Development and validation of a predictive statistical model to optimize accelerated solvent extraction of isoflavones from edamame soybean [Glycine max (L.) Merrill]

Three experiments were conducted to develop and validate predictive statistical models for isoflavone extraction from edamame soybean [Glycine max (L.) Merrill]. Soybean line NUTRIVEG Soy6407 was chosen to be the experimental sample material. The first set of extractions tested five factors; solvents, temperatures, pressures, extraction time per cycle and number of extraction cycles. Least squares means for six observed isoflavones (daidzin, glycitin, genistin, malonyl daidzin, malonyl glycitin and malonyl genistin) were used to evaluate trends for the effect of changes in each treatment factor. Pressures had no significant differences for any of the isoflavone extractions. Therefore, 500psi (the lowest pressure setting on the ASE instrument sufficient to maintain solvents in a liquid state when temperatures were raised above the boiling point), was used for all subsequent experiments. A second set of experiments was done to develop predictive regression models.Temperature had a significant effect on glycitin extraction, and extraction time had a significant effect on daidzin extraction. There were differences among solvents for malonyl daidzin, malonyl glycitin, malonyl genistin and total isoflavone content. Temperature and extraction time interacted with solvent, and the interaction was significant for daidzin, genistin and malonyl glycitin extraction. The final models predicted the quantity of extract for daidzin, glycitin, genistin, malonyl daidzin, malonyl glycitin and malonyl genistin respectively at 43.22, 53.57, 44.36, 672.13, 344.14 and 443.30 [mu]g per gram of dry weight edamame. Model validation was accomplished on a third set of extractions by calculating the means of percent absolute differences between true values and the predicted values in order to understand the accuracy of the model. Two organic solvents, 60% methanol and 80% methanol, were determined to have the most consistent extractions.The means of percent absolute differences were approximately 20% or less, and evidence led to the conclusion that the models for these two solvents were reliable.