AlphaFold prediction of structural ensembles of disordered proteins

DOI: 10.1038/s41467-025-56572-9Key Points:- Introduces AlphaFold-Metainference, a new method combining AlphaFold predictions with physics simulations- Notable because AlphaFold was trained on structured proteins but shows surprising accuracy with disordered ones- Validated against experimental data from 11 different disordered proteins using SAXS (small-angle X-ray scattering)- Further validated against detailed simulations of two neurodegenerative disease-related proteins (Abeta and alpha-synuclein)Major Implications:- Could revolutionize study of disordered proteins involved in diseases like Alzheimer's and Parkinson's- Suggests AlphaFold may be detecting fundamental principles of protein behavior beyond just structured proteins- Opens new possibilities for drug discovery and personalized medicine- May require rethinking our definition of protein "disorder"Technical Details:- AlphaFold provides distance predictions between amino acids- Limited to predictions up to ~22 angstroms- Metainference uses physics simulations to explore possible conformations within AlphaFold's predictionsLimitations Discussed:- 22 angstrom distance prediction limit means some longer-range interactions may be missed- Particularly relevant for very extended disordered regions- May need integration with other experimental techniques for complete picture