Der deutsche Education Blog

April, 2013

Microsoft Research Connections Blog

The Microsoft Research Connections blog shares stories of collaborations with computer scientists at academic and scientific institutions to advance technical innovations in computing, as well as related events, scholarships, and fellowships.

April, 2013

  • Microsoft Research Connections Blog

    Paris, Machine Learning, and the Wonders of Streaming


    Eiffel Tower, ParisStanding on the banks of the Seine, I found myself marveling at the beauty of “April in Paris” (cue the 1930s song). Perhaps it was the scent of the flowers in bloom in the Jardin du Luxembourg or the warmth of the Continental spring. But, most likely, my contentment stemmed from the success of the just-completed Microsoft Research Machine Learning Summit, which took place April 22 to 24 in the “City of Light” on Le Campus de Microsoft France.

    Andrew Blake, director of Microsoft Research CambridgeThe event brought together more than 230 attendees and presenters, including thought leaders from computer science, engineering, statistics, and mathematics. Through keynotes, demos, and panel discussions, we highlighted some of the key challenges in this new era of machine learning and explored the next generation of approaches, techniques, and tools that researchers and scientists need to exploit the information revolution for the benefit of society.

    Judea Pearl, professor emeritus at UCLAAs exciting as the in-person event was, I was equally enthused by the reception of our streaming broadcast of key presentations and interviews, which was viewed by some 3,000 people around the globe. The live, online presentation not only made it possible for many more people to view the summit, it gave a broader group of students and researchers an opportunity to engage directly with some of the top experts in the field of machine learning, including Andrew Blake, director of Microsoft Research Cambridge, and Judea Pearl, professor emeritus at UCLA. I was pleased that many from the online audience posed questions about computer vision to Professor Blake, and difficult questions about probability and causality to Professor Pearl.

    There was hardly an area of machine learning that wasn’t explored in depth at the summit—from the aforementioned topics of computer vision and causality, to insightful presentations on Bayesian statistics and the use of machine learning techniques in the realm of social media and large-scale learning.

    Of course the food was outstanding (it was Paris, after all), and meals were made all the more enjoyable by the stimulating conversation of our companions and the spectacular views of Paris from the thirty-fourth floor of our hotel. But for me, the most exciting moments were the intense discussions I observed taking place during breaks and the social events, and the sense that seeds of exciting new ideas were being planted that would germinate in the months and years ahead.

    Chris Bishop, Distinguished Scientist, Microsoft Research Cambridge

    Learn More

  • Microsoft Research Connections Blog

    HLA-C: An Underappreciated Force in HIV Control


    HLA-C: An Underappreciated Force in HIV ControlAlthough medical science has made great progress in managing HIV infection through modern drugs, 1.7 million people die of AIDS each year, with a disproportionate number of deaths in developing countries. Even access to life saving drugs cannot cure the disease: patients require lifelong drug maintenance and face the never-ending danger of developing resistance or adverse side effects to the medications.

    An HIV vaccine thus remains an utmost public health priority. To this end, studying the mechanisms by which some people are able to naturally control infection offers hope for researchers seeking insights into what constitutes an effective immune response—and how we might design a vaccine to illicit such a response. In the April 5 issue of Science, an investigative team, led by Richard Apps and Mary Carrington of the National Cancer Institute and aided by researchers in the eScience group at Microsoft Research, reported a new finding that sheds light on the protective potential of the human gene HLA-C, an often overlooked player in the adaptive immune response.

    Left untreated, the vast majority of HIV-infected individuals will progress to AIDS, marked by the loss of important cells of the immune system and the resulting onset of opportunistic infections. However, the rate of progression varies widely: the virus progresses within weeks in some individuals, while others control the virus and remain AIDS-free for decades.

    Epidemiologic studies of HIV control have repeatedly pointed to the importance of the MHC locus, a cluster of genes that encode proteins that the immune system uses to identify cells that have become virally infected. Of these genes, HLA-B has emerged as a dominant player. Its neighbor, HLA-C, has been largely ignored. The reasons for this are varied, including the relatively low cell-surface expression of HLA-C proteins compared to HLA-A and HLA-B, the observation that HIV actively down regulates surface expression of HLA-A and HLA-B but appears to ignore HLA-C, and the problem that HLA-B and HLA-C genes tend to be inherited together, so any positive effects that could be attributed to HLA-C are often assumed to be the result of neighboring HLA-B. The result is a relative dearth of scientific knowledge regarding the role HLA-C plays in controlling HIV.

    Recently, several genome wide association studies have been published that report common genetic variants that correlate with natural HIV control. One of the largest such studies, published in Science in 2010 and coauthored by many of the same investigators as the current study, found a number of important variations in MHC, but the most significant signal was immediately adjacent to the HLA-C gene. Several follow-up studies from Dr. Carrington’s group and others have provided circumstantial evidence that this genetic variant is an imperfect marker for variations in the level of HLA-C cell surface expression—that is, the number of HLA-C proteins present on the cell surface. Now, Dr. Carrington has provided epidemiological evidence that HLA-C expression directly correlates with control, while Microsoft Research Distinguished Scientist David Heckerman and I used models of sequence evolution combined with functional immune response data to provide a proposed mechanism and corroborating evidence that HLA-C expression modulates immune and viral responses. Thus, in contrast to HLA-A and HLA-B, it isn’t that individual variants of HLA-C proteins contribute to varying degrees of control (although that could also be the case), but that overall cell-surface quantities of the protein, regardless of variant, are directly correlated with control, rates of immune targeting, and magnitude of evolutionary pressure exerted upon the virus. These findings suggest a broader role for variations in HLA surface expression across a range of diseases. Indeed, in addition to the protective effect of HLA-C expression on HIV, we observed a correlation between HLA-C expression and increased susceptibility to Crohn’s disease, a complex inflammatory bowel disease that may be related to an overly active adaptive immune response.

    Although the finding that increased HLA-C expression levels can contribute to both pathogen control and disease susceptibility complicates our understanding of the immune system, it highlights the importance of this long-overlooked protein and may unlock new research into the mechanisms of natural control, providing potential new targets for vaccine design.

    Microsoft Research’s involvement in this study is the result of more than seven years of ongoing research in the HIV community. We have forged ongoing collaborations with more than a dozen labs and have developed statistical models of HIV evolution that have:

    Our ongoing research develops and uses tools derived from machine learning and applied statistics to move toward the development of an effective HIV vaccine.

    Jonathan Carlson, Researcher, eScience Research Group, Microsoft Research Connections

    Related Links

Page 2 of 2 (5 items) 12