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

The development of thin-layer chromatography (TLC) methods has long relied on empirical trial-and-error, resulting in low efficiency. Although machine learning has been successfully applied to predict retardation factors (R_F), a systematic approach for predicting tailing phenomena remains unexplored, limiting comprehensive evaluation of separation quality. To address this gap, this study aims to develop machine learning models for the simultaneous prediction of R_F values and tailing behavior in TLC. Experimental measurements were conducted to construct datasets covering R_F values and tailing behaviors of various tailing-prone compounds under different developing systems. Using the AutoGluon automated machine learning framework, three predictive models were established: an R_F regression model, a binary tailing classification model, and a minimum additive concentration prediction model. The results demonstrate excellent model performance: the R_F prediction model achieved a coefficient of determination (R²) of 0.888; the tailing classification model attained a balanced accuracy of 0.840; and the accuracy for predicting the minimum additive concentration required to suppress tailing reached 87.5 %. SHAP interpretability analysis further confirmed that the models' decision logic aligns closely with fundamental chromatographic principles, such as polarity interactions. The models developed in this study provide analysts with rapid and reliable decision support, significantly reducing experimental trial-and-error, and mitigating common issues such as tailing, thereby enhancing the efficiency and reliability of TLC method development.