Decision Tree QSRR Modeling of Gas Chromatographic Retention Time of Organic Solvents on a DB-۶۲۴ Column

Gas chromatography is routinely used in chemical and petrochemical analysis for separating organic compounds, yet reliable retention-time data are not always available during early method development. In such cases, a structure-based prediction model can help shorten experimental screening and support compound identification. In this study, the gas chromatographic retention time of ۲۴۶ organic solvents was modeled on a DB-۶۲۴ capillary column using a quantitative structure-retention relationship (QSRR) approach. The molecular structures were built and optimized in HyperChem through sequential MM+ and AM۱ calculations, and molecular descriptors were generated with Dragon software. After eliminating constant and non-informative variables, a genetic algorithm was used to select the descriptors most closely related to retention time. The selected seven-descriptor set was then applied in multiple linear regression and decision tree regression models. The GA-MLR model showed strong predictive performance, with R² values of ۰.۹۶۸۴۴, ۰.۹۷۱۳۰, and ۰.۹۶۸۹۸ for the training, test, and full datasets, respectively. The decision tree model, developed as an interpretable rule-based alternative, produced R² values of ۰.۹۷۰۲۶, ۰.۸۳۱۴۰, and ۰.۹۴۴۱۳ for the same datasets. The selected descriptors reflect key structural features associated with solvation tendency, hydrophilicity, molecular topology, and electronic distribution, suggesting that retention on the DB-۶۲۴ phase is governed by a combined contribution of polarity-related and structural effects. The results show that descriptor-based QSRR modeling can provide a reliable and transparent route for estimating retention times of organic solvents and can assist chromatographic method development in separation-based chemical and process analysis

کلیدواژه ها:

Gas chromatography ، QSRR ، Retention time prediction ، organic solvents ، Decision tree regression