Prediction of physicochemical properties of organic compounds using degree-based topological indices and machine learning models

This study presents a quantitative structure-property relationship (QSPR) framework that integrates graph theory with machine learning to predict key physicochemical properties of diverse organic compounds. A data set of 275 structurally diverse organic compounds, including alkanes, alkenes, alkynes, cyclic systems, and aromatic hydrocarbons–was represented as molecular graphs, and their topological features were encoded using two complementary degree-based indices: the Sombor index (SO) and the Modified inverse degree index (ID). These indices were employed to model the octanol-water partition coefficient (LogP), calculated LogP (CLogP), molar refractivity (MR), and critical pressure (CP). Linear regression established baseline correlations, while Random Forest and XGBoost regression models were implemented to capture nonlinear relationships and enhance predictive accuracy. Models were evaluated using a 70%/30% train-test split and five-fold cross-validation. The predicted values show good correlation with Actual values. Machine learning models are outperforming than linear regression. XGBoost has the highest prediction performance for all properties, with a best R² value larger than 0.97. Well-chosen topological indices and machine learning models, provide a highly efficient and effective way to predict the physicochemical properties of organic compounds with diverse scaffolds.