Summary
Coupling backbone and side chain prediction or design does not necessarily lead to better performance (1–2). Methods from early 2026 found strong performance in binder design when combining backbone design and inverse folding, and degraded performance when backbones were redesigned with ProteinMPNN afterwards. However, prior work found the opposite conclusion in protein backbone design (1,2). In parricular, Alphafold3 was found to be worse than Alphafold2 at sidechain packing (2).
1.
Chu AE, Kim J, Cheng L, El Nesr G, Xu M, Shuai RW, et al. An all-atom protein generative model. Proceedings of the National Academy of Sciences. 2024;121(27). Available from: https://doi.org/10.1073/pnas.2311500121
2.
Vangaru S, Bhattacharya D. To pack or not to pack: revisiting protein side-chain packing in the post-AlphaFold era. Briefings in Bioinformatics. 2025;26(3). Available from: https://doi.org/10.1093/bib/bbaf297
3.
Didi K, Zhang Z, Zhou G, Reidenbach D, Cao Z, Cha S, et al. Scaling Atomistic Protein Binder Design with Generative Pretraining and Test-Time Compute. 2026; Available from: https://arxiv.org/abs/2603.27950