Degree-Based Topological Indices and Machine Learning for QSPR Modeling of Arthritis Drugs

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Cheminformatics

Background:

• Accurate prediction of physicochemical properties is crucial for arthritis drug development.
• Traditional quantitative structure-property relationship (QSPR) models often struggle with complex molecular structures.
• Graph-theoretic descriptors offer a simplified representation of molecular complexity.

Purpose of the Study:

• To evaluate the efficacy of degree-based topological indices combined with machine learning for predicting arthritis drug properties.
• To compare the performance of XGBoost, Random Forest, and linear regression models.
• To establish a computationally efficient framework for drug design and virtual screening.

Main Methods:

• Computed nine degree-based topological indices for 50 diverse arthritis drugs.
• Utilized experimental property values including boiling point, melting point, and LogP.
• Developed and compared linear regression, Random Forest, and XGBoost predictive models.

Main Results:

• XGBoost demonstrated superior performance across all tested physicochemical properties, achieving high accuracy (e.g., r²=0.9945 for molar refractivity).
• Random Forest models also showed strong predictive capabilities, outperforming linear regression.
• Linear regression models captured some linear relationships but failed to model nonlinear structure-property dependencies.

Conclusions:

• Combining degree-based topological indices with advanced machine learning, particularly XGBoost, provides highly accurate QSPR predictions for arthritis drugs.
• This approach offers a cost-effective and computationally efficient alternative to traditional methods for drug design and virtual screening.
• The study highlights the potential of simple graph connectivity measures when integrated with nonlinear machine learning algorithms.