Deep-learning-based single-domain and multidomain protein structure prediction with D-I-TASSER

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Deep learning has dominated protein structure prediction, questioning traditional simulation methods.
• Existing deep learning models show limitations in predicting complex protein structures.

Purpose of the Study:

• To develop a hybrid approach combining deep learning and physics-based simulations for improved protein structure prediction.
• To assess the performance of the new method against state-of-the-art tools like AlphaFold2 and AlphaFold3.

Main Methods:

• Developed deep-learning-based iterative threading assembly refinement (D-I-TASSER).
• Integrated multisource deep learning potentials with iterative threading fragment assembly.
• Implemented a domain splitting and assembly protocol for large multidomain proteins.

Main Results:

• D-I-TASSER outperformed AlphaFold2 and AlphaFold3 in benchmark tests for both single-domain and multidomain proteins.
• Successfully folded 81% of human protein domains and 73% of full-chain sequences.
• Results are complementary to existing models, offering high accuracy for genome-wide applications.

Conclusions:

• The hybrid D-I-TASSER approach offers a novel integration of deep learning and classical simulations.
• This method provides a promising avenue for high-accuracy protein structure and function prediction.
• D-I-TASSER demonstrates significant potential for large-scale genomic applications.