Artificial intelligencemethods for protein folding and design.

Area of Science:

• Computational biology
• Biotechnology
• Protein engineering

Background:

• Machine learning (ML) has transformed protein structure prediction and design.
• Current ML models leverage evolutionary data for accurate structure prediction but face challenges in capturing protein folding physics.

Purpose of the Study:

• To review current methods for protein folding and inverse folding.
• To examine the potential and limitations of existing protein design tools.
• To provide perspectives on developing advanced models for protein engineering.

Main Methods:

• Review of state-of-the-art ML models including AlphaFold2, RoseTTAFold, and ESMFold for structure prediction.
• Analysis of protein design innovations like RFdiffusion, AF2-design, and sequence optimization methods (ProteinMPNN, ESM-IF).
• Examination of metrics for evaluating protein design and characterization of energy landscapes.

Main Results:

• ML models achieve high accuracy in structure prediction by utilizing evolutionary information.
• Repurposed prediction models have spurred innovations in protein design.
• Dedicated inverse folding methods design sequences based on target structures.

Conclusions:

• Current methods show promise but require further development to fully capture protein folding physics and energy landscapes.
• Advances in characterizing energy landscapes are crucial for accurate structure prediction and designing proteins with specific dynamics.
• Future developments could revolutionize novel protein engineering for biotechnological applications.