Multi-objective optimization for ultrasound-assisted extraction of phytochemicals from Moroccan Retama raetam using a solvent system: Comparison of mixture design methodology and artificial neural network modeling. Issue 8 (24th May 2023)

• Record Type: Journal Article
• Authors: El Baakili, Aafaf; Fadil, Mouhcine; Es-Safi, Nour Eddine
• Is Part Of: Separation science and technology. Volume 58:Issue 8(2023)
• Page Start: 1433
• Page End: 1449
• Publication Date: 2023-05-24
• Subjects: Retama raetam -- phenolic compounds -- antioxidant activity -- mixture design methodology -- Anns modeling; Separation (Technology) -- Periodicals; 660.284205
• DOI: 10.1080/01496395.2023.2191156
• Languages: English
• ISSNs: 0149-6395

Abstract

ABSTRACT: This study was designed to predict the optimized solvent mixture for extracting phenolic compounds from Moroccan Retama raetam . A mixture design methodology (MDM) and artificial neural networks (ANNs) were employed in a solvent system containing water, ethanol, and methanol to optimize the three responses: extraction yield, total phenol content (TPC), and antioxidant activity by 1, 1-diphenyl-2-picrylhydrazyl (DPPH) assay. After the statistical analysis of the MDM results, the responses were separately and simultaneously optimized according to the validated special cubic models. The results obtained through MDM showed that the optimal solvent system was a water:methanol (41:59 v/v) binary mixture and the simultaneous optimal responses were 30.29%, 250 mg GAE/g, and 79.8 µg/ml for extraction yield, total phenolic content, and DPPHIC50, respectively. The mixture design matrix was submitted to the ANNs approach based on a multi-layer consisting of five neurons in a single hidden layer and three input components associated with three output responses. The results obtained through ANN were close to those obtained through MDM with an optimal binary mixture of water:methanol (38:62 v/v) and optimal responses of 28.97%, 256.42 mg GAE/g, and 83.07 µg/ml for extraction yield, total phenolic content, and DPPHIC50, respectively. The comparison of the two optimization approaches showed that both MDM and ANNs could accurately predict the studied responses. However, ANNs models' parameters were slightly higher than MDM ones. Our findings can provide a practical way for using both ANNs and MDM approaches in optimizing the bio-industry process and studying the bioactivity of selected compounds. GRAPHICAL ABSTRACT: uf0001