Predictive performance and QSPR analysis of SARS-CoV-2 and tuberculosis drugs using distance-based topological descriptors

Topological descriptors from molecular graphs are crucial in quantitative structure-property relationship (QSPR) analysis, linking structure to physicochemical and biological properties. This work studies the detour distance-based index, Detour Eccentric Sum Index (DESI) and the established Eccentric Distance Sum index (EDS) for predictive capabilities. This study explores Quantitative Structure-Property Relationship (QSPR) modeling for SARS-CoV-2 and tuberculosis drugs using novel indices DESI and EDS. Various regression models-logarithmic, exponential, polynomial, and multiple linear-were compared to correlate these topological indices with physicochemical properties. Multicollinearity was assessed through Variance Inflation Factor (VIF) analysis and correlation matrices to ensure model stability. Leave-one-out cross-validation (LOOCV) validated predictions and prevented overfitting in small datasets. For SARS-CoV-2 drugs, second-order DESI models accurately predicted boiling point, enthalpy, and polar surface area; second-order EDS models excelled for polarizability; and logarithmic EDS fitted molar refraction best. In tuberculosis drugs, linear DESI models performed best for molar refraction and polarizability, second-order EDS for molar volume, and combined DESI-EDS multiple linear regression for enthalpy. These distance-based indices, validated through rigorous diagnostics, provide powerful, efficient tools for physicochemical property prediction in computational pharmaceutical design.