Drug Repositioning for Childhood Acute Lymphoblastic Leukemia Using an Explainable Regularized Bi-LSTM Ensemble and Molecular Docking Validation

Childhood acute lymphoblastic leukemia is a fast-progressing blood cancer marked by uncontrolled proliferation of immature lymphoid cells. Although modern treatment options such as chemotherapy and immunotherapy have increased survival rates, recurrence and drug resistance remain critical challenges. In this study, we present a novel explainable deep learning framework for predicting protein-ligand interactions and identifying repositionable drugs for this disease. Among the five tested architectures, a regularized Bi-LSTM + Bi-LSTM ensemble achieved the best performance, with a test mean absolute error (MAE) of 0.041 and a test loss of 0.210, outperforming other models. Explainability techniques such as Integrated Gradients and LIME revealed key sequence features in both SMILES and protein data contributing to predictions. Molecular docking validation using PandaDock confirmed strong binding affinities for selected ligands, such as Copanlisib (-207.11 kcal/mol) and Dutasteride (-171.17 kcal/mol), in alignment with model outputs. Literature-based evidence further supported the therapeutic relevance of identified compounds. This study demonstrates that explainable deep learning combined with molecular docking provides a robust strategy for drug repositioning in childhood acute lymphoblastic leukemia.