Naslov
Prediction of hesperidin content in orange peel extract using artificial neural network model

Autori
Jokić, Stela ; Šafranko, Silvija ; Jakovljević, Martina ; Cikoš, Ana-Marija ; Bilić, Mate ; Molnar, Maja

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, prošireni sažetak, znanstveni

Izvornik
ECCE12 & ECAB5 Book of Abstracts / - Milano : Italian Association of Chemical Engineering (AIDIC), 2019, 1141-1142

ISBN
978-88-95608-75-4

Skup
12th European Congress of Chemical Engineering (ECCE 12) ; 5th European Congress of Applied Biotechnology (ECAB 5)

Mjesto i datum
Firenca, Italija, 15.09.2019. - 19.09.2019

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
Citrus peel ; Hesperidin ; Artificial neural network ; By-product

Sažetak
During industrial citrus processing, large quantities of waste material is produced mainly as citrus peels. These food industry by-products represent a potential source of valuable components being important raw materials in the food, chemical and pharmaceutical industries. Hence, the utilization of these citrus residues rich in bioactive and functional components has become a study of interest. In recent years, artificial neural networks (ANNs) are receiving more attention from researchers as an effective predictive tool. It has also been reported that ANN models can be used to predict extraction yields. So in this study, the prediction of hesperidin content as the main bioflavonoid in orange peel extracts was studied by ANN. In this work, an ultrasound-assisted extraction (UAE) was performed from reused orange peel (after SC-CO2 extraction) in order to obtain the extracts rich in hesperidin - bioflavonoid with wide range of pharmacological properties. A three-layer feed- forward backpropagation artificial neural network (FFBP-ANN) was proposed to investigate the influence of four operating parameters: extraction temperature (30, 50, 70 °C), time (15, 30, 45 min), ethanol/water ratio (20%, 50%, 80% v/v) and solvent-solid ratio (10, 30 and 50 mL/g) on the extraction yield of hesperidin in UAE extracts. The performance of the developed ANN predictive models was evaluated based on the obtained mean square error (MSE) and coefficient of determination (R2) parameters. The experimental hesperidin yield was determined by reversed-phase high performance liquid chromatography (HPLC) and its content was in the range from 3.3 to 23.0 μg/mg. Comparing developed models based on the AAD (Average Absolute Deviation), MSE (Mean Square Error), and R2 coefficient (Coefficient of determination), the best performing ANN model was determined. These statistical parameters are useful in assessing model performance. The obtained AAD of 5.24 %, R2 value of 0.9769 and 0.9837 and minimum MSE of 0.0108 and 0.00796 during training and testing stage indicated that developed 4-5-1 FFBP-ANN model is the best performing model in predicting the hesperidin yield for studied dataset. Yield prediction of target components is of great importance and the first step for defining the optimal operating conditions, but also necessary for a successful regulation of extraction processes. It is well- known that citrus peel is a rich source of bioactive natural compounds, therefore it is essential to find an appropriate technique for optimization of extraction process. The ANN predictive model was found to be a suitable performing model for extraction hesperidin yield prediction from orange peel extract, as indicated by the statistical analysis.

Izvorni jezik
Engleski

Znanstvena područja
Kemija, Kemijsko inženjerstvo, Biotehnologija

Napomena
The Event is organized by AIDIC, the Italian
Association of Chemical Engineering, under the
auspices of the European
Federation of Chemical Engineering (EFCE), as well
as of the European Society of Biochemical
Engineering Science
(ESBES).

POVEZANOST RADA