基于机器学习及外部“探针”策略的HPLC保留时间预测的研究

目的：研究并建立径向基函数神经网络预测化合物色谱峰HPLC保留时间的方法。方法：使用Agilent TC-C₁₈色谱柱（250 mm×4.6 mm，5μm），甲醇-水为流动相等度洗脱，以毛蕊异黄酮葡萄糖苷、芒柄花素、山柰苷、山柰素、槲皮素、刺芒柄花苷、毛蕊异黄酮及异鼠李素8个化合物为研究对象，不同比例流动相洗脱条件下其中7个化合物色谱峰保留时间为特征，与待预测化合物色谱峰保留时间组成训练集各样本，生成并训练神经网络，使得该神经网络具有通过以上7个化合物色谱峰保留时间预测待预测化合物色谱峰保留时间的能力。结果：在使用同一型号色谱柱不同HPLC仪器的情况下，模型的保留时间预测误差不大于0.608 min。结论：本研究创建的方法能够对化合物保留时间进行有效和准确地预测。

Objective:To develop the method based on radial basis function neural network for retention time prediction in HPLC analysis. Methods:The study was performed on an Agilent TC-C₁₈ (250 mm×4.6 mm, 5 μm) column and the elution mobile phase consisted of methanol and water. In the paper, eight compounds, campanulin, formononetin, kaempferitrin, kaempferol, quercetin, ononin, calycosin and isorhamnetin, were used for the study. The retention time of peak of compound was predicted by a model with retention time of seven compounds provided after training set used in the model training process. Results:When the analyses were performed with same column but different HPLC instruments, the prediction errors were below 0.608 min. Conclusion:The method developed in this study can predict retention time in HPLC analysis in an effective and accurate way.

[1] 孙磊, 金红宇, 马双成, 等. 中药标准物质替代测定法技术指导原则[J]. 中国药学杂志, 2015, 50(4):284 SUN L, JIN HY, MA SC, et al. Guideline of substitute reference substance method for evaluation of traditional Chinese medicines[J]. Chin Pharm J, 2015, 50(4):284
[2] 中华人民共和国药典2015年版.一部[S]. 2015:303 ChP 2015. Vol Ⅰ[S]. 2015:303
[3] 孙磊, 金红宇, 逄瑜, 等. 双标多测法Ⅰ-双标线性校正技术用于色谱峰的定性[J]. 药物分析杂志, 2013, 33(8):1424 SUN L, JIN HY, PENG Y, et al. Two reference substance for determination of multiple components (Ⅰ):linear calibration using two reference substances for identification of chromatographic peaks[J]. Chin J Pharm Anal, 2013, 33(8):1424
[4] MILLER TH, MUSENGA A, COWAN DA, et al. Prediction of chromatographic retention time in high-resolution anti-doping screening data using artificial neural networks[J]. Anal Chem, 2013, 85(21):10330
[5] GORYSKI K, BOJKO B, NOWACZYK A, et al. Quantitative structure-retention relationships models for prediction of high performance liquid chromatography retention time of small molecules:endogenous metabolites and banned compounds[J]. Anal Chim Acta, 2013, 797:13
[6] JIAO L, XUE Z, WANG G, et al. QSPR study on the relative retention time of polybrominated diphenyl ethers (PBDEs) by using molecular distance-edge vector index[J]. Chemometr Intell Lab, 2014, 137:91
[7] MITCHELL MT. Machine Learning[M]. Westlake Village:McGraw-Hill Education, 1997:2
[8] YACIN SM, CHAKRAVARTHY VS, MANIVANNAN M. Reconstruction of gastric slow wave from finger photoplethysmographic signal using radial basis function neural network[J]. Med Biol Eng Comput, 2011, 49(11):1241
[9] YUAN LF, DING C, GUO SH, et al. Prediction of the types of ion channel-targeted conotoxins based on radial basis function network[J]. Toxicol In Vitro, 2013, 27(2):852