ИСТИНА

Fluorescence excitation-emission matrices (EEMs) are well matched to the PARAFAC decomposition due to the direct correspondence between the underlying physical process and the mathematical model being fitted. However, scattering signal typically present in EEMs is not well described by PARAFAC and must therefore be handled prior to calculating the decomposition. Interpolation of the areas containing the scattering signal [1] is a simple and widely used preprocessing method in PARAFAC decomposition. Compared to the insertion of missing data, it typically gives better results by avoiding local minima and physically impossible solutions allowed by the zero-error weight in the scatter areas. However, if a fluorophore overlaps with a second diffraction order scattering band, removing and interpolating the band may hide the component. Moreover, some interpolation methods are sensitive to noise in the spectral data and may introduce artefacts in the shape of the estimated PARAFAC components. In this work, an approach combining PARAFAC and a bilinear model is suggested for the purpose of modelling both fluorescence and scattering signal at the same time, similar to multivariate curve resolution (MCR) with trilinear constraints [2]. On every iteration of the algorithm, PARAFAC and MCR fit each other’s residuals with appropriate constraints (nonnegativity for both methods; MCR loadings fixed to zero outside the scattering bands). The approach has been tested on various EEM datasets, including synthetic data, amino acid mixtures, sugar process data, and seawater DOM. By eliminating a preprocessing step and taking more of the EEM information into account, we can conclude that the decomposition results obtained this way are more reliable than the classical methods. The reported study was funded by the Russian Foundation for Basic Research according to the research project No. 20-33-90280. References 1. M. Bahram, R. Bro, C. Stedmon, A. Afkhami. Journal of Chemometrics. 20 (2006), 99–105. 2. R. Tauler, I. Marqués, E. Casassas. Journal of Chemometrics. 12 (1998), 55–75.