Improved prediction of protein-protein interactions using AlphaFold2
Problem
Predicting the structures of heterodimeric protein complexes, fundamental for understanding protein function, was unreliable with existing methods.
Why it matters
Protein interactions underpin most cellular functions and are key targets in drug discovery.
What the AI did
AlphaFold2 was applied to predict heterodimeric complex structures; optimising the multiple sequence alignment substantially improved accuracy over docking alternatives.
Results
44% accurate with default AF2, 59% with optimised MSA vs RoseTTAFold 10%, template docking 35%, traditional docking 22%; acceptable/incorrect discrimination AUC 0.85
Impact
Peer-reviewed with 1,009 citations; improved complex prediction over all compared methods.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2022 |
| Work type | article |
| Authors | Patrick Bryant, Gabriele Pozzati, Arne Elofsson |
| Citations | 1,010 (OpenAlex, at import time) |
| References | 68 |
| Model or method | AlphaFold2 (compared with RoseTTAFold) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-022-28865-w |
| Open access | gold |
| Keywords | AlphaFold2, protein-protein interactions, structure prediction, docking, MSA optimisation |
Importance rating
significant — thousands to millions affected
Abstract states predicting interacting protein structures is fundamental for understanding protein function; optimized AF2 reaches 59% accuracy versus 10-35% for alternatives, with tools freely released to the community.