Recently, a collaborative team led by Professor Mo Fanyang from our institute and Professor Zhang Dongxiao from East China University of Technology has, for the first time, clearly revealed the quantitative relationship between Thin-Layer Chromatography (TLC) and Column Chromatography (CC) by combining statistical and machine learning methods. This study proposes a knowledge discovery technique, establishes an interpretable formula, and transforms "Chemist’s experience" into "AI experience", providing theoretical support for determining and optimizing experimental conditions for chromatographic separation. The related achievements were published inNature Communications.

Figure 1: The related work was published in Nature Communications on January 19

Both thin-layer chromatography and column chromatography are analytical separation methods based on chromatographic principles, widely used in synthetic chemistry laboratories. When conducting column chromatography experiments, it is often necessary to first perform thin-layer chromatography analysis and use the retention factor (RF value) to evaluate the relative polarity between components in the mixture and the mobile phase. In actual operations, the composition and proportion of the mobile phase are usually adjusted as needed to ensure that the RF value of the target compound is approximately between 0.2 and 0.3. This experience-based method is highly effective, but the underlying principles have not been fully clarified, resulting in a phenomenon of "knowing the result but not the reason". This hinders researchers' in-depth understanding of the chemical essence of chromatographic separation.

Figure 2: Quantification of the relationship between expert experience in chromatographic separation and data-driven approaches

To address this scientific question, the research team adopted a data-centric perspective, attempting to directly identify the potential coupling relationship between thin-layer chromatography and column chromatography from a large amount of experimental data and express it in the form of a concise equation. To this end, the research team developed an automated column chromatography platform, systematically collected column chromatography retention volumes of 192 compounds under different experimental conditions, and obtained a total of 5984 data entries. Based on this, the research team analyzed the relationship between the retention factor (RF value) of thin-layer chromatography and the retention volume of column chromatography using machine learning methods, and derived an explicit mathematical formula through symbolic regression.

Figure 3: Formula identification and prediction effect of the relationship between column chromatography retention time and thin-layer chromatography RF value