Choose what we're going to take as the 'true positive' results

[NathanSkene]

Take marker genes from Human Cell Atlas? We don't want it to be circular

Might be easiest to just pick e.g. 100 celltype specific genes, then add random genes until it's borderline significant