Summary
Protein structure prediction methods are unable to predict the energetics of a conformational landscape unless explicitly trained for that purpose (1–2), although Monteiro da (5) disagrees and provides some anecdotal examples. (6) show that at least for fold-switching proteins, the sampled conformation is memorized by the exact protein folding method, whi (7) show the same for binders that alternate between open-closed conformations. A step in the right direction is how BioEmu reproduces values to within 1 kcal/mol when modeling folded and unfolded states and comparing their populations (8); however, later results showed that it isn’t as effective at predicting the equilibrium dynamics of mutated enzymes as well as inverse folding NNs.
1.
Jing B, Berger B, Jaakkola T. AlphaFold Meets Flow Matching for Generating Protein Ensembles. In: GenBio@NeurIPS2023. 2023. Available from: https://openreview.net/forum?id=yQcebEgQfH
2.
Vani BP, Aranganathan A, Tiwary P. Exploring Kinase Asp-Phe-Gly (DFG) Loop Conformational Stability with AlphaFold2-RAVE. Journal of Chemical Information and Modeling. 2023;64(7):2789–97. Available from: https://doi.org/10.1021/acs.jcim.3c01436
3.
Riccabona JR, Spoendlin FC, Fischer A-LM, Loeffler JR, Quoika PK, Jenkins TP, et al. Assessing AF2’s ability to predict structural ensembles of proteins. Structure. 2024;32(11):2147-2159.e2. Available from: https://doi.org/10.1016/j.str.2024.09.001
4.
del Alamo D, Sala D, Mchaourab HS, Meiler J. Sampling alternative conformational states of transporters and receptors with AlphaFold2. eLife. 2022;11. Available from: https://doi.org/10.7554/elife.75751
5.
Monteiro da Silva G, Cui JY, Dalgarno DC, Lisi GP, Rubenstein BM. High-throughput prediction of protein conformational distributions with subsampled AlphaFold2. Nature Communications. 2024;15(1). Available from: https://doi.org/10.1038/s41467-024-46715-9
6.
Schafer JW, Porter LL. AlphaFold2’s training set powers its predictions of fold-switched conformations. openRxiv; 2024. Available from: https://doi.org/10.1101/2024.10.11.617857
7.
Lazou M, Khan O, Nguyen T, Padhorny D, Kozakov D, Joseph-McCarthy D, et al. Predicting multiple conformations of ligand binding sites in proteins suggests that AlphaFold2 may remember too much. Proceedings of the National Academy of Sciences. 2024;121(48). Available from: https://doi.org/10.1073/pnas.2412719121
8.
Lewis S, Hempel T, Jiménez-Luna J, Gastegger M, Xie Y, Foong AYK, et al. Scalable emulation of protein equilibrium ensembles with generative deep learning. Science. 2025;389(6761). Available from: https://doi.org/10.1126/science.adv9817