Multi-domain and complex protein structure prediction using inter-domain interactions from deep learning
Problem
Predicting 3D structures of multi-domain proteins and protein complexes, where AlphaFold2 still underperforms.
Why it matters
Accurate domain-domain interaction modeling is key to understanding protein function and designing structure-based drugs.
What the AI did
A deep learning network infers inter-domain interactions, and a population-based evolutionary algorithm assembles domains into multi-domain proteins and complexes.
Results
On 219 multi-domain proteins, inter-domain distance precision 22.7% higher than AlphaFold2; 13.1% accuracy improvement on 164 low-accuracy structures.
Impact
Peer-reviewed improvement over AlphaFold2 on a hard sub-problem, but complex modeling remains a challenge per abstract.
Record
| Venue | Communications Biology |
|---|---|
| Published | 2023 |
| Work type | article |
| Authors | Yuhao Xia, Kailong Zhao, Dong Liu, Xiaogen Zhou, Guijun Zhang |
| Citations | 47 (OpenAlex, at import time) |
| References | 55 |
| Model or method | DeepAssembly |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s42003-023-05610-7 |
| Open access | gold |
| Keywords | protein structure prediction, AlphaFold2, deep learning, multi-domain proteins, protein complexes |
Importance rating
significant — thousands to millions affected
Protein structure prediction underpins a foundational scientific field and structure-based drug design; abstract frames domain-domain interactions as key to both, with demonstrated gains over AlphaFold2.