하민우

AI-Driven Advances in Protein–Ligand Docking

Journal of The Korea Society of Computer and Information

-

1598-849X

30(11)

KCI

In this paper, we review AI-based protein-ligand docking, highlighting its evolution from traditional
techniques to modern approaches utilizing deep learning and diffusion models. Computer-Aided Drug Design
(CADD) accelerates discovery, with docking central to pose prediction and virtual screening. Conventional
workflows split pose sampling (GA/MC/MD) and scoring (force-field/empirical/knowledge-based), but suffer
from receptor rigidity and binding-site dependence. AI mitigates these via CNN/GNN rescoring, learned site
prediction, and generative models. Diffusion docking (DiffDock) denoises translation/rotation/torsions, boosting
top-k accuracy. An EGFR-Gefitinib study (1M17) contrasts AutoDock Vina, GNINA, and DiffDock, motivating
hybrid AI-physics pipelines.

김재현, 권찬혁, 이수미*, 하민우*

28 November 2025

10.9708/jksci.2025.30.11.223