Fermentation process tracking through enhanced spectral calibration modeling.

Area of Science:

• Analytical Chemistry
• Chemometrics
• Process Analytical Technology (PAT)

Background:

• The FDA's Process Analytical Technology (PAT) initiative drives the adoption of spectroscopic instrumentation.
• Effective calibration models are crucial for maximizing information from spectroscopic data, especially in complex systems like fermentation.
• Traditional calibration methods using Partial Least Squares (PLS) on full or subsets of spectra have limitations in handling spectral complexity and environmental impacts.

Purpose of the Study:

• To develop a robust and informative calibration modeling procedure for spectroscopic data analysis.
• To address the challenges in interpreting complex spectral data from fermentation processes.
• To improve the accuracy and interpretability of calibration models compared to existing methods.

Main Methods:

• A novel calibration modeling approach combining wavelength selection via Spectral Window Selection (SWS).
• Automatic selection of spectral windows, which form the basis for calibration models.
• Ensemble modeling by combining multiple models generated from non-unique windows using stacking to enhance robustness.

Main Results:

• The proposed methodology was applied to on-line Near-Infrared (NIR) and Mid-Infrared (MIR) spectroscopic data from an industrial fermentation process.
• The new calibration modeling procedure demonstrated superior performance compared to traditional calibration methods.
• The approach successfully identified critical spectral regions relevant to the fermentation process.

Conclusions:

• The developed calibration modeling strategy offers a significant improvement for analyzing spectroscopic data in industrial fermentation.
• Spectral Window Selection (SWS) combined with stacking provides a robust method for building accurate and interpretable calibration models.
• This approach effectively extracts maximum information content from spectroscopic measurements, supporting the goals of Process Analytical Technology (PAT).