Real-time monitoring of the column chromatographic process of Phellodendri Chinensis Cortex part I: end-point determination based on near-infrared spectroscopy combined with machine learning

Abstract

As a widely used technique in the area of natural medicine, column chromatography lacks appropriate process monitoring methods. In this work, near infrared spectroscopy was applied to develop a robust calibration model to determine the end-point of the column chromatographic process of Phellodendri Chinensis Cortex. Experimental batches of column chromatographic process (n = 10) with different concentrations of sample solution were designed for the research. The calibration models were established using partial least squares (PLS) regression. Additionally, two novel machine learning algorithms, namely Gaussian process (GP) regression and extreme learning machine (ELM), were utilized to improve the models. Compared to ELM and PLS regression, GP regression exhibited the greatest potential to develop a robust model, which could provide ideal results for the end-point determination of berberine hydrochloride, phellodendrine chloride and the total alkaloids. As a calibration-free method, the moving block standard deviation (MBSD) algorithm could accurately detect the elution end-points of berberine hydrochloride and the total alkaloids in a more convenient way. The current findings suggest that both GP regression and MBSD can be used as efficient tools to determine the end-point of the column chromatographic process of Phellodendri Chinensis Cortex.