Summary
Protein folding neural networks make local optimizations in the absence of coevolutionary information (1). This was observed using AlphaFold2 and includes side chain optimizations and resolution of steric clashes. The observation is consistent with the hypothesis put forth by Roney & Ovchinnikov (2) that they learn an energy function but require evolutionary info to traverse the structural space.
Figures
Ref (1)
1.
Gut JA, Lemmin T. Dissecting AlphaFold2’s capabilities with limited sequence information. Bioinformatics Advances. 2024;5(1). Available from: https://doi.org/10.1093/bioadv/vbae187
2.
Roney JP, Ovchinnikov S. State-of-the-Art Estimation of Protein Model Accuracy Using AlphaFold. Physical Review Letters. 2022;129(23). Available from: https://doi.org/10.1103/physrevlett.129.238101