Accounting for the unbound state improves inverse-folding prediction of binding-affinity changes

Summary

Accounting explicitly for both bound and unbound states improves inverse-folding estimates of mutation-induced changes in protein-protein binding free energy (1).

Details

Boltzmann Alignment subtracts unbound-state log-likelihood contributions and improved unsupervised and supervised Spearman correlations on SKEMPI v2 from 0.2632 to 0.3201 and from 0.4324 to 0.5134, respectively (1). The method was also evaluated for antibody optimization.

See also

• Zero-shot protein stability prediction using inverse folding models can be improved by subtracting predictions from residue in isolation

1.

Jiao X, Mao W, Jin W, Yang P, Chen H, Shen C. Boltzmann-Aligned Inverse Folding Model as a Predictor of Mutational Effects on Protein-Protein Interactions. In: International Conference on Learning Representations. 2025. Available from: https://proceedings.iclr.cc/paper_files/paper/2025/hash/2e10d50dfd2a9d52c06fbcd4ed89a022-Abstract-Conference.html