in reply to Using Perl in a personalized medicine application.
Unless you do this only as a learning exercise, I think that the main question you should answer for yourself: is it not unnecessary? I am not quite sure, but I think EnsEMBL, UniProt, Entrez between them already have a lot, perhaps even all of the things that that PyCon quote mentions. EBI, EMBL, NCBI, they all have large teams working on both data, website, and interfaces. Make sure you find a useful addition/niche before even starting to work on it.
You also omit to mention what this would be used for. What is the audience? What kind of queries do you want to answer? How fast, how many, how reliably.
In general, I think the size of the source-databases involved (chebi, GO, human and perhaps other genomes) would make a database necessary, unless you envisage a system that makes use their respective API's.
I don't think an ORM (or which ORM) should be your first concern. Such things are pretty much downstream of the above questions.
A last consideration: have you contacted the poster presenters? Most people love it when others show interest in their work, and they are possibly quite willing to share their work. Even if you don't use it it may give you good ideas.