Reversed degree-based graph invariants in machine learning-driven QSPR analysis of antiviral drugs

Résumé

The prediction of physicochemical properties of antiviral drugs is an important challenge in pharmaceutical research due to the high cost and time associated with experimental methods. This study investigates the use of reversed degree-based topological indices (RTIs) combined with machine learning (ML) techniques for predicting physicochemical properties of antiviral compounds. Eighteen antiviral drugs were represented as molecular graphs using canonical SMILES notation, and several RTIs, including Reverse Randić, Reverse Atom-Bond Connectivity, Reverse Zagreb, and Reverse Hyper Zagreb indices, were computed and used as descriptors in Quantitative Structure–Property Relationship (QSPR) models. Support Vector Regression (SVR) and Random Forest Regression (RFR) were employed to evaluate the predictive performance of the descriptors. The novelty of this work lies in integrating reversed degree-based topological descriptors with non-linear ML models for antiviral drugs analysis, providing an alternative to conventional descriptor-based QSPR approaches. The results showed that RFR outperformed SVR for most physicochemical properties, including molecular weight, surface tension, and melting point, with lower Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) values, while SVR performed slightly better for boiling point and polarization. Overall, the findings demonstrate that combining RTIs with ML provides a reliable, efficient, and cost-effective framework for predicting physicochemical properties of antiviral drugs and may contribute to accelerating early-stage drug discovery.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0