Experiment-guided AlphaFold3 resolves measurement-consistent protein ensembles.

Area of Science:

• Structural biology
• Computational biology
• Biophysics

Background:

• AlphaFold3 accurately predicts protein structures but often yields single conformations, ignoring inherent heterogeneity and experimental conditions.
• Experimental techniques like NMR, X-ray crystallography, and cryo-EM provide crucial data on protein structure and dynamics.

Purpose of the Study:

• To guide AlphaFold3 predictions using experimental data for more accurate and heterogeneous structural ensembles.
• To develop a methodology that integrates diverse experimental data, including dynamics information.

Main Methods:

• Guiding AlphaFold3 with data from nuclear magnetic resonance (NMR) spectroscopy, X-ray crystallography, and cryogenic electron microscopy (cryo-EM).
• Incorporating site-resolved order parameters to represent molecular dynamics.
• Generating compact structural ensembles consistent with experimental observables.

Main Results:

• The guided AlphaFold3 approach produces structural ensembles with improved agreement with experimental data.
• Fewer distance restraint violations were observed compared to traditional NMR structures.
• Unmodeled alternate conformations were identified in electron density maps.

Conclusions:

• This methodology enables experimentally aware predictive models for protein structures.
• The approach generates structural ensembles consistent with multi-modal experimental measurements.
• Future work can refine these ensembles using thermodynamic principles and energetics.