Proteins can tolerate many single-letter changes in their sequence with little effect. But some single amino acid substitutions cause local rearrangements or even disease. This study measures how far and how strongly single-point mutations change experimentally observed protein structures, and whether current prediction tools can recover those changes.

The researchers built a careful dataset from the Protein Data Bank of pairs of x-ray crystal structures: a wild-type sequence and a corresponding single-amino-acid mutant. Each wild-type sequence in the set was represented by at least five independent experimental structures. They excluded structures that were part of heteromeric complexes, bound to nucleic acids, or that contained non-standard amino acids or missing residues at the mutation site. All structures were cleaned and completed using standard tools before analysis.

To quantify structural change, the team measured, for each residue, the root-mean-square change in distances between that residue’s backbone carbon atom and its nearby residues. They then normalized this “mutation-induced deformation” by the typical variation seen among duplicate wild-type structures. This normalization filters out the protein’s normal thermal or experimental fluctuations. Averaged over their dataset, the deformation is largest at the mutated position and falls off quickly with increasing spatial distance from that site.

The authors also tested AlphaFold3 by predicting all-atom structures for the mutant sequences and computing the same deformation measure against the experimental wild-type structures. They report that AlphaFold3’s accuracy falls as the experimentally observed mutation-induced deformation grows. In other words, AlphaFold3 tends to recover small, local rearrangements but performs worse for mutations that cause large structural changes or effects far from the mutation site. The authors therefore caution that current structure-prediction methods are not reliable for identifying the most strongly perturbative mutations.