Predicting the Physicochemical Properties of Herbal Glycosides Using Topological Indices and Regression Modelling

Herbal glycosides play a vital role in phytopharmaceutical applications, and accurate determination of their physicochemical properties is essential for drug research. The study aims to develop quantitative structure property relationship (QSPR) models using degree-based topological indices to predict the key physicochemical properties of herbal glycosides. Six herbal glycosides like Aloe-emodin, Chrysophanol, Emodin, Rhein, Hypericin, and Sennoside-A were selected, and degree-based topological indices including the First Zagreb Index M1(G), Second Zagreb index M2(G), Forgotten index F(G), Sum-connectivity index S(G), Yemen index Y(G), and the Degree index D(G) were calculated. Linear regression was applied to predict the Boiling Point (BP), Flash Point (FP), Polarizability (P), Surface Tension (ST), and Molar Volume (MV). The regression models showed very strong predictive power for Boiling Point, Polarizability, and Molar Volume with correlation coefficients (R) above 0.98 and coefficients of determination (R2) greater than 0.90, all statistically significant at p<0.01. Similarly, polarizability predictions using M2(G) and F(G) achieved R2 values of 0.972 and 0.995, respectively. In contrast, predictions of flash point and surface tension were support moderately correlated, with R values around 0.5-0.7 and R2 values less than 0.40, indicating the need for additional descriptors. The findings confirm that degree-based topological indices are effective structure-based descriptors fpr predicting Boiling Point, Polarizability, and Molar Volume of herbal glycosides, thereby offering a rapid and cost-effective computational alternative to experimental methods. However, Flash Point, Surface Tension require complementary descriptors for reliable prediction. This study provides a promising foundation for integrating topological indices into QSPR models in phytopharmaceutical research, with future improvements possible through larger datasets, 3D descriptors, and machine learning approaches.