Evaluating data-mining strategies for label-free Raman microspectroscopic analysis of cellular processes in vitro:...

Understanding dynamic intracellular metabolism is essential for elucidating cellular function and disease mechanisms. However, analysing time-resolved spectral data is challenging due to overlapping signals, high dimensionality, and biological variability. This study evaluates the applicability of Multivariate Curve Resolution-Alternating Least Squares (MCR-ALS) for resolving complex cellular spectral data and introduces a framework that integrates Principal Component Analysis (PCA) of sequential time points as a preprocessing step. PCA was used to extract dominant spectral variance associated with biological kinetics (PC1), and MCR-ALS was subsequently applied to the sequential PC1 loadings to enhance the recovery of kinetic trends. To validate the approach, simulated Raman datasets were generated by combining predefined component spectra with known kinetic profiles, representing sequential (glycolysis-like) and parallel (glycolysis + glutaminolysis) models. These signals were superimposed onto real cellular spectra ("Sim-Cell" data) to include intercellular variability and background interference. Standard MCR-ALS was first assessed to determine the optimal number of components and, when applied to cell-free datasets, resolved up to three components. It was then applied to Sim-Cell datasets to evaluate performance under simulated biological conditions. MCR-ALS resolved the sequential model up to a cellular background weight of 2 but failed for the parallel model, highlighting limitations in complex, overlapping systems. In contrast, PCA-MCR-ALS substantially improved performance under strong cellular background, maintaining accurate resolution up to weight = 5 for sequential and weight = 8 for parallel models. Overall, this study benchmarks chemometric performance in label-free Raman microspectroscopy and highlights PCA-assisted MCR-ALS for analysing time-resolved intracellular spectral data.