• You could use (the now defunct ?)(Has it ever been used for the scientific literature ?) Life Science Identifier LSID: http://en.wikipedia.org/wiki/LSID
• a Publisher Item Identifier (PII) http://en.wikipedia.org/wiki/Publisher_Item_Identifier (you can find it in some pubmed records)
• a Pubmed central identifier ( http://www.nlm.nih.gov/pubs/techbull/ma08/ma08_pmcid.html )
• you could also imagine to put a RDF fragment describing your annotation (that would contain the date when the paper was read, who read the paper, etc... ).

I once defined an RDF scheme for capturing metadata about biological objects. I liked the OMG's Bibliographic Query Service's data model for references. (http://www.omg.org/spec/BQS/)

It has been many years since I looked at this, but I remember thinking it had the flexibility to handle pretty much anything that we'd come across. You could look at it for ideas of what you might see.

One downside for using a PubMed identifier alone is that there are some useful papers published in things that PubMed doesn't index. Even working primarily with biology, I occasionally come across a chemistry paper or a crystallography paper that is in a journal PubMed doesn't cover.

I would go for DOI, because you can immediately tell from its appearance what it is: a DOI.

With PubMed IDs or Pubmed Central IDs you are sometimes just presented with a number (e.g., 123456 instead of PMC123456), which can be confusing. The conversion between Pubmed Central IDs and Pubmed IDs (http://www.ncbi.nlm.nih.gov/sites/pmctopmid) is also not that straightforward, because not all Pubmed Central IDs resolve to Pubmed IDs immediately (you might have to wait until the mappings are updated).

So, DOIs seem attractive due to their easily identifiable notation and their functionality as a link (as Chris said). They are also ubiquitous in the sense that they are also applied outside Pubmed Central / Pubmed covered areas.