Integrating machine learning with QSPR for property prediction of anti-vertigo drugs

Vertigo is a debilitating vestibular disorder affecting approximately 10-30% of the global population, yet no dedicated computational chemistry study has systematically characterised the properties of its pharmacological agents using machine learning. In this study, we present the first Quantitative Structure-Property Relationship (QSPR) analysis of 47 clinically established anti-vertigo drugs using nine Neighborhood degree-based topological indices, combined with three machine learning algorithms such as Ridge Regression, Random Forest (RF) and Gradient Boosting Regression (GBM). The topological indices were correlated with five key properties like Molecular Weight, Complexity, Molar Refractivity, Polarizability, and Molar Volume, using Pearson correlation analysis. All 47 index-property correlations were statistically significant at p < 0.0001. Molecular weight (MW) had the highest correlation with ND₅ index (R = 0.9684), while molar refractivity (MR) and polarizability had their strongest correlations with NH index (R = 0.9672, R = 0.9673), respectively. Complexity had the highest correlation with the forgotten index (R=0.9477), where Molar Volume had highest correlation with the index NH (R= 0.9152). Within the machine-learning model analysis, GBM achieved the most accurate predictions for MW(R²=0.9757), complexity(R²=0.7975), polarizability(R²=0.9652), and molar volume (R²=0.9456), while RF was superior to other models in terms of the accuracy of the prediction of MR (R²=0.9648). Additionally, LOO-CV was used to evaluate model validity and returned Q² above the minimum level for each of the Molecular Weight (Q²=0.9283), Polarizability (Q²=0.9192), Molar Refractivity (Q²=0.9189), Complexity (Q²=0.8753), and Molar Volume (Q²=0.7633) models. This study provides a validated computational framework for rapid predictions of physicochemical properties and drug discovery for future vestibular drugs.