A comparison of PCA, ASCA, and vASCA for enhancing the interpretability of biological Raman spectroscopy

Raman spectroscopy is often used to analyze complex biological systems, but interpreting the resulting spectra remains difficult because of overlapping vibrational bands and the limitations of standard multivariate approaches such as Principal Component Analysis. This work assesses whether ANOVA-Simultaneous Component Analysis and its variable-selection extension can address these interpretability issues. More specifically, the goal was to determine if this approach could isolate spectral features that reflect metabolic syndrome and its progression over time.

Serum Raman spectra were collected from control rats and those with metabolic syndrome at four time points. The datasets were then analyzed comparatively using Principal Component Analysis, ANOVA-Simultaneous Component Analysis, and Variable-Selection ANOVA-Simultaneous Component Analysis, with attention not only to statistical separation but also to the interpretability of the resulting features in a biochemical context. Variable-Selection ANOVA-Simultaneous Component Analysis shows a narrower set of statistically relevant spectral regions. Among these, a Raman band in the 907-926 cm^-1 range appeared as a candidate signature associated with metabolic syndrome. Temporal variation was also evident, particularly in lipid- and protein-related bands between 8 and 52 weeks, although some overlap across groups persisted. Compared with the other methods, this approach yielded a clearer feature-based separation and, in most comparisons, stronger statistical contrasts.

The use of Variable-Selection ANOVA-Simultaneous Component Analysis makes the relationship between spectral features and the physiology of the disease clearer. Our approach paves the way to establish an interpretable tool to identify relevant spectral signatures in biological Raman spectra datasets.