Anubis author here. I looked into protein folding. The problem is that protein folding requires scientific data, which can easily get into the gigabyte range. That is more data than I want to serve to clients. Unless there's a way to solve the big data problem, a lot of "compute for good" schemes are frankly unworkable.
Zero knowledge proofs are basically arbitrary proof of work models. There is some interesting work being done with MPC and ZK proving, so only a small part needs to live on the client. I wonder if this would make it feasible again
