Application of total quadratic indices of molecular pseudographs: a study on Nonsteroidal Anti-Inflammatory Drugs (NSAIDs) to predict physicochemical properties using weighting schemes

Abstract This study investigates the relevance of physicochemical properties of 34 NSAIDs to their chemical characteristics using total quadratic indices q k ( x ) modeled as pseudographs with atomic number and atomic radius weighting schemes, compared against traditional indices modeled using unweighted graphs. The objective was to assess the predictive capabilities of these indices in modeling drug properties. Multiple Ordinary Least Squares (OLS) regression revealed significant multicollinearity in the models. Principal Component Analysis (PCA) was employed to address this issue, leading to a reduction in multicollinearity, although it was accompanied by a slight decrease in model performance metrics such as R 2 . Despite this, quadratic indices demonstrated competitive performance when compared to traditional indices. Notably, the quadratic indices showed higher relevance in predicting the physicochemical properties complexity (C), molecular weight (MW), refractivity (RV) and polarizability (P) compared to traditional indices. The model validated by parameters like coefficient of determination ( R 2 ), p - value, standard error (S.E), F-statistic and Durbin-Watson (DW) statistics. These results underscore the potential of quadratic indices in enhancing predictive modeling of drug properties, offering improved insights compared to traditional topological indices.