Five‐descriptor model to predict the chromatographic sequence of natural compounds

Despite the recent introduction of mass detection techniques, ultraviolet detection is still widely applied in the field of the chromatographic analysis of natural medicines. Here, a neural network cascade model consisting of nine small artificial neural network units was innovatively developed to predict the chromatographic sequence of natural compounds by integrating five molecular descriptors as the input. A total of 117 compounds of known structure were collected for model building. The order of appearance of each compound was determined in gradient chromatography. Strong linear correlation was found between the predicted and actual chromatographic position orders (Spearman's rho = 0.883, p < 0.0001). Application of the model to the external validation set of nine natural compounds was shown to dramatically increase the prediction accuracy of the real chromatographic order of multiple compounds. A case study shows that chromatographic sequence prediction based on a neural network cascade facilitated compound identification in the chromatographic fingerprint of Radix Salvia miltiorrhiza. For natural medicines of known compound composition, our method provides a feasible means for identifying the constituents of interest when only ultraviolet detection is available.