![]() |
ИСТИНА |
Войти в систему Регистрация |
ФНКЦ РР |
||
The last decade witnessed the use of carbon nanoparticles as drug carriers becoming more and more widespread. USA Food and Drug Administration has already approved nearly 20 nano-based drugs. The use of any drug is associated with the need to control its clearance from the body in a certain period, including control of drug carrier clearance, which in this study are carbon nanoparticles. This study presents an approach to solve the problem of controlling the elimination of carbon dots (CD) -based nanoagents from the body via urine using optical spectroscopy and artificial neural networks (ANN). The use of ANN is conditioned by significant variability of urine optical properties. Thus, the shape of the urine fluorescence spectrum depends on many factors that are difficult to control such as donor's age, gender, nutrition and health status, time of sampling, etc. That is why standard methods of processing spectroscopic data are not applicable to this problem. In this study 624 suspensions of CD nanocomplexes with an anti-cancer drug – doxorubicin (Dox) – adsorbed on them in urine were prepared. CD and Dox concentrations varied from 0 to 1.2 mg/L with 0.05 mg/L increment, and from 0 to 1 mg/L with 0.042 mg/L increment, respectively. The problem was solved using three sets of optical spectroscopy data: fluorescence spectra of aqueous suspensions of CD and Dox in urine upon excitation by radiation with wavelengths of 405 nm, 532 nm, as well as optical absorption spectra in the range from 190 to 800 nm. Fully connected neural networks – multilayer perceptrons (MLP) – were applied to the obtained spectral data, which ensured the determination of CD and Dox concentrations in urine. To increase the accuracy of monitoring CD and Dox clearance via urine, autoencoders were additionally used. This study contains a comparative analysis of the results obtained using the mentioned spectroscopic methods. The factors influencing the results of using artificial neural networks are discussed in detail. It was found that the smallest error in determining the desired parameters from the absorption spectra is provided by an MLP with one hidden layer and 64 neurons in it. The mean absolute error in determining the concentrations of CD and Dox was 0.042 mg/L (3.6% of the maximum value) and 0.024 mg/L (2.4% of the maximum value), respectively. This study has been supported by Russian Foundation for Basic Research No. 19-01-00738 (K.A. Laptinskiy). The contribution of O.E. Sarmanova (programming and ANN training) was supported by the Foundation for the Advancement of Theoretical Physics and Mathematics “BASIS” (Project No. 19-2-6-6-1).