Microsoft combs through millions of strains of human immunodeficiency virus (HIV)

Feb 23 2005

Microsoft Research has pioneered promising new ways to combat one of humankind's most deadly viruses with advanced software typically used to analyze large computer databases and complex digital images, or to separate spam from legitimate e-mail.

Today at the 12th Conference on Retroviruses and Opportunistic Infections (CROI), Microsoft Research will show how medical researchers can use machine-learning, data-mining and other software techniques to comb through millions of strains of human immunodeficiency virus (HIV) to find the genetic patterns necessary to train a patient's immune system to fight the virus. The first of these vaccine designs are currently undergoing laboratory testing.

Microsoft Corp. researchers David Heckerman and Nebojsa Jojic are the first to use algorithms similar to those in Microsoft Corp.'s database and anti-spam software to uncover hidden patterns within the genetic mutations of the virus and the immune system of the patient. The researchers, in collaboration with doctors and scientists from the University of Washington (UW) in Seattle and Australia's Royal Perth Hospital, plan to exploit these patterns to create improved vaccine designs that pack more HIV-fighting genetic markers into vaccines. Microsoft researchers Christopher Meek and Carl Kadie and Jojic's brother (and former Microsoft Research intern), Vladimir, also contributed to the project.

"Microsoft has helped us make a tremendous leap forward in our efforts to halt a virus that has already killed nearly 30 million people worldwide," said Simon Mallal, professor and executive director of the Centre for Clinical Immunology and Biomedical Statistics at Royal Perth Hospital and Murdoch University. "Microsoft Research's contributions enabled us to filter patient data 10 times faster than any previous research technique we've used and produced vital clues about the building blocks of a vaccine -- clues that were all but impossible to find in our growing stockpile of medical data."

"The potential for these vaccines is a powerful example of how computer science is transforming medical research and other areas of science," said Dr. James Mullins, professor in the UW Department of Microbiology. "These Microsoft Research technologies weren't initially conceived as medical research tools, but they may prove to be critical to the ongoing battle to slow down or halt HIV and other deadly viruses."

The Microsoft Research-aided vaccine designs are currently undergoing laboratory testing at the University of Washington. The tests are being conducted on samples of immune cells taken from HIV-infected patients to determine how effectively the models uncover the appropriate genetic patterns. Similar tests are planned at the Royal Perth Hospital in Australia. Initial results should be available later this year.

Researchers plan to use the same techniques to analyze HIV strains from different parts of the world to gain a global understanding of vaccine components in a fraction of the time it would otherwise take. The new vaccine models may also help in the development of treatments for hepatitis C and other mutating viruses.

"Science is changing rapidly with the explosion of new data, and we've only scratched the surface of what computers can do to help advance this kind of research," said David Heckerman, senior researcher and manager of the Research Machine Learning and Applied Statistics Group at Microsoft. "Our goal is to accelerate scientific insight and radical breakthroughs by advancing the state of the art in machine learning and statistics. I'm inspired by the idea that new algorithms and software we have developed could potentially benefit so many people some day."

Technology offers unprecedented potential to transform science through advanced software and computer science techniques. In addition to their work on HIV vaccine design, Microsoft researchers are working with colleagues in other fields of science to apply their know-how and resources to the toughest problems outside traditional computer science. Other collaborative efforts include the following: