Phytosterol glycosides (PGs), comprising both acylated steryl glycosides (ASGs) and steryl glycosides (SGs), are active ingredients with benefits for human use. Here, we aimed to optimize the silica-gel adsorption technique for the extraction of PGs from soybean lecithin powder, which contains 5 to 10% of these glycolipids. Both response surface methodology (RSM) and artificial neural networks (ANNs) were applied to optimize the PG extraction parameters (X1 = silica-gel dosage, X2 = adsorption temperature, and X3 = lecithin concentration) for high-purity phospholipid and PG production, and their prediction and optimization accuracies were compared. Although both models fitted well with the experimental data, the ANN model demonstrated better accuracy for predicting and optimizing the conditions using four interrelated dependent variables (Y1 = phospholipid yield, Y2 = ASG recovery, Y3 = SG recovery, and Y4 = PG purity) and had a higher coefficient of determination and lower root mean square error and absolute average deviation. After digitally setting the percentages of the four dependent variables for phospholipid and PG production, the ANN-optimized phospholipid product (Y1 = 88.07%, Y2 = 98.89%, Y3 = 100%, and Y4 = 49.03%) was acquired at X1 = 3.54 g/g, X2 = 26 °C, and X3 = 43 mg/mL, whereas the PG product (Y1 = 83.83%, Y2 = 97.64%, Y3 = 100%, and Y4 = 59.21%) was obtained at X1 = 2.00 g/g, X2 = 28.38 °C, and X3 = 41 mg/mL. In conclusion, the ANN method was better than RSM for the optimization of the silica-gel adsorption technique for PG extraction from soybean lecithin powder. PRACTICAL APPLICATION: This paper lays a theoretical foundation for the optimization of the industrial production of phytosterol glycosides and the comprehensive utilization of lecithin resources.

中文翻译：

---

响应面法和人工神经网络模型优化硅胶吸附技术从大豆卵磷脂粉中提取植物甾醇苷

植物甾醇糖苷 (PG) 包括酰化甾醇糖苷 (ASG) 和甾醇糖苷 (SG)，是对人类有益的活性成分。在这里，我们旨在优化从大豆卵磷脂粉末中提取 PG 的硅胶吸附技术，其中含有 5% 到 10% 的这些糖脂。应用响应面方法 (RSM) 和人工神经网络 (ANN) 来优化高纯度磷脂和 PG 生产的 PG 提取参数（X1 = 硅胶用量，X2 = 吸附温度，X3 = 卵磷脂浓度），并比较了它们的预测和优化精度。尽管两种模型都与实验数据吻合得很好，ANN 模型使用四个相互关联的因变量（Y1 = 磷脂产量，Y2 = ASG 回收率，Y3 = SG 回收率，Y4 = PG 纯度）展示了预测和优化条件的更高准确性，并且具有更高的决定系数和更低的均值平方误差和绝对平均偏差。在对磷脂和 PG 生产的四个因变量的百分比进行数字设置后，在 X1 = 3.54 处获得了 ANN 优化的磷脂产物（Y1 = 88.07%、Y2 = 98.89%、Y​​3 = 100% 和 Y4 = 49.03%） g/g、X2 = 26 °C 和 X3 = 43 mg/mL，而 PG 产物（Y1 = 83.83%、Y2 = 97.64%、Y3 = 100% 和 Y4 = 59.21%）是在 X1 = 2.00 时获得的g/g，X2 = 28.38 °C，X3 = 41 mg/mL。综上所述，ANN 方法在优化硅胶吸附技术从大豆卵磷脂粉中提取 PG 方面优于 RSM。实际应用：本文为植物甾醇苷的工业化生产优化和卵磷脂资源的综合利用奠定了理论基础。