A Computational Framework for Drug Property Prediction Using K Banhatti Topological Descriptors and Hybrid Machine Learning Modeling Approaches
Quantitative Structure-Property Relationship modeling is a valuable resource to predict drug-related physicochemical properties and supports rational drug development.Knee disorders are the most common musculoskeletal diseases which affect the knee joints, including arthritis and various inflammatory diseases.These disorders require effective pharmacological intervention to address their high prevalence and respective stability.In this study, distinct K-Banhatti indices were systematically evaluated for knee disorder drugs used to treat knee disorders.Their relationships with different physicochemical properties were studied to determine the property that could be predicted most accurately and the K-Banhatti index that was most suitable for predicting it.Regression models were applied and validated in Python using a dataset to develop a quantitative descriptor-property relationship.Among curvilinear regression models, cubic regression based on the product connectivity Banhatti index achieved the best performance with the lowest Root Mean Square Error (RMSE) = 1.707,Mean Absolute Error (MAE) = 1.347 and the largest R 2 = 0.963 value.Additionally, machine learning (ML) models [Artificial Neural Networks (ANN), Random Forest] were used to enhance the predictive accuracy of the models.All of the nonlinear models developed in this study significantly outperformed compared to the linear models across all evaluation metrics.Specifically, the ANN model has better accuracy with RMSE = 1.182,R 2 = 0.982, and MAE = 0.885 and a smaller error distribution in comparison to other models.The models effectiveness were also demonstrated by a bar chart plot.It was observed that polarizability was identified as the best predicted property, based on its strong dependence on molecular descriptors.K-Banhatti indices combined with ensembled ML offer an effective approach for predicting physicochemical properties of musculoskeletal drugs.This computational framework enables and prioritizes promising drug candidates, minimizing unnecessary synthesis and experimental screening.
Authors
- Hafiz Muhammad Atif
- Hifza .
Publication Details
- Journal
- Cureus Journal of Computer Science.
- Published
- 2026-11-14
- DOI
- https://doi.org/10.7759/s44389-026-00367-6
- Primary Topic
- Graph theory and applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00