NMR Spectroscopy for the Validation of AlphaFold2 Structures

Area of Science:

• Biochemistry
• Structural Biology
• Computational Biology

Background:

• AlphaFold revolutionized protein structure prediction using artificial intelligence (AI).
• Combining AI predictions with experimental data offers potential for higher accuracy and reduced experimental effort.
• Nuclear Magnetic Resonance (NMR) spectroscopy provides valuable experimental structural information.

Purpose of the Study:

• To develop and test hybrid computational-experimental methods for protein structure determination.
• To assess the accuracy of AI-predicted protein structures using experimental NMR data.
• To establish a framework for integrating AlphaFold predictions with NMR spectra.

Main Methods:

• Developed heuristics to compare N-edited NOESY spectra with AlphaFold predicted structures.
• Compiled a dataset linking the Biological Magnetic Resonance Data Bank (BMRB), Protein Data Bank (PDB), and AlphaFold Database.
• Utilized a support vector machine to evaluate NMR data consistency with predicted structures.

Main Results:

• Demonstrated the ability of new heuristics to identify inaccurate AlphaFold structures.
• Established a comprehensive dataset for developing and testing hybrid AI-NMR methods.
• Successfully applied the developed methods to determine the structure of the engineered protein LoTOP.

Conclusions:

• Hybrid approaches leveraging AI and NMR spectroscopy can enhance protein structure prediction.
• The developed heuristics and machine learning models provide a robust framework for validating AI-based structures.
• This work facilitates more accurate and efficient protein structure determination through integrated computational and experimental strategies.