That certainly makes sense since the problem has a higher likelihood of stemming from the same root cause, but also speaks to how little is known.

So include the social graphs of people who are on antidepressant discussion boards and pick up on friends, family, socioeconomic features ...

I so want to use ML tools on medical treatment data, but it's so hard to come by. Have toyed with a couple of startup ideas based on the tendency of people who share a condition to establish ties. Crohn's disease, cutaneous lymphoma. But the groups are too small. Mood disorders on the other hand...