Kirsanov Dmitry, Panchuk Vitaly, Agafonova-Moroz Marina, Khaydukova Maria, Lumpov Alexander, Semenov Valentin, Legin Andrey
Institute of Chemistry, St. Petersburg State University, St. Petersburg, Russia.
Analyst. 2014 Sep 7;139(17):4303-9. doi: 10.1039/c4an00227j.
The experimental design of mixtures for multivariate calibration is introduced. The idea of this design is based on uniform distribution of experimental points in a concentration hypercube. Unlike the already reported uniform designs this one is pretty simple and not computationally demanding. The suggested approach does not employ the concept of fixed "levels" and allows for designs with any number of experimental mixtures and any number of components depending on "time and money" considerations for each particular calibration experiment. The performance of the design is assessed with a UV-Vis spectroscopic experiment for simultaneous quantification of four inorganic components in complex mixtures. The performance of the PLS regression models derived from the design is compared with that of cyclic permutation and Kennard-Stone designs. The suggested approach allows for comparable or higher prediction accuracy with the lower number of experimental points.
介绍了用于多元校准的混合物实验设计。该设计的理念基于实验点在浓度超立方体内的均匀分布。与已报道的均匀设计不同，此设计非常简单且计算要求不高。所建议的方法不采用固定“水平”的概念，并允许根据每个特定校准实验的“时间和成本”考虑，设计具有任意数量实验混合物和任意数量组分的方案。通过紫外可见光谱实验评估该设计的性能，以同时定量复杂混合物中的四种无机成分。将从该设计导出的偏最小二乘回归模型的性能与循环排列设计和肯纳德-斯通设计的性能进行比较。所建议的方法能够以较少的实验点数实现相当或更高的预测精度。
