@

초록

Immunosuppressive therapy is essential for long term graft survival after liver transplantation, yet optimizing drug selection and dosing remains challenging due to complex structure–property relationships. Existing Quantitative Structure–Property Relationship [QSPR] studies often rely on limited descriptors or linear models, restricting predictive accuracy and practical relevance. In this work, an integrated graph theoretic and machine learning framework is proposed to model physicochemical properties of clinically approved immunosuppressive drugs, namely tacrolimus, cyclosporine, and prednisone. Degree-based topological indices are analytically investigated, efficiently computed using a reproducible Python framework, and employed as molecular descriptors in supervised learning models. Linear regression is used as a baseline, while Random Forest and Extreme Gradient Boosting capture nonlinear dependencies. Ensemble models achieve excellent predictive performance, with coefficients of determination exceeding 0.99 and low mean absolute error [MAE] and Root mean square error [RMSE] values across key properties such as molecular weight, boiling point, polarity, and enthalpy. The study introduces new analytical relations among topological indices and demonstrates that combining theoretical insights with ensemble learning substantially improves Quantitative Structure–Property Relationship prediction. These results highlight the practical potential of graph-based machine learning models for rapid drug property screening and support their application in personalized immunosuppressive therapy and computational drug design for liver transplantation.

키워드