Answered Dec 21
A recent example is:
Community Wide Experiment on the Critical Assessment of Techniques for Protein Structure Prediction
They run an annual competition, where they publicly post primary sequences of proteins.
If I understand correctly, they already know the 3-D structures, but haven’t publicly released those.
Then, many research groups around the world use the primary sequences to attempt to solve the final structures, and submit their predictions.
This year’s winner, by far, was Google’s Deep Mind AI AlphaFold project.
AlphaFold apparently uses a large database of publicly known sequence and structure solutions. Then, it compares those solved sequences with the new sequences. Apparently it makes a number of possible predictions, and finds the one with the lowest final Gibbs energy level.
It’s pretty cool, although it isn’t clear whether their system is just trying to go directly from primary sequence to final structure. Or whether it attempts to find the series of intermediary conformations that real proteins go through. That is important, since one intermediary fold may be required to trigger the next fold.