My rough plan is this: determine the set of paralogs in zebrafish that have orthologs in human such
that one human gene maps to two paralogous zebrafish genes. I would reinforce this by making sure
the paralogs have 2 to 2 orthologs in fugu. However, as you point out, this is not very precise at
all and the Brunet 2006 paper makes use of genes tress and synteny on top of orthology and paralogy.