Okay here is an interesting bit of news. My two bit for the Gmail invite !
Whatizit: An analysis tool that tries to make sense of a paper for you. You can upload an abstract of a paper, or a document of PDF file.
From the EBI newsitem:
The Rebholz group have released the annotation tool 'Whatizit'. It identifies protein-protein interactions, mutations and terminology from submitted text. Whatizit can tell you the meaning of words found in your text, depending on the kind of information you want to see highlighted.
Comments
Will there ever be AI?
I tried this with the pasted text (straight off a web page) from a paper entitled "Adenovirus protein VII condenses DNA, represses transcription, and associates with transcriptional activator E1A", available in J. Virol. at this link and got it to look for protein interactions.
It's quite interesting. There are basically 2 kinds of highlights: (1) links to an SRS search at the EBI for named proteins and (2) what the software has picked as interactions. Some are useful:
"...interaction between protein VII and E1A..."
Others less so:
"...GST protein bound to glutathione-agarose beads..."
Now, I can say "useful or not" because my brain recognises instantly that the first one is a biological process, the second refers to the methodology that they were using. In fact, I can scan the abstract in about the same time as Whatizit takes and extract more useful information from it.
I regard these things as interesting curiosities, but they don't improve my research. Until a computer thinks like me, they never will.
text mining
Quite right, textmining technology is still in its infancy, but the overall goal (what I understood from this talk I attended last week) is to be able to parse large quantities of text (PubMed) for example, and be able to extract interesting bits of information. This does not mean searching fulltext for particular keywords etc, but more in the nature of using a well indexed enclycopaedia. So for example, if you were interested in p52 interactions, this sort of technology would pick out not only p52, but its synonyms, binding studies, microarray experiments etc. So you would end up with a set of references which are likely to be very specific and yet offer a wide coverage.
On second thoughts
I was originally inclined to agree with Neil on this, but on second thougths this would make an excellent web service. Give it an http link and have it return the XML markup directly.
Alternatively let me have an option to conver the data to RDF. Then link it in with other RDF formatted data. Lots of possibilies...
Interesting
I tried the utility, but unfortunately I got an XML error on the output. I can see the point though, it looks interesting.
I sent you a gmail invite, however the email that you registered with was not working. Can you please email me at greg[at]nodalpoint[dot]org and I'll send anther invite, thanks.