Exploration of multivariate curve resolution- alternating least squares (MCR-ALS) for datamining kinetically evolving complex cellular spectroscopic data (Spectralomics)

Multivariate curve resolution- alternating least squares (MCR-ALS) approach for datamining the complex spectral fingerprints from kinetically evolving cellular Raman spectroscopy data was explored in this study. Principal components analysis and partial least squares- discriminant analysis indicated the metabolic changes were captured in individual metabolic conditions (Control, Stimulation and Inhibition) as a function of time; however, MCR-ALS could not resolve the spectral components accurately. Hence simulated datasets were generated to test the limit of resolution which revealed the significance of initial estimation of spectral components in the MCR, and the effect of equality constraints in the ALS was studied. The resolved rate constants for the time evolution of the components were not quantitatively accurate at higher cellular background overlayed on the evolving components, although they did exhibit a consistent qualitative trend across the modulated conditions. Hence, the cellular data was analysed qualitatively, and the initial estimates constraint in MCR along with a kinetic hard model constraint in ALS was deduced to be the best strategy for datamining complex cellular spectra. The spectral fingerprints of both glycolytic and non-glycolytic cellular processes were resolved in all the modulated conditions, highlighting the high-content insights from the label-free approach. The study demonstrates the potential of Raman spectroscopy coupled with a spectralomics approach for datamining of the complex spectral fingerprints as a function of time and highlights its limitations. This approach could potentially find applications in high-content drug screening, drug discovery, disease diagnostics and process analytical techniques for monitoring bioprocesses.