Overall it's pretty kickass, and the papers themselves tell why it's HIV in particular: HIV evolves extremely quickly within a specific host, so any general solution needs to be able to account for a virus that can adapt to your specific treatment regimen.
And that's the link to spam-fighting: rapidly changing strategies to fool your immune system into thinking it's a good cell that should be let in, which is amenable to similar spam-fighting approaches.
These are great links, and answer many of the questions that I had. But there must be something more to the story; PhyloD is from 2007 and the news article is from 2011. Tangentially related, the research article that you linked is cited by a paper in PNAS that describes extraordinarily constrained coevolving group of sites in HIV (which appears to be a site of protein-protein interaction in the Gag protein). Such a group is likely a good target for multiple simultaneous therapeutics, as multiple mutations there likely reduce HIV's intrinsic fitness: http://www.pnas.org/content/108/28/11530.full
Checking David Heckerman's web page shows this paper from October 2011 in the Journal of Virology. I don't know how it relates to the news article specifically but it's interesting if you'd like to follow this stream of work.
My take is that MR is contributing compute resources as much as they are contributing algos. It's too bad that neither article really describes the problem domain nor the solution space. If this were a genetic study of HIV resistance to ARVs (and it seems like it is but again, few details), one could at least imagine having to look for 3-way interaction terms across the HIV genome, a large search space. Would be interesting to know if this is the problem they are tackling, and if MR's major contribution is algorithms or compute.
I love reading stories like this, but with a background in fighting spam I'm a bit skeptical. I would love to hear more details.
For those of you who don't know, most of Microsoft's anti-spam efforts are from techniques I suspect are hard to translate to the HIV domain. Microsoft catches the overwhelming majority of spam (98%) using IP address blocking. Most of the remaining spam is caught using long lists of regular expressions managed by humans. I would not expect researchers to be crafting regular expressions or mapping blocklists to protein sequences. Maybe they are, but the article makes it sound like some algorithmic approach.
Obviously there are more modern techniques for fighting spam, but Microsoft isn't using them yet, and I hardly think of Microsoft as a leader in this space.
The people in Microsoft Research are very different people than the ones in charge of Hotmail. It could be that the Research people are more familiar with content analysis methods.
I've said it before, but I really think Microsoft would be a completely different (and better) company if Microsoft Research was a bigger part of their product development. Just look at the insane stuff they're doing with Kinect, for instance.
While you mentioned that you would like to know more details, at the same time you make it sound like you know the current inner workings of how Microsoft spam filters work.
Do you work for Microsoft spam filtering/email software teams and know the details you mentioned and thus have the inside information? Even then, I am not sure how you can extend that to what are the spam filtering techniques that Microsoft Research has been working on lately.
Fair questions and no details to share :). I don't work for Microsoft, but my path has crossed with the FOPE team a few times where we got into specifics about their platform. Hence my surprise to see this article about how their tech is used by research.
We are using a key principle that we use in fighting spam to fight HIV. In the case of spam, we have spammers changing their emails to get around the spam filters. So, we go after their Achilles heel: their need to extract money. In the case of HIV, we have HIV mutating to get around our immune system. Here, we’re again looking for the Achilles heel(s) of HIV—vulnerable spots on the virus that, if they mutate, compromise the function of the virus. One step in this approach is to catalog the spots along HIV that our immune system can target. This is where PhyloD comes in. Here are some of the articles describing our search for targets using PhyloD:
• G. Alter, D. Heckerman, A. Schneidewind, L. Fadda, C. Kadie, J. Carlson, C. Oniangue-Ndza, M. Martin, B. Li, S. Khakoo, M. Carrington, T. Allen, M. and Altfeld M. HIV-1 adaptation to NK-cell-mediated immune pressure. Nature, 476 (7358): 96-100, August 2011.
• A. Bansal, J. Carlson, J. Yan, O. Akinsiku, M. Schaefer, S. Sabbaj, A. Bet, D. Levy, S. Heath, J. Tang, R. Kaslow, B. Walker, T. Ndungu, P. Goulder, D. Heckerman, E. Hunter, and P. Goepfert. CD8 T cell response and evolutionary pressure to HIV-1 cryptic epitopes derived from antisense transcription. JEM, 10.1084/jem.20092060, January 2010.
• C. Berger, J. Carlson, C. Brumme, K. Hartman, Z. Brumme, L. Henry, P. Rosato, A. Piechocka-Trocha, M. Brockman, P. Harrigan, D. Heckerman, D. Kaufmann, and Ch. Brander. Viral adaptation to immune selection pressure by HLA class I-restricted CTL responses targeting epitopes in HIV frameshift sequences. JEM, 10.1084/jem.20091808, January 2010.
http://mscompbio.codeplex.com/wikipage?title=PhyloD
Overall it's pretty kickass, and the papers themselves tell why it's HIV in particular: HIV evolves extremely quickly within a specific host, so any general solution needs to be able to account for a virus that can adapt to your specific treatment regimen.
And that's the link to spam-fighting: rapidly changing strategies to fool your immune system into thinking it's a good cell that should be let in, which is amenable to similar spam-fighting approaches.