Combining Docking Pose Rank and Structure with Deep Learning Improves Protein–Ligand Binding Mode Prediction over a Baseline Docking Approach
Problem
Predicting how drug-like ligands bind to target proteins (binding mode and activity) is difficult, and standard docking approaches alone are inaccurate.
Why it matters
Binding mode prediction guides drug design, so errors delay development of new medicines.
What the AI did
A dual-graph convolutional neural network takes docking-generated protein-ligand complex structures, separates ligand-identity bias from true interaction effects, and combines docking pose rank with structure to predict binding mode.
Results
Qualitative improvement over baseline docking claimed in title; metrics not stated in abstract
Impact
Title claims improved binding mode prediction over baseline docking for computational drug discovery.
Record
| Venue | Journal of Chemical Information and Modeling |
|---|---|
| Published | 2020 |
| Work type | article |
| Authors | Joseph A. Morrone, Jeffrey K. Weber, Tien Huynh, Heng Luo, Wendy D. Cornell |
| Citations | 91 (OpenAlex, at import time) |
| References | 52 |
| Model or method | graph-based convolutional neural network (dual-graph architecture) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1021/acs.jcim.9b00927 |
| Open access | green |
| Keywords | deep learning, protein-ligand docking, binding mode prediction, graph neural network, drug discovery |
Importance rating
significant — thousands to millions affected
Binding mode and activity prediction guide drug design; a core computational problem in drug discovery, but the abstract states no scale of affected populations or programs.