Improving protein tertiary structure prediction by deep learning and distance prediction in CASP14
Problem
Predicting protein 3D structure from sequence, especially for template-free (ab initio) targets, remains difficult.
Why it matters
Protein structures are key to understanding function and drug design, but experimental determination is slow and costly.
What the AI did
The MULTICOM system added a deep-learning inter-residue distance predictor, enhanced template-based modeling, and deep-learning quality assessment for CASP14.
Results
Ranked 7th of 146 predictors in tertiary structure prediction and 3rd of 136 in inter-domain prediction at CASP14
Impact
CASP14 blind experiment independently benchmarked the method against 146 predictors.
Record
| Venue | Proteins Structure Function and Bioinformatics |
|---|---|
| Published | 2021 |
| Work type | article |
| Authors | Jian Liu, Tianqi Wu, Zhiye Guo, Jie Hou, Jianlin Jack Cheng |
| Citations | 31 (OpenAlex, at import time) |
| References | 62 |
| Model or method | MULTICOM (deep learning-based distance and quality assessment) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1002/prot.26186 |
| Open access | hybrid |
| Keywords | protein structure, CASP14, deep learning, distance prediction, MULTICOM |
Importance rating
significant — thousands to millions affected
Protein structure prediction is foundational to understanding function and drug design; strong CASP14 ranking among 146 predictors shows significant field-level advance, but no impact scale is quantified.