Protein structure prediction powered by artificial intelligence: from biochemical foundations to practical...

Area of Science:

• Structural biology
• Computational biology
• Biochemistry

Background:

• Protein three-dimensional structure dictates biological function, making its determination crucial.
• Experimental methods (X-ray crystallography, NMR, cryo-EM) face limitations in throughput and cost.
• Traditional computational methods struggle with novel proteins and complex folding.

Purpose of the Study:

• To review biochemical principles of protein folding.
• To summarize recent AI-driven advances in protein structure prediction.
• To discuss applications and future directions in the field.

Main Methods:

• Deep learning and large-scale protein language models.
• Integration of evolutionary information and geometric constraints.
• End-to-end neural architectures (e.g., AlphaFold3, RoseTTAFold) and single-sequence approaches (e.g., ESMFold).

Main Results:

• AI models achieve near-experimental accuracy in protein structure prediction.
• Single-sequence models provide significant speed and scalability improvements.
• AI advancements are transforming drug discovery, enzyme engineering, and disease research.

Conclusions:

• AI has overcome major hurdles in protein structure prediction.
• Continued advancements promise broader applications and deeper biological insights.
• Addressing current challenges will further refine AI's role in structural biology.