Аннотация:Canonical tensor decomposition (also known as PARAFAC) is well matched to the underlying physical model of fluorescence signal and is frequently used to model fluorescence excitation-emission matrices (EEMs). However, EEMs typically contain scattering signal, which has to be handled separately as it doesn’t fit the assumptions of the PARAFAC model. Use of missing data in place of the scattering areas and their interpolation [1] are two widely used methods. In particular, interpolation makes it possible to avoid the local minima and convergence problems resulting from use of the PARAFAC model with a lot of missing data. On the other hand, interpolation may discard potentially useful information or result in artefacts, e.g. in case when second diffraction order scattering band overlaps with a fluorescence peak. In this work, a combination of PARAFAC and a bilinear model is suggested in order to model fluorescence and scattering signals together, similar to multivariate curve resolution (MCR) with trilinear constraints [2]. Currently, the model is fitted by alternating between the PARAFAC step and a constrained matrix factorisation step. Implementations in MATLAB and R programming languages are available. 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.
