in cell culture, are herein examined to highlight common pitfalls, best practices, and opportunities for improvement. Ultimately, this report seeks to challenge the idea that chemometrics is inaccessible and provide practical insights to enable researchers to develop accurate and reliable models for real-world bioprocess applications.