A comparative study of machine learning models in QSPR analysis of antihypertensive drug compounds

Hypertension is one of the most prevalent ailments globally, and it predisposes individuals to heart conditions. To develop improved drugs that reduce hypertension, one should understand the molecular properties that influence the effect of a drug. In this study, we explore the quantitative structure–property relationship models to predict significant properties of antihypertensive drug molecules using machine learning methods. We consider various molecular properties such as molecular weight, heavy atom count, molecular complexity, molar refractivity, polarizability, and molar volume by employing structural properties such as degree-based, degree-sum, and reverse degree topological indices to characterize the molecular structures. The data are analyzed using random forest, extreme gradient boosting, and artificial neural networks techniques in which the developed models are evaluated using leave-one-out as well as 5-fold, and 10-fold cross-validation. We also conduct Y-randomization tests to ensure that the results are not obtained by mere coincidence, and statistical measures are discussed for model performance and consistent predictions.