Raman-guided sample subset selection for cost-efficient offline calibration in bioprocesses

Area of Science:

• Bioprocess Engineering
• Process Analytical Technology (PAT)
• Spectroscopy

Background:

• Model-based methods are crucial for understanding bioprocesses but are hindered by high costs and inefficient sampling.
• Insufficient prior knowledge complicates the development of accurate models for novel species or strains.

Purpose of the Study:

• To develop a model-free Process Analytical Technology (PAT) strategy for optimizing offline reference assays in bioprocess development.
• To introduce Raman-guided sample subset selection (RGSS) for prioritizing informative samples before model calibration.

Main Methods:

• RGSS was implemented using constrained vector quantization (CVQ) with full-spectrum and analyte-specific wavenumber selection.
• The strategy was demonstrated on Saccharomyces cerevisiae fed-batch fermentations.
• Performance was evaluated using auto- and cross-validation (CV) and identifiability analysis, benchmarked against uniform-in-time, random, and Kennard-Stone (KS) sampling.

Main Results:

• RGSS, using only five samples per analyte, achieved offline CV accuracy (NRMSE = 3.49) comparable to using all 22 samples (NRMSE = 3.29).
• KS selection was competitive with ten samples (NRMSE = 3.79).
• Uniform-in-time sampling showed significantly higher CV errors (8.47 for five samples, 65.57 for ten samples).

Conclusions:

• RGSS offers a cost-efficient, spectrally informed framework for prioritizing offline assays in model-based bioprocess development.
• The approach effectively reduces the number of required reference analyses without compromising model accuracy.
• RGSS enhances the efficiency and reduces the cost of developing models for bioprocesses.