A Docking-Informed Machine Learning Framework for Prioritizing Tyrosine Kinase Inhibitors

In this study, kinase-targeting compounds were evaluated using molecular docking and machine learning approaches. Tyrosine kinases are key therapeutic targets in cancer due to their role in regulating cellular proliferation and survival. A total of ۴۷۲ compounds were screened against two tyrosine kinase targets (۶LUD and ۴XG۷) and a non-kinase control protein (۱A۳N) to evaluate binding selectivity. Docking validation showed reliable performance with an RMSD of approximately ۱.۰ A. On the other hand, machine learning models, including Random Forest (RF), Support Vector Machine (SVM), Logistic Regression (LR), and Gradient Boosting (GB), were developed for compound prioritization. Among them, LR achieved the best performance (accuracy = ۰.۹۲۶۳; AUC ۰.۹۷). SHAP analysis revealed that physicochemical descriptors and docking scores were the most influential features. As a result, several compounds with favorable binding interactions, including Deucravacitinib and Lomedeucitinib, both non-receptor tyrosine kinase (NRTK) inhibitors, were ranked as top candidates. Potent clinical oncology-relevant kinase inhibitors such as Lorlatinib, Entrectinib, Radotinib, Dasatinib, and Quizartinib were also identified among the top candidates. Overall, the integration of molecular docking and machine learning provided a robust framework for prioritizing kinase-binding compounds, while emphasizing the need for biological validation to improve anticancer specificity and reduce false-positive predictions.

کلیدواژه ها:

Machine learning ، Random Forest (RF) ، Tyrosine kinase inhibitors (TKI) ، Molecular docking ، SHAP analysis