Synthetic spectral libraries for Raman model calibration.

Raman spectroscopy has become increasingly popular in the process analytical technology (PAT) landscape due to its versatility and predictive capability in bioprocesses. However, model building remains a time-consuming and cost-intensive task. Building upon a fast calibration workflow based on physical pure compounds spiking in water, this work explores the novel use of in silico spiking of pure spectral fingerprints of various analytes. Through data fusion, a synthetic spectral library (SSL) is created that combines base spectra information from mammalian cell culture runs with matrix variability, as well as pure component spectra in water, aiming to greatly reduce the cost and time required for efficient model building. The findings indicate that the in silico addition of pure compounds provides spectral information comparable to physically spiked measurements. Consequently, this approach allows for the generation of an extensive number of information-rich spectra, forming a robust foundation for various regression algorithms and enhancing Raman calibration of existing spectral databases.