Deep-learning-based single-domain and multidomain protein structure prediction with D-I-TASSER
Problem
Accurately predicting atomic-level 3D structures of large multidomain proteins remains challenging for deep-learning methods alone.
Why it matters
Accurate protein structure models are foundational for biological research and medicine.
What the AI did
D-I-TASSER integrates multisource deep-learning potentials with iterative threading fragment-assembly simulations, plus a domain splitting and assembly protocol for large multidomain proteins.
Results
Outperformed AlphaFold2 and AlphaFold3 on single-domain and multidomain proteins in benchmark tests and CASP15; folded 81% of human proteome domains and 73% of full-chain sequences.
Impact
Benchmark and community CASP15 results indicate state-of-the-art performance on tested proteins.
Record
| Venue | Nature Biotechnology |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Wei Hui Zheng, Qiqige Wuyun, Yang Li, Quancheng Liu, Xiaogen Zhou, Chunxiang Peng, Yiheng Zhu, Lydia Freddolino et al. |
| Citations | 99 (OpenAlex, at import time) |
| References | 98 |
| Model or method | D-I-TASSER |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41587-025-02654-4 |
| Open access | hybrid |
| Keywords | D-I-TASSER, protein structure prediction, AlphaFold, CASP15, deep learning, multidomain proteins |
Importance rating
major — millions of lives, an industry or a field
Protein structure prediction is foundational for biological research and medicine; abstract shows human-proteome-wide folding (81% of domains) usable in genome-wide applications.