Biotech Data Mining 33
Roland Piquepaille writes "In the last ten years, biotech companies have been busy accumulating mountains of data. And it's becoming more and more difficult to find useful information about interactions between genes and proteins for example. It's one of the reasons why the European Union has started the BioGrid project. In Mining biotech's data mother lode, IST Results describes this project. Among current results, the researchers involved in it have delivered a better search engine for PubMed by analyzing over-expressing genes and predicting the protein interactions that are likely occurring. And many of the tools developed by BioGrid are available for public use -- even by yourself. For more information, this overview contains additional details, pictures and references about this project."
I worked for a bioinformatics company ... (Score:4, Informative)
What they're hoping for is to find commonalities among individuals with certain traits, genetic diseases obviously, but also predisposition for certain drug therapies. Medications work for some people and not for others. It would be marvelous if you could do a genetic scan of someone and predict whether a certain drug therapy would work on them. It could also bring drugs to market that aren't already because they don't help some people and are in fact dangerous to others.
Great idea, but there are serious obstacles. First, the dataset that comes off a DNA sample is enormous. You can currently sample DNA in about 500,000 places, each location represented by letters, those letters representing the type of molecule present in a certain position on the double helix. There are about a half-dozen of those, so what you end up with is half a million of these letters juxtaposed. And it's not as easy as "if there's an n at position 232.922, this drug won't work for you." It's more like "if there's a series of letters in one place, and there's a series of letters in another, then you can either have a series in this place or a series that one, etc.".
Unfortunately, you don't know how long the series is and you don't know where, you don't know what depends on what. You just don't know. It's a statistical nightmare. You should see the algorithms they throw at those things.
And of course, environment may play a role in such things, too, so you don't even know if what you're looking for means anything.
Great idea, hope it works, not holding my breath.