Machine Learning-Guided Discovery of Novel Scaffolds for EGFR Inhibition via Classification Models and ScaffoldGVAE
Tạp chí Nghiên cứu Dược và Thông tin Thuốc · 22.06.2026
DOI: 10.59882/1859-364x/402 ↗Kurz gefasst
Fachartikel in der Zeitschrift Tạp chí Nghiên cứu Dược và Thông tin Thuốc, online erschienen am 22.06.2026. Verfasst von Thuan Nguyen Thi und 5 weiteren Autor:innen.
Abstract (Originalsprache)
TThe Epidermal Growth Factor Receptor (EGFR) is a high-priority target in cancer therapy. Reliance on existing chemical space in conventional virtual screening often restricts the discovery of truly novel inhibitors. To address this, a computational workflow combining machine learning with generative artificial intelligence was developed to discover structurally diverse EGFR inhibitors. A curated dataset of 5,276 compounds was used to train classification models, with the best model achieving an overall accuracy of 88.12% and a Potent F1-score of 0.7847. Subsequently, the ScaffoldGVAE generative model was used for scaffold hopping, yielding 12,417 unique and chemically valid molecules. Structural analysis using Principal Component Analysis and Tanimoto similarity (mean Tc = 0.34) demonstrated that the generated molecules expanded the original chemical space. Notably, among the top 1% of candidates predicted as potent (n = 124) ranked by prediction probability, only one compound shared the same Bemis-Murcko scaffold as the training dataset. A dual-filter strategy, combining high prediction probabilities (>0.8) with molecular docking scores superior to those of Lapatinib, identified four final candidates. These molecules retain several interaction patterns and structural features, such as Michael acceptors and halogenated phenyl groups, despite containing modified central scaffolds. Finally, pharmacokinetic, drug-likeness, and medicinal chemistry evaluations showed variable gastrointestinal absorption, limited blood-brain barrier permeability, and moderate synthetic accessibility, with drug-like profiles similar to several EGFR drugs. Overall, this study provides an efficient and reliable strategy for discovering novel EGFR inhibitors
Zitieren
APA
Thi, T. N., Thao, M. N., Thi, D. L., Minh, Q. T., My, H. V. T., & Mai, D. D. T. (2026). Machine Learning-Guided Discovery of Novel Scaffolds for EGFR Inhibition via Classification Models and ScaffoldGVAE. Tạp chí Nghiên cứu Dược và Thông tin Thuốc. https://doi.org/10.59882/1859-364x/402
Vancouver
Thi TN, Thao MN, Thi DL, Minh QT, My HVT, Mai DDT. Machine Learning-Guided Discovery of Novel Scaffolds for EGFR Inhibition via Classification Models and ScaffoldGVAE. Tạp chí Nghiên cứu Dược và Thông tin Thuốc. 2026. doi:10.59882/1859-364x/402
BibTeX
@article{Thi2026,
author = {Thi, Thuan Nguyen and Thao, My-Kristína Nguyen and Thi, Dung Le and Minh, Quang Truong and My, Hue Van Thi and Mai, Dung Do Thi},
title = {Machine Learning-Guided Discovery of Novel Scaffolds for EGFR Inhibition via Classification Models and ScaffoldGVAE},
journal = {Tạp chí Nghiên cứu Dược và Thông tin Thuốc},
year = {2026},
doi = {10.59882/1859-364x/402}
}
EUVIMED