It's laughable to declare protein structure prediction a "solved" problem. The reported precision in the best predicted models is still 10-20 times worse than coordinate precision in typical experimental structures. https://twitter.com/MoAlQuraishi/status/1333383634649313280
Especially since protein function often relies on sub-angstrom positioning of atoms (like in enzyme active sites) and/or alternative conformations. How do we learn about function from a model that has 2A RMSD relative to the truth?
Currently, there are bad homology models, or worse ones. Rarely can they replace experiments. Not a solved problem.
I'd like to clarify, the advance in prediction is impressive. However, making a big step forward is not the same as "solving" a decades old problem in biology and chemical physics.