The Boltz-2 affinity module cannot be effectively repurposed for PPI affinity prediction

Summary

The Boltz-2 affinity module for small molecule-protein affinity prediction cannot be effectively repurposed for protein-protein interaction affinity prediction (1). They found that it underperformed fine-tuning from ESM embeddings. However, appending Boltz features to ESM features improves affinity prediction. This limitation is specific to repurposing the native affinity module: PreFold-dG uses distance-weighted intermediate representations in a separate predictor and reports strong $\Delta G$ and $\Delta \Delta G$ performance (2).

See also

• All-atom structure and affinity predictors partially generalize to point mutants

1.

King J, Cornwall L, Nica AC, Day J, Sim A, Dalchau N, et al. On fine-tuning Boltz-2 for protein-protein affinity prediction. 2025; Available from: https://arxiv.org/abs/2512.06592

2.

Park S, Yim S, Kim D, Yoo K, Hwang D, Lee K, et al. PreFold-dG: estimating binding affinity of protein-protein interaction from intermediate representations of protein folding model. Bioinformatics. 2026;42(Supplement_2):btag489. Available from: https://doi.org/10.1093/bioinformatics/btag489