Аннотация:The PARAFAC decomposition is widely used in the analysis of fluorescence data, owing to itsdirect correspondence to the underlying physical processes. Unfortunately, excitation-emissionmatrices (EEMs) typically contain scattering signals in addition to fluorescence, making it arequirement to handle the scattering signal either before or as part of the PARAFAC decomposition.Interpolation of the areas affected by the scattering signal [1] makes it possible to performPARAFAC decomposition of EEMs while avoiding the local minima, but it may introduce artefactsin the shape of the resulting loadings, or even hide a component if it happens to fully overlap with asecond order scattering band.In this work, an approach based similar to multivariate curve resolution (MCR) with trilinearconstraints [2] is applied to the task of modelling both the fluorescence and the scattering signal. Onevery iteration of the algorithm, PARAFAC and MCR fit each other’s residuals, converging towardsfluorescence being described by the PARAFAC model and the scattering signal in the MCR model.Typical limitations of this approach are nonlinearities arising from the detector being close tosaturation when measuring scattering signal. Constraints fixing the values of the MCR componentsoutside the scattering bands to zeros, use of multiple MCR components, and strategic positioning ofmissing data may be required to deal with these limitations. The approach has been tested onvarious EEM datasets, including fully synthetic, amino acid mixtures, sugar process data, andseawater DOM. Example results of the suggested approach – emission and excitation PARAFACloadings and the scattering signal loadings – are presented in the Figure. We can also conclude thatthe suggested approach provides better level of precision in reconstruction of original components.The reported study was funded by RFBR, project number 20-33-90280.[1] Bahram M.; Bro R.; Stedmon C.; Afkhami A.; Handling of Rayleigh and Raman scatter for PARAFAC modeling offluorescence data using interpolation, Journal of Chemometrics. 2006, 20, 99–105.[2] Tauler R.; Marqués I.; Casassas E.; Multivariate curve resolution applied to three-way trilinear data: Study of aspectrofluorimetric acid–base titration of salicylic acid at three excitation wavelengths, Journal of Chemometrics, 1998,12, 55–75.
