Real-time concentration prediction in column chromatography purification using NIR optical sensing and evolutionary...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Natural Product Chemistry

Background:

• Ginsenosides are key bioactive compounds in American ginseng, necessitating precise quality control during purification.
• Traditional methods for monitoring ginsenosides are slow and unsuitable for real-time process adjustments.
• Low-abundance ginsenosides pose a challenge for accurate quantification in complex mixtures.

Purpose of the Study:

• To develop a rapid, non-destructive method for real-time monitoring of ginsenoside concentrations during column chromatography.
• To integrate near-infrared (NIR) spectroscopy with an advanced deep learning model for dynamic prediction.
• To enhance the efficiency and accuracy of natural product purification processes.

Main Methods:

• Utilized a near-infrared (NIR) optical sensing system for high-speed spectral data acquisition.
• Developed and applied an Evolutionary Attention-based Long Short-Term Memory (EA-LSTM) model for dynamic ginsenoside concentration prediction.
• Trained and validated the model using offline spectra from eight elution batches, cross-referenced with Ultra-Performance Liquid Chromatography (UPLC) data.
• Simulated high-frequency sampling to replicate industrial application conditions.

Main Results:

• The NIR-EA-LSTM system achieved high accuracy in predicting ginsenoside concentrations (R² = 0.9841, RPD = 7.94) under offline conditions.
• The system demonstrated robust performance under dynamic conditions, with excellent accuracy (R² = 0.9694, RPD = 5.72).
• Achieved a rapid response time of 0.34 ms, enabling near real-time monitoring.

Conclusions:

• The developed NIR-EA-LSTM system offers a fast, non-destructive, and accurate solution for monitoring ginsenosides during chromatography.
• This technology has significant potential for real-time, in-line quality control and intelligent process optimization in natural product purification.
• The approach addresses the limitations of conventional assays for dynamic process monitoring, especially for low-abundance compounds.