Raman spectroscopic quantification of phycocyanin and process metadata-based data fusion for dry cell weight in Spirulina platensis cultivation.

Rapid and non-destructive quantification of intracellular target products is crucial for rapid at-line process control in modern microalgae biorefineries. This study established a small-sample (50 independent biological samples) quantitative monitoring framework based on Raman spectra for live Spirulina platensis cells. For intracellular phycocyanin (PC) content quantification, a linear combination of raw full-range spectra, regression coefficients (RC) feature selection, and partial least squares regression (PLSR) delivered the optimal performance, achieving a prediction coefficient of determination (R2) of 0.944 and a relative prediction deviation (RPD) of 4.22. To further extend this monitoring framework to the macro-biomass indicator, dry cell weight (DCW) quantification was simultaneously evaluated. Pure Raman spectral signals, however, encountered non-linear matrix interferences for DCW (RPD < 2.00). To overcome this bottleneck, a multi-source information fusion strategy combining process metadata (strain, light, and cultivation time) side-by-side with optimized spectral features into a fused input matrix was developed. The resulting non-linear Savitzky-Golay (SG) smoothing, successive projections algorithm (SPA) feature selection, and support vector regression (SVR) framework significantly improved the R2 and RPD for DCW to 0.937 and 3.98, respectively. Furthermore, Shapley Additive exPlanations (SHAP) analysis successfully demystified the model's black-box nature, revealing that cultivation time exerted a dominant contribution (mean absolute SHAP value = 0.855) to DCW accumulation. This strategy reduces the reliance on massive databases, offering a highly transparent, low-sample-cost, and lightweight novel solution for implementing intelligent microalgae process analytical technology.