A multivariable signal conversion strategy for quality assessment of Gastrodiae rhizome based on convolutional neural network

Abstract

Chromatographic fingerprinting is a time-consuming analytical method used to evaluate the overall chemical properties of natural products. While NIR spectroscopy is a rapid analytical tool, it is typically used to predict only a few quality markers. In this study, we proposed a multivariable signal conversion strategy that can rapidly assess the quality of natural products by reproducing HPLC signals from NIR spectra. The strategy converted the NIR spectroscopic fingerprints to chromatographic fingerprints using PLS and CNN, respectively, and evaluated the performances of conversion using Pearson correlation coefficient and nearness index. Hierarchical clustering analysis and correlation analysis were used to evaluate the similarity between the converted and real fingerprints to reflect the goodness of the conversion. The CNN models showed comparable and satisfactory results to PLS models in predicting quality markers and converting fingerprints. Moreover, CNN models did not require spectral pretreatment, which makes them more advantageous than PLS models. The proposed strategy combines the strengths of NIR spectroscopy, such as information-rich and high-speed analysis, and the benefits of chromatographic fingerprinting, which provides a better understanding of overall chemical properties. Therefore, it has great potential for application in the field of natural product analysis.