Generic Chemometric Models for Metabolite Concentration Prediction Based on Raman Spectra

Area of Science:

• Biotechnology
• Process Analytical Technology (PAT)
• Chemometrics

Background:

• On-line process monitoring using chemometric models is established in pharmaceutical bioprocesses.
• Current models require extensive calibration and are inflexible to system or process changes, necessitating recalibration.
• This limits the widespread adoption and efficiency of real-time bioprocess monitoring.

Purpose of the Study:

• To develop generic partial least squares regression (PLSR) models for predicting metabolic compound concentrations in Chinese Hamster Ovary (CHO) cell cultivations.
• To demonstrate the flexibility and transferability of these models across different cell cultures, sites, and Raman spectrophotometers without recalibration.
• To reduce the calibration effort and improve the adaptability of chemometric models in bioprocess monitoring.

Main Methods:

• Compilation of a large and diverse Raman spectroscopic dataset from various CHO cell cultures across different sites and companies.
• Development of generic partial least squares regression (PLSR) models using this comprehensive dataset.
• Testing the developed generic models on independent datasets, including a dilution series in FMX-8 mod medium and an independent CHO cell culture, using different Raman spectrometers and setups.

Main Results:

• Generic PLSR models reliably predicted concentrations of glucose, lactate, and glutamine in CHO cell cultivations.
• The models demonstrated flexibility, accurately predicting compound concentrations from spectra acquired with different Raman spectrometers and setups.
• Prediction errors in testing were predominantly within an acceptable range (<10% relative error).

Conclusions:

• Generic chemometric models can be created and are transferable between processes without recalibration, given careful selection of diverse calibration data.
• This approach significantly reduces calibration effort and enhances the flexibility of on-line process monitoring in pharmaceutical bioprocesses.
• The study validates the potential of robust, generic chemometric models for reliable and adaptable bioprocess monitoring.