Comment by DoctorOetker

Are you referring to alphafold etc?

My understanding is they don't actually simulate or calculate (meta)stable states of proteins, but rather extrapolate on known folds of experimentally confirmed proteins (basically peeking at what types of folds are found in similar sequences in other proteins. then known as homologous proteins).

How proteins get folded, unfolded, refolded etc depends on the exact cellular or vacuolar environment.

AlphaFold isn't trained on the environment, it only sees the known mappings from genetic sequence to protein structure. It is patently unaware of any environmental aid or frustration in correctly folding a protein.

An incorrectly folded protein structure (putative prion structure) and its correctly folded structure share the same genetic sequence. AlphaFold is effectively blind, it was just trained on correctly folded proteins with known structure.

Unless future versions of alphafold use ML to speed up actual QM or molecular modelling calculations

I don't see how alphafold can help enumerate all potential misfolds of all proteins generated or preserved in an animal of species A and consumed in an animal of species B, and calculate all possible ways a misfolded protein from A may act as a prion in B.