AlphaFold can accurately predict the stability of mutant proteins

MOLECULAR & COMPUTATIONAL BIOLOGY

9/2/2024

There is no denying that DeepMind's AlphaFold algorithm, which can predict protein structures with high accuracy solely from their coding genes, represents a major breakthrough in the field of biology. However, AlphaFold has limitations, as it was trained on stable proteins that remain folded at physiological temperatures, meaning that the predicted structures cannot be guaranteed to fold perfectly and remain stable under all conditions. To address this, researchers from the Center for Algorithmic and Robotized Synthesis and the Institute for Basic Science conducted a series of experiments using AlphaFold to predict fine structural deformations caused by single mutations, measured through strain and compared them with experimental datasets of relevant folding energy changes (ΔΔG). In the study published in Physical Review Letters in August 2024, the researchers demonstrated that AlphaFold's physical strain predictions, without additional data or calculations, correlated almost as well with energy-based ΔΔG predictors, enabling the prediction of the stability of mutated proteins. These findings underscore the significant potential of AlphaFold in protein engineering, particularly in the development of drug technologies and therapies for diseases caused by protein misfolding, such as Alzheimer's.

Source: https://doi.org/10.1103/PhysRevLett.133.098401