AlphaFold-predicted deformation probes changes in protein stability

AlphaFold2 (AF) is an excellent structure predictor, but it does not unambiguously indicate whether the predicted structure is stable. Could it be that AF contains sufficient information to predict protein stability, and one just needs a way to extract it? To address this question, we investigate single-mutation effects on protein stability using a large empirical dataset of changes in free energy of folding ΔΔ G , and calculate the deformation, measured by strain, between AF-predicted mutant and wild-type protein structures. Our initial motivation was to assess if anything could be inferred about protein stability from AF-predicted structures, but our results showed unexpectedly that strain alone – without any additional data or computation – correlates almost as well with ΔΔ G as state-of-the-art energy-based and machine-learning predictors. In particular, high strain in buried residues generally gives rise to large changes in stability. Thus, our findings indicate that AF-predicted structures encode significant information on stability, suggesting that (de-)stabilizing effects of mutations may be estimated using AF. This paves the way for the development of new structure-based algorithms that accurately predict and explain how mutations affect stability. ### Competing Interest Statement The authors have declared no competing interest.