Machine-learning prediction of retardation factor and tailing propensity in thin-layer chromatography

Area of Science:

• Analytical Chemistry
• Chromatography
• Machine Learning

Background:

• Thin-layer chromatography (TLC) method development traditionally relies on empirical trial-and-error, leading to inefficiencies.
• While machine learning predicts retardation factors (R_F), predicting tailing phenomena systematically remains an underexplored area, hindering comprehensive separation quality assessment.

Purpose of the Study:

• To develop machine learning models for the simultaneous prediction of R_F values and tailing behavior in TLC.
• To address the limitations of empirical methods and improve TLC analysis efficiency.

Main Methods:

• Experimental measurements were performed to create datasets for R_F values and tailing behaviors of various compounds under different developing systems.
• The AutoGluon automated machine learning framework was utilized to build three predictive models: R_F regression, tailing classification, and minimum additive concentration prediction.

Main Results:

• The R_F prediction model achieved a coefficient of determination (R²) of 0.888.
• The tailing classification model demonstrated a balanced accuracy of 0.840.
• The minimum additive concentration prediction model reached an accuracy of 87.5% in predicting concentrations to suppress tailing.
• SHAP analysis confirmed model logic aligns with chromatographic principles like polarity interactions.

Conclusions:

• The developed machine learning models offer rapid and reliable decision support for TLC analysts.
• These models significantly reduce experimental trial-and-error and mitigate common issues like tailing.
• The study enhances the overall efficiency and reliability of thin-layer chromatography method development.