Open Access
Peer-Reviewed
Research Articles
Hierarchical Graph Neural Networks for Zero-Shot Protein-Ligand Complex Binding Affinity Prediction
Integrating Equivariant 3D Coordinate Projections with Multi-Scale Molecular Graph Topology
Abstract
Accurately predicting binding affinities of drug-like small molecules to target proteins in silico is fundamental to modern structure-based therapeutic discovery. Here, we present EquiAffinity-GNN, an SE(3)-equivariant hierarchical graph architecture that unifies atomic-level geometry with pocket-level residue interactions. Benchmarked against the standard PDBbind v2020 refined core set, our framework achieves a Pearson correlation of R = 0.864 and RMSE = 1.12 kcal/mol, significantly outperforming conventional docking force fields and flat GNN baselines.
Keywords
Graph Neural Networks
Molecular Biology
Protein-Ligand Docking
Machine Learning
Computational Chemistry
Declarations & Ethics
Funding:
Funded by the Wellcome Trust (Grant WT21098/Z/22/Z).
Conflicts of Interest:
The authors declare no competing financial or institutional interests.
Peer Review:
Double-blind peer reviewed by international subject specialists.
License:
Creative Commons Attribution 4.0 International (CC BY 4.0).
How to Cite This Article
APA / MLA / BibTeX
Sterling, et al. (2026). Hierarchical Graph Neural Networks for Zero-Shot Protein-Ligand Complex Binding Affinity Prediction. Journal of Advanced Computing and Applied Sciences, 4(2). https://doi.org/10.1234/jcas.2026.040202
Sterling, et al. "Hierarchical Graph Neural Networks for Zero-Shot Protein-Ligand Complex Binding Affinity Prediction." Journal of Advanced Computing and Applied Sciences, vol. 4, no. 2, 2026. https://doi.org/10.1234/jcas.2026.040202
Sterling, et al. "Hierarchical Graph Neural Networks for Zero-Shot Protein-Ligand Complex Binding Affinity Prediction." Journal of Advanced Computing and Applied Sciences 4, no. 2 (2026). https://doi.org/10.1234/jcas.2026.040202