“…Using the same methods as the first combined UV‐Vis and Raman calibration, we created new stock solutions containing 12.01 g/L total protein and 0.3% aggregation in the monomer‐enriched sample and 12.04 g/L total protein and 6.3% aggregation in the aggregates‐enriched sample. We were confident from our previous results (Figure 3) that our system was capable of producing stable mixtures for calibrating both UV‐Vis scans and Raman spectra, thus we updated our analysis of the Raman spectra to include methods that have been recently successful for calibration model building, including partial‐least squares (PLS) regression modeling (Brestrich et al, 2018; Wei et al, 2021) and PCR (McAvan et al, 2020; Ramakrishna et al, 2022; Silva et al, 2020), as well as nonlinear models such as Gaussian processes (GPs) (Tulsyan et al, 2020). Model hyperparameter selection was made by subjective judgment from published values and included: 10 for the number of PLS and PCA latent variables, and 60% as k , for the KNN model that we used instead of a GP model to reduce risk of overfitting.…”