Answers are grounded in this paper and cite their sources.
Protein mutations can affect how a protein binds a ligand and this may contribute to drug resistance as well as variability in patient responses to treatment. However, experimentally measuring binding affinities across numerous variants in real life is costly. Here we evaluated the accuracy of Boltz-2, a protein computational model used for biomolecular structure and binding-affinity prediction, by using experimentally determined reference-variant protein-ligand pairs collected from four independent data sources under the same experimental conditions. Boltz-2 predictions generally align with experimental trends, but the magnitude of large affinity changes, especially for mutations with indirect effects on ligand binding, is systematically underestimated. Structural analysis between predicted protein models by Boltz-2 and structures from lab data was compared side by side, further revealing that even accurate predictions between mutant protein-ligand structures from Boltz-2 could not guarantee accurate affinitychange predictions. These findings suggest that Boltz-2 is useful for variant prioritization and hypothesis generation but has limitations in quantitative mutation-effect prediction. They also highlight the value of matched reference-variant affinity data for benchmarking and improving future biomolecular models.
No takes yet. Share an insight, caveat, or question.
Yu et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: