Distance-AF improves predicted protein structure models by AlphaFold2 with user-specified distance constraints

Area of Science:

• Structural Biology
• Computational Biology
• Biophysics

Background:

• Protein three-dimensional structure is crucial for understanding biological functions.
• Computational protein structure prediction is vital but faces challenges with complex targets like multi-domain proteins or those with multiple conformations.
• Existing methods like AlphaFold2 show promise but can be improved for difficult cases.

Purpose of the Study:

• To enhance the performance of AlphaFold2 for protein structure prediction.
• To develop a computational method that incorporates distance constraints to improve model accuracy.
• To address limitations in predicting structures of proteins with multiple domains or conformations.

Main Methods:

• Developed Distance-AF, a method that integrates distance constraints into the AlphaFold2 framework.
• Evaluated Distance-AF on a test set of 25 protein targets, comparing its performance against AlphaFold2, Rosetta, and AlphaLink.
• Assessed the root mean square deviation (RMSD) between predicted and native structures.

Main Results:

• Distance-AF reduced the average RMSD by 11.75 Å compared to AlphaFold2.
• Distance-AF achieved an average RMSD of 4.22 Å, outperforming Rosetta (6.40 Å) and AlphaLink (14.29 Å).
• Demonstrated successful applications in fitting structures to cryo-electron microscopy data, modeling protein conformations, and generating ensembles for NMR data.

Conclusions:

• Distance-AF significantly improves protein structure prediction accuracy, especially for challenging targets.
• The method offers potential for accelerating structural biology research and drug discovery.
• Distance-AF provides a foundation for integrating experimental and computational approaches to study protein dynamics.