Those are cool application ideas. We don't require a lab, although that would be good to have. It's probably safe to say that there is more Bio data than people resources available to comb through and understand it.
I like the approach of that protein folding project, they used AI/ML to find and imitate the best protein folding approaches from a list of approaches generated by humans. That way the software is optimizing something that is already known to work in some fashion. That seems more doable and immediately useful than a %100 learning agent.
We could create a variation of the project they have done, for protein folding, or we could make a new game for gene to evolution/function/cross species matching, using BLAST:
Or an IGEM related simulator:
Another big problem, is that there are more Bio papers in existence than can be understood by anybody, often even in niche areas like for 1 type of bacteria. So the researcher is presented with a serious problem of having to search through thousands of journal papers. An NLP knowledge representation, matching and search program would be a huge benefit. Such as 'show me the 100 most relevant papers related to how molecule x may be involved in disease y.'
Here is a great free book on AI and Bio, especially the first chapter: