> Having better tox models seems like the highest value, albeit very difficult, route here. Which to me is a separate, more general, problem than a specific disease model.

Agreed. Better toxicity testing would go a long way towards solving some aspects of the search problem.