Exploring the Role of Topological Descriptors to Predict Physicochemical Properties of Curcumin Compounds using Supervised Machine Learning Algorithms

Abstract

Curcumin and its derivatives have attracted significant attention due to their diverse pharmacological activities and potential applications in medicinal chemistry, drug design, and material science. Accurate prediction of their physicochemical properties is essential for understanding molecular behavior, optimizing bioavailability, and accelerating compound development. In this study, we explore the effectiveness of topological descriptors in predicting key physicochemical properties of curcumin compounds using supervised machine learning algorithms. A dataset comprising structurally diverse curcumin analogs was curated, and multiple graph-theoretic topological descriptors were computed to capture molecular connectivity and structural characteristics. Supervised learning models, including Linear Regression, Decision Tree Regression, Random Forest Regression, Support Vector Regression, and Gradient Boosting methods, were implemented and evaluated for predictive performance. Model assessment was carried out using statistical metrics such as coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE). The results demonstrate that topological descriptors provide significant predictive capability for physicochemical parameters, with ensemble learning techniques outperforming conventional regression approaches in terms of accuracy and robustness. Feature importance analysis further revealed that specific molecular topology indices strongly influence the predicted properties of curcumin derivatives. The findings highlight the potential of combining cheminformatics-based descriptors with machine learning techniques as an efficient and cost-effective framework for physicochemical property prediction and rational molecular design.