Structure-Aware Multimodal Deep Learning for Drug–Protein Interaction Prediction
Problem
Predicting which drugs interact with which proteins is crucial for drug discovery, but existing methods use biased datasets and weak protein representations.
Why it matters
Better drug-protein interaction prediction speeds up drug discovery and reduces costly experimental screening.
What the AI did
Built STAMP-DPI, a structure-aware multimodal deep learning model using a graph neural network from protein sequence with predicted contact maps, trained on a new industry-scale benchmark dataset (GalaxyDB).
Results
not stated in source
Impact
Provides a curated industry-scale benchmark dataset and a more robust benchmark study for DPI prediction.
Record
| Venue | Journal of Chemical Information and Modeling |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Penglei Wang, Shuangjia Zheng, Yize Jiang, Chengtao Li, Junhong Liu, Chang Wen, Atanas Patronov, Dahong Qian et al. |
| Citations | 75 (OpenAlex, at import time) |
| References | 37 |
| Model or method | STAMP-DPI (structure-aware graph neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1021/acs.jcim.2c00060 |
| Open access | green |
| Keywords | drug-protein interaction, deep learning, graph neural network, benchmark dataset, virtual screening |
Importance rating
significant — thousands to millions affected
Drug-protein interaction prediction is called crucial to drug discovery, a foundational industry; model plus new industry-scale benchmark could reduce costly screening, though no scale is quantified.