Study On Rapid Identification Models Of Sandalwood And Agarwood,and Their Chemical Components,Based On Proton Nuclear Magnetic Resonance And Machine Learning

Rapid identification model meant to classify samples from different groups by computer in a short time.Because of the rare of sandalwood and agarwood,the phenomenon of mixing genuine and counterfeit in the market needs to improve.This study set up ¹H-NMR fingerprints of 7 groups of Chinese national standards sandalwoods as below: Dalbergia bariensis,Dalbergia cochinchinensis,Dalbergia frutescens,Dalbergia oliveri,Pterocarpus erinaceus,Pterocarpus macrocarpus and Pterocarpus santalinus,and 2 groups of non-Chinese national standards sandalwoods,as following: Gluta sp.and Pterocarpus tinctorius.The machine learning methods were applied in this study was k-Nearest Neighbor?k NN?,Decision Tree?DT?and Support Vevtor Machine?SVM?.Ada Boost algorithm was added to the best performing model to improve accuracy.UPLC/Q-TOF-MS was applied to detect the chemical components of sandalwoods,and ESI-MS fragmentation pattern was summaried.The results showed DT algorithm combine with Ada Boost algorithm can build a robust model of sandalwood.The accuracy of training set and test set of Ada Boost DT model were more than 95%.One hundred and eighty-three components were identified by UPLC/Q-TOF-MS,and 99 components were known-component.Combined with the results of UPLC/Q-TOF-MS,14 components were assigned by ¹H-NMR and ^13C-NMR.The methanol extracts of sandalwoods contains a large number of flavonoids.It could preliminary infer by analyzing the larger contribution rate of variables of model,that the reasons of difference in sandalwood species were as bellow: firstly,the difference of conneting position between C3 and aromatic group;secondly,the difference of conneting position between diferent substitude groups and aromatic groups.A method of building rapid identification model,based on ¹H-NMR and Decision Tree-basis algorithm,was built.To validating this method,natural agarwood,artificial agarwood,Aquilaria sinensis and counterfeits were applied to build a rapid identification model.The results showed the accuracy of training set was more than 90%,and the acuuracy of test set was more than 85%.Comparing with HCA model,PCA model and PLS-DA model,the classification of Decision Tree model was more effective.It could preliminary infer by analyzing the larger contribution rate of variables of model,that the reasons of difference in agarwood and non-agarwood species was the content of 2-?2-phenylethyl?chromone.There are more 2-?2-phenylethyl?chromone in agarwood than in non-agarwood.