Reproduce “Learning Protein-Ligand Binding in Hyperbolic Space” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jianhui Wang, Wenyu Zhu, Bowen Gao, Xin Hong, Ya-Qin Zhang, Wei-Ying Ma, Yanyan Lan
HypSeek is a hyperbolic representation learning framework designed to predict protein-ligand binding by embedding molecular entities into a Lorentz-model hyperbolic space. It demonstrated a 20.7% improvement in EF$_{1\%}$ on DUD-E for virtual screening and achieved a Pearson r of 0.7742 on JACS for affinity ranking, surpassing Euclidean-based deep learning methods.
Key insights
- Hyperbolic geometry, with its exponential volume growth and capacity to preserve hierarchies, provides a more suitable inductive bias than Euclidean geometry for modeling the nuanced and hierarchical relationships in protein-ligand interactions.
- The proposed cone-hierarchy learning, based on radial and angular constraints in hyperbolic space, effectively guides the model to learn fine-grained affinity distinctions, allowing structurally similar but functionally different ligands to be separated.
- A multimodal, three-tower architecture combining ligand, pocket, and protein sequence information, when projected into hyperbolic space, creates more expressive and robust representations for both virtual screening and affinity ranking tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.