Raman spectral unmixing for quantitative analysis of multicomponent amino acid mixtures.

Real-time quantification of individual amino acids in mammalian cell culture using Raman spectroscopy remains challenging due to significant spectral overlaps and reliance on empirical regression chemometrics. Presented herein is Raman spectral unmixing which offers a mechanistically grounded alternative based on linear spectral additivity for quantitative analysis of amino acids in aqueous mixtures at millimolar concentrations, thus systematically linking intrinsic molecular spectral features to prediction performance. Initially, a comprehensive Raman spectral library comprising all 20 proteogenic amino acids was constructed under aqueous conditions. The single-component Raman spectra were analyzed to characterize spectra with distinctive patterns and spectral features such as peak intensity, and to provide a basis for interpreting subsequent spectral unmixing results. Spectral unmixing was then applied to binary, ternary, senary, and full 20-component mixtures to assess scalability under increasing spectral overlap. Prediction performance was evaluated by R