Sente: like iTunes, for biomedical literature

·

Anyone who's spent a lot of time literature searching using PubMed will have noticed that the indexing system it uses doesn't allow phrase searching: for example, it's impossible to search for the phrase "T cell costimulation in cancer" without it being split into separate words. Using Google goes some of the way to alleviating this: it indexes a lot of PubMed's abstract pages, so limiting a search to the NCBI site allows phrase searching in abstracts. I had thought that Google didn't index all of PubMed's abstracts, but today it seems that a text search for HIV produces 137603 results in PubMed direct and about 131000 in PubMed via Google (which also provides PageRank for the articles, but there's probably not much linking data to base that on yet).

The obvious solution for the phrase searching problem though, apart from licensing the whole MEDLINE database (which is free for internal, non-commercial research use, but takes up about 50GB of disk space) would be to create a local MySQL database of all the abstracts in a particular area of research and use that for phrase searching. I was about to set that up (fetching the abstracts as XML is fairly straightforward using the E-Utilities web service) when I remembered Sente.

I stand by what I said before: Sente is still ridiculously overpriced at $80 (and the reviewers at MacUpdate agree), but I see now how valuable it could be. When James Howison talked about managing your literature in the way that iTunes works for music, this is probably at least part of what he had in mind. Basically, Sente runs your search against PubMed (say "immune tolerance[MESH]" for example, which would match all the articles on the subject of immune tolerance), downloads all the abstracts (in this case 10,000) and metadata using E-Utilities, then creates a local database that you can search using a Cocoa interface.

Narrowing down a search is easy, by typing keywords like in iTunes. There's also a similar browse mode that divides the window into panes, allowing you to filter through various categories. PubMed IDs can be used to link to external resources using customisable URLs (such as http://www.pmbrowser.info/pmdisplay.cgi?uids={id} to link to the HubMed page for an article, http://www.ncbi.nlm.nih.gov/entrez/eutils/elink.fcgi?cmd=prlinks&dbfrom=pubmed&id={id}&retmode=ref to link to the fulltext and http://www.pmbrowser.info/pmris.cgi?pmid={id} to import the citation into EndNote. That last link highlights one of the problems with Sente: it doesn't contain any functions for bibliography management, so you can only send citations to Word indirectly from Sente when writing a paper.

If you want to try Sente, there's a demo version available. Searching and downloading abstracts only works for 30 days, but you can continue to browse the downloaded results after the trial runs out. If it was $20 I'd buy it straight away - at $80 I might be tempted to see how easy it would be to add similar features to BibDesk, so that searching and bibliography management can be combined in one application.