I could be wrong, but that makes it sound algorithmically similar to AlphaFold 1 where they used ML to learn properties of the folded protein and then optimized

I could be wrong, but that makes it sound algorithmically similar to AlphaFold 1 where they used ML to learn properties of the folded protein and then optimized the final structure to fit those properties