One-point calibration for Raman models in CHO fed-batch control

Primary source Biotechnology and Bioengineering: Enabling Raman Spectroscopy-Based Cell Culture Process Control via Real-Time Chemometric Model Calibration 4 Nov 2026

A peer-reviewed paper from authors at Johnson & Johnson Innovative Medicine (BioTherapeutics Development & Supply, and Manufacturing Statistics, Malvern, Pennsylvania) proposes a way to make Raman-based monitoring and control of cell culture easier to deploy. It was published online on 4 November 2026. The paper is not open access, so this brief is limited to what the abstract says. We did not read the full text.

The problem the paper targets

Raman spectroscopy paired with chemometric models, such as Partial Least Squares (PLS, a statistical method that links spectra to measured values), is described by the authors as a powerful enabling technology for real-time monitoring and control of industrial cell culture. The measurement is non-destructive and near real time. According to the abstract, industrial implementation has nonetheless been limited.

The abstract gives two reasons. First, chemometric model development needs extensive datasets, because dozens to hundreds of cell-culture constituents are known to contribute to Raman scattering. Second, experimental artifacts and anomalies from different sites, scales or equipment make the problem worse.

The proposed approach

The authors introduce what they call a novel one-point calibration methodology. It changes the training and validation goal from the “best chemometric model for the historical data” to the “best model for the most recent and representative data”. It is implemented within an adaptive-learning framework.

The abstract states that this approach drastically reduces dependence on extensive historical datasets. It also says the approach better accommodates site-, process- and run-to-run variability. The processes investigated were 9 to 14 day fed-batch Chinese Hamster Ovary (CHO) cell cultures.

What the abstract does not say

The abstract gives no quantitative results. It reports no prediction error, no number of calibration points, no list of analytes modeled and no comparison against a named baseline model. It does not say whether the method was run in closed-loop control at manufacturing scale. Those details would be in the full text, which we did not open.