Comparative study of response surface methodology and artificial neural network in the optimization of the ultrasound-assisted extraction of diarylheptanoid phytoestrogens from Curcuma comosa rhizomes

Abstract

The aim of this study was to establish ultrasound-assisted extraction (UAE) conditions needed to maximize diarylheptanoid recovery from Curcuma comosa Roxb. The performances of a response surface methodology (RSM) and an artificial neural network (ANN) were compared in terms of the predicted yield of the target diarylheptanoids. Key parameters, including the amplitude of the ultrasonic wave (X1), the extraction temperature (X2) and the irradiation time (X3), were studied by using ethanol as extraction solvent. The experimental data were also used to train a multilayer perceptron for creating an ANN model with a 3-7-3 architecture. The optimal UAE conditions predicted by RSM to provide the highest yield of the three diarylheptanoids were 57.32% X1, 64.60 °C X2, and 17.02 min X3. By contrast, the ANN model offered the optimal values for X1, X2 and X3 of 60.27%, 63.92 °C and 16.60 min, respectively. Both statistical tools offered satisfactory results, and superior predictability was observed with the ANN. A vaginal epithelium cornification assay verified the estrogen-like properties of the plant extract. The integrated approach was not only effective at achieving a high recovery of the desired phytoestrogens but could also serve as a useful guide for the large-scale production of diarylheptanoid-rich C. comosa extract.