A Perspective on the Prospective Use of AI in Protein Structure Prediction.

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• AlphaFold2 (AF2) and RoseTTaFold (RF) are leading AI tools for protein structure prediction.
• Their integration with experimental methods is transforming structural biology research.

Purpose of the Study:

• To evaluate the impact and limitations of AF2 and RF across various protein types.
• To identify areas for improvement in deep learning-based protein structure prediction.

Main Methods:

• Analysis of AF2 and RF performance in predicting structures of membrane proteins, intrinsically disordered proteins (IDPs), and oligomers.
• Review of integration strategies with experimental techniques like X-ray crystallography, mass spectrometry, and NMR.
• Exploration of complementary computational methods such as molecular dynamics (MD) simulations.

Main Results:

• AF2 and RF show high reliability but struggle with membrane protein conformational ensembles and IDP dynamics.
• These tools enhance experimental pipelines, aiding phase problem resolution and structure determination.
• Promising results for oligomeric models, outperforming traditional docking, with AlphaFold-Multimer showing enhanced performance.

Conclusions:

• Deep learning models show significant potential but require further refinement for specific protein classes.
• Integrating experimental data and advanced simulations is key for future advancements.
• A "wish list" for model improvement includes experimental constraints and consideration of binding partners and PTMs.