Adaptive modeling optimized by the data fusion strategy: Real-time dying cell percentage prediction using capacitance...

Area of Science:

• Bioprocess Engineering
• Chemometrics
• Biotechnology

Background:

• Previous research utilized partial least squares (PLS) regression with capacitance spectra for cell death prediction.
• Existing models showed limitations in accuracy and adaptability during bioprocess operations.

Purpose of the Study:

• To enhance the accuracy and robustness of cell death prediction models.
• To develop an adaptive data fusion strategy for improved predictive performance.

Main Methods:

• Implemented a data fusion approach incorporating variables from the Cole-Cole model, conductivity (and its derivatives), and Mahalanobis distance into the predictor matrix.
• Utilized partial least squares (PLS) regression as the underlying modeling technique.
• Evaluated model performance using root-mean squared error of prediction (RMSEP) and robustness against reference spectrum selection.

Main Results:

• The data fusion model achieved a substantial reduction in RMSEP by approximately 50% compared to the prior PLS model based solely on capacitance spectra.
• Incorporating Cole-Cole parameters improved prediction accuracy, while conductivity data and Mahalanobis distance enhanced model adaptability and mitigated fluctuations.
• The model demonstrated consistent performance across different reference spectrum time points, confirming its robustness.

Conclusions:

• Data fusion significantly enhances the accuracy of cell death prediction models in bioprocesses.
• The adaptive modeling strategy offers a robust and improved approach compared to models relying solely on capacitance spectra.
• This methodology provides a more reliable tool for monitoring cell viability during biomanufacturing.