Summary
AlphaFold3 ipTM can distinguish antibodies that bind and those that don’t with an AUC of 0.86 (1). This was corroborated in one subsequent prospective study(2), whereas another study found this to be target-dependent(3) (see figure below for details). However, previous studies have not found the same for AlphaFold2-generation models (4). Meanwhile fine-tuned RosettaFold was also unable to distinguish these, suggesting a very high base level of performance is required to distinguish binders and nonbinders.
Figures
Ref(3)
See also
- PAE weakly correlates with Ab-Ag binding
- Protein structure prediction and design confidence metrics do not correlate with binding affinity
1.
Bennett NR, Watson JL, Ragotte RJ, Borst AJ, See DL, Weidle C, et al. Atomically accurate de novo design of antibodies with RFdiffusion. Nature. 2025;649(8095):183–93. Available from: https://doi.org/10.1038/s41586-025-09721-5
2.
Sang Z, Xiang Y, Huang W, Sargunas PR, Kim YJ (Jeffery), Feng Z, et al. Repertoire-scale antibody structural prediction informs therapeutic design. Science Advances. 2026 Apr;12(17). Available from: http://dx.doi.org/10.1126/sciadv.aef7163
3.
Harvey EP, Smith JS, Hurley JD, Granados AJ, Schmid EW, Liang-Lin JG, et al. In silico discovery of nanobody binders to a G-protein coupled receptor using AlphaFold-Multimer. Nature Communications. 2026 Apr; Available from: http://dx.doi.org/10.1038/s41467-026-72093-5
4.
Lourenço A, Subramanian A, Spencer R, Anaya M, Miao J, Fu W, et al. Protein CREATE enables closed-loop design of de novo synthetic protein binders. openRxiv; 2024. Available from: https://doi.org/10.1101/2024.12.20.629847