DrLungker: A Deep Ensemble Learning Framework for Predicting Anti‐Lung Cancer Compound Activity and Validating Multitarget Potency through WaterMap, DFT, MD Simulations, and MM‐GBSA Analysis

ABSTRACT Lung cancer, with more than 50 approved drugs, is still the deadliest cancer, with 1.80 million annual deaths, necessitating rapid drug development, which can be accelerated by AI‐driven prediction of potent candidates. In this study, we downloaded the lung cancer BioAssay data from ChEMBL and PubChem and filtered at a 5.0 µ m threshold, yielding 4,537 and 8,661 unique active compounds, respectively, and equal inactive molecules are extracted from the big inactive compound library, totalling 26,396 unique, balanced compounds are taken for descriptor computations with QikProp and AlvaDesc software. Mean imputations and standard scaling with PCA for feature sorting, followed by three Deep Learning Models—Residual Neural Network, Feed Forward Neural Network, and Recurrent Neural Network—with an 80:20 split, 50–100 epochs, Adam optimizer, 0.001 learning rate, 32 batch size, early stopping, and ensembled (majority voting, averaging, and stacking) to enhance robustness, accuracy, generalization, stability, and confidence in predicting Activity scores from 1 to 10. A user interface is built to deploy the trained models (h5) for scoring unlabeled compounds (scores 5–10 as highly active), achieving 0.99–1.0 accuracy and F1 scores. The top predicted compound library is docked (HTVS, SP, XP, MM‐GBSA) against ALK, HSP5, KRas, MMP‐8, and tRNA DHDS2, identifying the top three multitargeted hits (PubChem CIDs: 144074375, 440810382, and 48426893) with docking scores from –10.8 to –5.6 kcal/mol and MM‐GBSA energies from –67.7 to –10.4 kcal/mol. Pharmacokinetics and DFT analyses confirmed the drug‐likeness of the compound, while 5 ns WaterMap simulations revealed implicit water roles in interactions, and 100 ns MD simulations showed deviations and fluctuations within 2 Å, with numerous intermolecular interactions. The entire in‐silico study supported and validated the deep learning predictions, identifying the computational potency of compounds against lung cancer proteins—warranting experimental validation.