FTIR spectroscopy combined with DD-SIMCA and Two-dimensional correlation spectroscopy (2DCoS) for judgement about walnut origins

【Objective】This work was to develop rapid and facile methods for the determination and classification of the walnut geographical origin based on kernel and shell samples by Fourier transform infrared spectroscopy (FTIR).【Methods】A total of 120 in-shelled walnut samples were collected from 4 areas of China, including Xinjiang, Guizhou, Sichuan and Yunnan. The walnuts were split into 120 kernel samples and 80 shell samples. After freeze- drying and oven- drying respectively, samples were numbered to maintain traceability throughout the process and were stored in a sealed bag for use. These samples were investigated under solid-state conditions in KBr pellets utilizing FTIR spectra in the range of 4000-400 cm-1 . Several mathematical pre-treatment methods including baseline correction, SavitskyGolay (S.G.) smoothing, standard normal variate (SNV) and multiplicative scatter correction (MSC) were tested in the original spectral matrix. Then, the principal component analysis (PCA) was performed on the whole dataset for qualitative analysis of the spectra. Subsequently, four different classification models were applied to the FTIR data. Partial least squares discriminant analysis (PLS-DA), and support vector machine (SVM) were used to identify and categorize the spectra data from walnut kernel samples and shell samples. The class modeling techniques for soft independent modeling by class analogy (SIMCA) and data-driven soft independent modeling of class analogy (DD-SIMCA) were evaluated by using walnut kernel samples data. The DD-SIMCA was chosen because it was a powerful class-mod-eling technique to test whether a sample was a target category or not, to assign an unknown sample to a class (target category) or not (outlier). The DD-SIMCA model was established on the PCA decomposition of each class separately. The efficiency of such chemometrics models was evaluated in terms of correct- prediction percent, sensitivity and specificity. Besides, the synchronous two- dimensional correlation spectrum (2DCoS) maps based on temperature perturbations of the walnut kernel were computed and analyzed in 1800~700 cm-1 wavenumber regions.【Results】The results showed that both walnut kernel and shell samples from different origins performed high similarities of the FTIR spectral profiles, especially in the position of the absorption peak. The resemblance of FTIR spectra was a result of its similar chemical composition. Separately, 14 characteristic absorption peaks were obtained from the walnut kernel average spectrum. As indicated by their FTIR spectra, their differences occurred in the region of the characteristic peaks 1750~1430 cm-1 and the fingerprint region (1240-600 cm-1 ). In particular, using the variable importance in the projection (VIP) technique analyzed the peak value and provided that the main differences occurred in characterizing the absorption peaks of proteins (1649 cm- 1 and 1539 cm- 1 , VIP>1.0). It was difficult to classify them successfully only based on FTIR spectra. Therefore, a chemometric analysis must be conducted to detect these potential distinctions. Based on PCA and SIMCA classification performance comparison among these pretreatment methods, S.G. smoothing + MSC was superior to other methods. The samples can be clustered into four non-overlapping categories according to their collection areas. Also, the SIMCA results indicated the highest specificity and sensitivity, with a performance above 67%. In this work, all classification models used this method to pre-process spectra data. The recognition rate of the walnut kernel and shell prediction set was 73% (R2 cv=0.78) and 100% (R2 cv=0.95) by PLS-DA. Using the SVM pattern, 97% and 100% of prediction set samples were correctly classified in pre-treated spectra. The SIMCA model and DD-SIMCA model were selected for walnut kernel pre-treated spectra further analysis. The prediction ability of SIMCA resulted in 87% accuracy. The accurate and stable model (100% of sensitivity and 100% of specificity) was obtained with walnut kernel spectra data for the DD-SIMCA. In contrast to the four models, the DD-SIMCA model was a suitable one for walnut origin category classification. The synchronous 2DCoS constructed from the FTIR changes in the temperature gradient range of 15-55 ℃ revealed that an autocorrelation spectrum intensity in the region of 1800-1600 cm-1 was obviously different. Manifestation in (1745 1745) cm-1 and (1650 1650) cm- 1 showed a strong autocorrelation peak and the cross peak at (1655 1745) cm- 1 and (1540 1745) cm- 1 . Bands due to the combination of carboxylic acid (ester) carbonyl C = O stretching, carbonyl (amide I) C = O stretching, and protein N-H bending vibrations were varied. The 2DCoS acted out greater diversity than the one-dimensional spectrum, which can be classified.【Conclusion】Adopting a suitable spectral pre- processing (S.G. smoothing + MSC) method can significantly improve the classification and discriminatory correct- prediction percent of the model. Both walnut shells and kernels can be used to identify the origin of walnuts. The FTIR combined with DD-SIMCA and 2DCoS was potentially an ideal tool for rapidly and sensitively identifying the geographical origin of walnut samples.