Topological indices and machine learning techniques for quantitative structure property relationship analysis of antidiabetic drugs

Abstract

Diabetes mellitus is a persistent metabolic disorder marked by disrupted glucose homeostasis, ultimately giving rise to serious complications across multiple organ systems. Although therapeutic options have advanced considerably, existing treatments remain inadequate for fully controlling disease progression, simulating the necessity for innovative strategies in drug discovery and optimization. In contemporary quantitative structure-property relationship (QSPR) studies, topological indices that encode the structural and connectivity characteristics of molecules play a pivotal role. These indices facilitate the prediction of essential physicochemical properties that inform drug likeness and therapeutic potential. In the present work, degree-based, degree-sum-based, and reverse-degree-based topological indices are examined under a bond partitioning framework for a series of antidiabetic drug molecules. The QSPR analysis reveals strong correlations between these indices and key physicochemical attributes, confirming their predictive relevance. Using these descriptors, predictive models are constructed to estimate properties such as boiling point, complexity, heavy atom count, molecular weight, molar refractivity, polarizability, flash point, molar volume, and enthalpy of vaporization. To enhance predictive performance, advanced machine learning algorithms including random forest and XGBoost are employed. These ensemble methods effectively capture complex, nonlinear dependencies between molecular descriptors and physicochemical properties, yielding robust and interpretable models.