In this work, different chemometric calibration models were developed and validated for the purpose of determining of ternary mixture of oral antidiabetic drugs; vildagliptin (VDG), saxagliptin (SAX) and sitagliptin phosphate (STG). The used models were Partial least squares (PLS) and Artificial Neural Networks (ANN). However, on these various models the impact of genetic algorithm (GA) as a form of variable selection was also investigated. The UV spectral data was used as basis in the quantitative study of the drugs analyzed in bulk and product formulations. The concentration range of the calibration curves of VDG, SAX and STG were 10–22 μg mL⁻¹, 24–40 μg mL⁻¹ and 82–130 μg mL⁻¹, respectively. The calibration set included nineteen mixtures and the others six were used as a validation set to test the predictability of the developed multivariate models. The validation parameters of the evaluated methods were statistically determined. For the analysis of drugs studied in laboratory-prepared mixtures and their dosage forms, PLS-1, GA-PLS-1, ANN, and GA-ENN were successfully employed. The results obtained by the developed methods were compared to those given by a reported method and there were no statistically significant differences regarding accuracy and precision.

在这项工作中，为了确定口服降糖药的三元混合物，开发并验证了不同的化学计量学校准模型；维格列汀（VDG），沙格列汀（SAX）和西他列汀磷酸酯（STG）。使用的模型是偏最小二乘（PLS）和人工神经网络（ANN）。但是，在这些各种模型上，也研究了遗传算法（GA）作为变量选择形式的影响。紫外线光谱数据被用作对散装和产品配方中所分析药物进行定量研究的基础。VDG，SAX和STG的校准曲线的浓度范围分别为10-22微克毫升^-1，24-40微克毫升^-1和82-130微克毫升^-1， 分别。校准集包括19种混合物，其余6种用作验证集，以测试开发的多元模型的可预测性。统计评估方法的验证参数。为了分析实验室制备的混合物及其剂型中研究的药物，成功使用了PLS-1，GA-PLS-1，ANN和GA-ENN。通过开发的方法获得的结果与通过报告的方法获得的结果进行了比较，在准确性和精确度方面没有统计学上的显着差异。