Deep learning techniques have significantly impacted protein structure prediction and protein design

Area of Science:

• Biochemistry
• Computational Biology
• Artificial Intelligence

Background:

• Protein structure prediction and design are inverse problems.
• Progress in these fields has been slow for two decades.
• Deep neural networks offer new computational approaches.

Purpose of the Study:

• To review the impact of deep learning on protein structure prediction and design.
• To highlight advancements in computational protein science.
• To discuss future directions integrating artificial intelligence.

Main Methods:

• Application of deep neural networks for spatial constraint prediction.
• End-to-end model training for protein structure prediction.
• Leveraging neural network models for functional protein design.

Main Results:

• Deep learning significantly improved protein structure prediction accuracy.
• The problem of protein folding at the fold level is largely solved for single-domain proteins.
• Neural network models have enabled advancements in functional protein design.

Conclusions:

• Deep learning integration is a transformative future direction for protein folding and design.
• Continued application of AI will impact protein science.
• This approach promises to accelerate biological discovery and engineering.