Bioprocesses often depend on variables such as biomass and protein concentration that are difficult to measure continuously in real time. This creates a challenge for monitoring and control, especially when fast decisions are needed during production.
Our new publication in Chemical Engineering Transactions (Acosta-Pavas et al., 2025) explores how machine-learning-based soft sensors can help bridge this gap. The study focuses on recombinant nanobody production in E. coli for potential use in scorpion antivenom, testing whether routinely measured online process data can be used to estimate biomass and protein concentrations that would otherwise require offline analysis.
The researchers compared four machine learning approaches and found strong predictive performance for both protein and biomass estimation. K-Nearest Neighbors and Radial Basis Function Support Vector Machine performed particularly well for protein prediction, while all tested models showed strong results for biomass estimation. The study also used SHAP analysis to examine which process variables contributed most strongly to the model predictions, adding an important layer of interpretability.
To test whether the approach could remain useful under different process conditions, the biomass soft sensors were also evaluated in a scale-down setting using only four online variables: temperature, pO₂, agitation and pH. Performance decreased compared with the full model, but the results still showed that soft sensors can support biomass estimation with a reduced set of measurements. The authors conclude that these tools offer promising potential for real-time monitoring and faster decision-making in bioprocesses, with future work aimed at industrial-scale applications.
Read the whole publication:
Acosta-Pavas, J. C., Corrales, D. C., Alonso Villela, S. M., Georgakilas, G., Bouhaouala-Zahar, B., Cescut, J., Fillaudeau, L., & Aceves-Lara, C. A. (2025). Soft-Sensors Strategy for Estimating the Cell Biomass and the Nanobody-Based Scorpion Antivenom Production: from ML Algorithms Selection to Scale-Down Bioprocess Application. Chemical Engineering Transactions, 119, 349–354. https://doi.org/10.3303/CET25119059
