Re: Protein folding thread
Posted: Sun Dec 24, 2023 3:00 pm
Alphafold's error rate is higher than most people realize.
https://www.science.org/content/blog-po ... lace-worldThe estimate is that AlphaFold structures have somewhere between 7 and 20% of the side chain residues in an incorrect orientation - sometimes slightly off, sometimes way off. Those percentages seem to be the same for functional side chains as well as ones that are away from the action.
...That makes the current state of the art in protein structure prediction very useful as a hypothesis generator (and far beyond anything we had before), but it also means that, at least as we move into 2024, that it cannot replace experimental data, either. My own guess is that improvements in accuracy may turn out to be a sort of "last mile" problem that is unlikely to be solved by the sort of sudden advances that got us this far.