© 2020 All Rights Reserved

Unraveling molecular profiles in skin in vivo using Raman microspectroscopy and non-negative matrix factorization

A.V. Venets 1, B.P. Yakimov 1,*, J. Schleusener 3, V.V. Fadeev 1, M.E. Darvin 3, E.A. Shirshin 1,2

1 M.V. Lomonosov Moscow State University, Faculty of physics, 1-2 Leninskie Gory, Moscow, 119991, Russia

2 Institute of Spectroscopy of the Russian Academy of Sciences, Fizicheskaya Str., 5, 108840, Troitsk, Moscow, Russia

3 Charité – Universitätsmedizin Berlin, corporate member of Freie Universität Berlin, Humboldt-Universität zu Berlin, and Berlin Institute of Health, Department of Dermatology, Venerology and Allergology, Center of Experimental and Applied Cutaneous Physiology, Charitéplatz 1, Berlin, 10117, Germany


Understanding the molecular composition of the skin is essential for numerous applications in dermatology and cosmetology. Molecular composition at different depths of the epidermis and dermis can be non-invasively assessed using Raman microspectroscopy. However, the optical Raman response obtained from the skin is a complex mixture of signals emitted from a wide variety of proteins, amino acids, lipids, which are components of the skin. Therefore it is of great concern to establish which main components contribute to the optical signal and what physiologically important information they carry. A standard approach to determine the molecular profiles of various skin components is spectral decomposition into predefined components, such as keratin, ceramides, cholesterol, natural moisturizing factor, etc. However, fixing the components and their spectra can lead to the loss of important information related to both the presence of other components and possible variations in the spectra of specified components due to changes in the physicochemical conditions that occur in vivo.

In this work, we investigate an alternative approach to the analysis of the Raman spectra of skin based on non-negative matrix factorization [1]. Non-negative matrix factorization is an unsupervised learning technique that attempts to represent input data with non-negative values (such as Raman spectra) as a weighted sum of independent non-negative components. In contrast to other unsupervised methods, for example, the principal component analysis, the imposed restriction on non-negativity of decomposition makes it possible to identify physically relevant components that are easy to interpret. Thus, we show that the application of this technique to the depth-resolved Raman spectra of skin makes it possible to isolate the molecular components that have the largest contribution to the Raman spectra, to obtain information on the conformation of proteins, and also to determine the profiles of penetrating agents, without a priori information on the spectra of individual skin component. We believe that this technique could be a convenient supportive tool in the analysis of Raman spectra of the skin and determination of molecular profiles of various components.

This work was supported by the Russian Foundation for Basic Research (grant No. 19-02-00947).

[1] Lee, Daniel D., and H. Sebastian Seung. "Learning the parts of objects by non-negative matrix factorization." Nature 401.6755 (1999): 788-791.

File with abstract


Boris Yakimov
Lomonosov Moscow State University


File with report


Ask question