Accurate prediction of protein structures and interactions using a three-track neural network
Problem
Predicting accurate 3D protein structures and protein-protein complexes from sequence information alone remained a major challenge.
Why it matters
Protein structures underpin drug discovery and disease understanding, but experimental determination is slow and costly.
What the AI did
A three-track neural network successively transforms and integrates 1D sequence, 2D distance map, and 3D coordinate information to predict structures and protein-protein complexes directly from sequence.
Results
Accuracies approaching DeepMind's CASP14 predictions; rapid solution of challenging x-ray crystallography and cryo-EM modeling problems (no numbers stated)
Impact
Enabled rapid, accurate structure and complex prediction and was made available to the research community.
Record
| Venue | Science |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Minkyung Baek, Frank DiMaio, Ivan V. Anishchenko, Justas Dauparas, Sergey Ovchinnikov, Gyu Rie Lee, Jue Wang, Qian Cong et al. |
| Citations | 5,468 (OpenAlex, at import time) |
| References | 82 |
| Model or method | three-track neural network |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1126/science.abj8754 |
| Open access | hybrid |
| Keywords | protein structure prediction, three-track network, CASP14, protein-protein interactions |
Importance rating
major — millions of lives, an industry or a field
Abstract says the network provides insights into proteins of unknown structure, solves challenging crystallography/cryo-EM problems, and is made available to speed biological research; foundational for protein science.