Surface desorption atmospheric pressure chemical ionization mass spectrometry for identification of lotus seeds freshness based on PCA and BP-ANN

Abstract: In order to realize fast discrimination of lotus seeds freshness, the surface desorption atmospheric pressure chemical ionization mass spectrometry (DAPCI-MS) and principal component analysis (PCA) with back propagation artificial neural network (BP-ANN) were used to distinguish the freshness of lotus seeds produced from 2009 to 2012. Without any sample pretreatments, 60 dried lotus seeds of each year, for a total of 240 individuals were tested and distinguished. The seeds were randomly picked from samples supplied by the Chinese Lotus Seeds Research Academy, which were cultured in the same field in Guangchang County, Jiangxi Province; and were grown with the same standardized method. Each lotus seed was longitudinally sliced to 2 mm for the DAPCI-MS investigation, and tested in the center of the slice with 6 replicates to obtain the averaged results. Experiments were performed using a commercial linear ion trap mass spectrometer (LTQ-XL, Finnigan, San Jose, CA, USA) installed with a homemade DAPCI ion source in negative ion detection mode, and coupled with N2 (0.1 MPa) through a methanol: water (1:1) solution, and a high voltage of 3.0 kV. The mass range m/z was 50-500 and the ion transfer tube temperature was 150℃. The mass spectra were rapidly recorded by DAPCI-MS and the data were processed by PCA. Its main components were selected as the input variables for classification mode of BP-ANN. PCA and BP-ANN were performed by Matlab7.0 software. The results showed that DAPCI-MS was a practical, convenient tool for the detection of matrix bases of lotus seeds. The signal peaks occurred increasingly over the storage time, and the observation correlates well with previous studies of aging cereals such as rice and wheat. The PCA's first 50 components, whose cumulative contribution reached 99.99% and maintained almost all of the original information of the samples, were selected as the input layer of the BP-ANN model which included 50 input layer nodes, 48 hidden layer nodes, and 2 output layer nodes for the crusted and fresh lotus seeds with 30 iterations, and 4 output layer nodes for the different years lotus seeds with 37 iterations; and the learning rate, training time and testing time were 0.01, 10 and 10 respectively. This model successfully distinguished the fresh lotus seeds from the aged samples with the training set accuracies of 92.5% and 100% and testing set accuracies of 95.0% and 91.7%. It also provided a classification of production year of the samples with the training set accuracies of 97.5%, 100%, 97.5%, and 100%, and with the testing set accuracies of 90%, 85%, 85%, and 90%. The whole time of one sample injected 6 times did not exceed 2 min with the full spectrum scan time at 100 ms, and the relative standard deviation (RSD) of the sample was 15.4%. Therefore, the method demonstrates that DAPCI-MS is a fast, convenient and accurate tool for detection of the different quality of lotus seeds, and has a reliable reference value for authentication of food with sufficient sensitivity and high throughput.