SIM-CaliRam: A workflow combining simulation-informed references and biomass-augmented features for in-line Raman calibration of Saccharomyces cerevisiae.

Authors

Terrance Wilms, Steffi Knorn

Published in

Biotechnology progress. Nov 03, 2026. Epub Nov 03, 2026.

Abstract

Raman spectroscopy is a promising in-situ Process Analytical Technology for upstream bioprocesses, but transferring trained calibration models to new, independent fermentation runs without re-calibration remains a challenge due to sparse sample references (SaR) and prediction drift caused by biomass-induced fluorescence and turbidity effects. We present SIM-CaliRam, a calibration workflow combining two physically motivated interventions. First, Simulation-Informed References (SiR) densify glucose, ammonium, and phosphate labels using a feed-driven, non-reactive mass balance on a dedicated uninoculated non-reactive feed run, where substrate dynamics are determined by feeding and dilution. SiR labels agreed closely with the SaR values. Additionally, Biomass-augmented Feature (BaF) appends Raman-predicted biomass as an additional substrate-model regressor, providing a proxy for residual biomass-induced spectral confounding in the feature space. The workflow is demonstrated on Saccharomyces cerevisiae fermentations and evaluated within a systematic grid-search over established preprocessing options and six calibration methods spanning linear, Bayesian nonparametric, and neural network approaches. Conventional SaR calibration serves as the baseline. Training uses two AV runs, with model selection on a separate CV run and final evaluation on a held-out TEST run. The best selected models achieve NRMSE_TEST of 0.203 (biomass, SaR), 0.206 (glucose, SiR), 0.350 (ammonium, SiR + BaF), and 0.598 (phosphate, SiR). SiR improved phosphate prediction (R² = 0.64) relative to SaR (R² = -1.2), although the remaining prediction error limits standalone process-control application. Selected preprocessing, learner, and augmentation configurations differed across analytes. These results support target-specific model selection rather than a universal calibration pipeline.

PMID:
42977682
Bibliographic data and abstract were imported from PubMed on 04 Nov 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.