Thanks to the wonderful Ensembl, Unichem, RGD and MGI for providing the underlying data for these mappings.

However, there is a lot of call for simple access to Termite in other languages (particularly as java and python). As as Java programmer myself, i understand the value of being able to do such a thing, particularly if you have a lot of your code in that language.

One thing I’m just adding to our Termite download pack is an example of how you might use the perl Inline functionality to write code accessing termite in your own language. It supports a number of languages, including Python and Java. For example, lets say we really like Python for processing Termite data.

Red: Perl code just booting up termite

Blue: Python code actually doing the text mining

# Here we create  simple single termiter and then use Python to do all the work
# We ask for the results in json as that makes life a lot easier for python
use strict;
use Termite::TermiteControl;

# Perl's job is just to boot up our search objects
# We need to create a subroutine just to return our Termite instance
sub get_termite { 
   my $s = Termite::TermiteControl->new()->getNew("COMPANY"); 
   return $s; 
}

# Thats it for the perl. Now Code your python here
use Inline Python => <<E;
print "hello there from python... running termite...";
result  = perl.get_termite().runSimple("Pfizer and merck");

for item in result:
  print item.toJson();

E

As you can see we used Perl just to set up the Termiter object (so yeah, still a little perl needed). But once thats done, you can run it from Python. Of course, you wouldn’t write all your python in this script – you’d write a python function/package to process the data and just import it and call it from this script.

The main advantages here are that you aren’t having to boot up an entirely new instance of perl (e.g. if you did a system() command from another language) nor do you have the network overhead that comes from using the Termite socket service.

So, there you go. I think this doesnt fit with every use-case, and its still a perl script you’re running.  But its certainly a very useful method for using our text mining suite.

Each row is a different gene. The bar chart tells you how significant the gene-disease linkage was, per year, from 2002-now. For each gene-disease combination we review all the articles published that year and score them. A high score means that yes, this article was “about” the gene – disease relationship. A low score means that the mention was spurious or not immediately relevant. While we do also log actual counts (the number of articles that mention the gene and disease), we find that the score is as good a guide as anything else.

So, you can see all the usual suspects at the top. For instance, we can see the first few have all been mentioned very significantly with MS from 2002-2013.

these are well precendented genes for this disease and you can see that each year brings a raft of highly relevant publications discussing the two topics together.

But lets scroll down a bit in the >900 results (!)

now we can see things like paraoxonase 1 that have only been linked to MS within the last few years, there are many more of these “recent” discoveries that can easily be found by a quick glance at the data.

I am particularly interested in the apparent “dead ends” – those things that were mentioned (perhaps excitedly) with MS a while ago but have had very little follow up since. For instance, the always interesting sirtuin-1, was linked to MS from about 2008-2011 but we’ve not heard anything else in the literature on this until yesterday (!)

Others such as PADI4 stop abruptly due to a fairly conclusive negative finding.

Then there’s interesting proteins such as neurofascin, which is sort of bobbling along with only a few papers per year. 

and finally, things like CD73 which again showed some significant interest a good few years back but have not delivered many papers since.

So, as you can see, using the literature space API gives you an unprecedented overview of the Multiple Sclerosis gene-disease relationship matrix. And of course, this can be repeated for many different diseases or the inverse (show me diseases linked to gene X).

If you want to play with the MS data, the spreadsheet is linked below. Our API makes it incredibly simple to generate this data and provides a lot more detail, including the pubmed IDs of every relevant document.

Data: multiple_sclerosis_gene_trends

OK, back to scibite, so we’re working on a ton of great stuff with SureChem, some more details on that soon. In the meantime, SciBite+ is progressing well. The first part of this is to create the underlying data mining methods and API calls needed. One of the things we’ve just finished is the “Literature Space” methods. These allow you to identify correlations between topics in the historical medline literature. Some of you know that at Pfizer we created such a system called Pharmamatrix that did this (described here and here). It was a very useful tool – being able to rapidly ask “what are all the diseases associated with this protein” really aids drug repurposing, while the corresponding question “what are all the proteins associated with this disease” provides a  rich hunting ground for new target discovery. And there are many more uses of course.

When creating SciBite i was struck by the lack of a freely available, computational infrastructure for this kind of thing. There’s quite a few sites that will offer things along these lines, but nowhere that provides it in an easy to consume API, and connects it with emerging trends across the world wide web.

The API call is pretty simple:

curl -v  -X GET "<scibite_services_server>/site/sciapi/v3_1/InsightLitSpaceService?q=GENE:PDE5A&entityType=MESHDISEASE"

Running this will give you all of the diseases associated with the gene PDE5A in the last decade and their associated articles for further verification. And of course, each correlation is identified and scored using our proprietary Termite technology which gives accurate and insightful results.

So, if you’re interested in this type of analysis please get in touch. If you’re not an API person, then a user interface is coming later this year as part of SciBite+, so stay tuned to the blog for more information on that.

Finally, for those who do use our API, we put together a much needed extended user guide (using the simply wonderful Documenter tool) containing code examples and other useful tips. The API guide (screenshot below) is only available to registered API users, so please get in touch if you’d like access.

Thats it for now,

Lee

• InsightTopTypesService: This provides a set of the most mentioned topics of a given type (e.g. Drugs) in the news over the last “n” months. So you can ask what are the top drugs in recent patents, the top proteins in grants or any combination really.
• InsightJointInSourceService: This is a bit more in depth than the first one. Here we take a particular topic of interest (a gene, a disease, a company) and ask for other topic of a given type which have been most recently connected. So, a question like what targets have been associated with Pfizer in the recent patent literature, could be answered with this API method

If you’re interested in accessing our API, just get in touch!

cheers

• Firstly hello to all our new users, the press releases, BioIT world and other contacts are driving some healthy new visitors to the site, so welcome.
• Our work with SureChem is progressing nicely, it’ll be a while before we can show what we’ve done but its already looking promising
• Dave is doing some awesome work on our thesauri – a new version of our disease terminology is almost ready with lots of good new stuff in it. We’ve also very mcuh in the deep-end of our drug thesaurus update, still a lot of work to do there, but we have a version0.00001 which is something, believe me!
• The bookmarklet (webconnect, see last post) is also coming along nicely. Hopefully will be able to release it to the world in a week or so
• SciBite+ is still very much under construction. Some nice new API calls should be ready in the next couple of months, followed by the website sometime after.

Thats it for now!

SciBite WebConnect is a bookmarklet. You simply go to a page you’re interested in and click on a special bookmark in your browser toolbar… Instantly, SciBite scans the page looking for the drug discovery topics it knows about. Here’s a screenshot on a pubmed abstract:

As you can see, we correctly recognised the drugs, indications, genes and other major topics in the article. You simply hover-over the term of interest and you’re presented with links to the live newsfeed at scibite.com.

The bookmarklet should be available in the next week or two, we’ll post info here when it is.