Biologists have worked for decades and spent hundreds of millions of dollars to develop a safe and effective vaccine against HIV, but with little success. Today, the medical community is turning to computer scientists, computational biologists, and statisticians to help understand how HIV continues to evade our best efforts to activate the immune system against it. Together, they hope, we can finally design a vaccine to beat the virus.
Four years ago, the Duke Human Vaccine Institute called on Professor Scott Schmidler, an Associate Professor of Statistical Science with a secondary appointment in Computer Science, to join the effort. Schmidler specializes in models and algorithms for complex, dynamic systems in the physical and biological sciences. The Institute, directed by Barton Haynes, hoped Schmidler could use that expertise to simulate the behavior of antibodies -- proteins that the immune system produces to attack viruses.
It has been a challenge in HIV vaccine research to evoke an antibody response in people because HIV antigens are constantly mutating, and our bodies simply don't produce the right antibodies to bind those ever-changing antigens. Several years ago, scientists identified a group of antibodies that broadly inhibit many HIV strains. Hopeful, the researchers identified the antigens that bind those antibodies, injected the antigens into humans, and waited. But the patients did not produce the desired antibodies in their immune system response.
Working with Tom Kepler at the Vaccine Institute and postdoctoral fellow Kevin Wiehe, Schmidler has spent the last four years developing algorithms to simulate how immune system antibodies and HIV antigens bind to each other to better understand why some antigens evoke antibody responses and others donít. The algorithms will allow the researchers to identify which antibodies bind to which antigens and how strongly they attach.
The algorithms sample the vast set of possible shapes that antigens and antibodies can take as a way to identify the best and most likely binding combinations. "We've been working on algorithms for quickly sampling this high dimensional conformational space," says Schmidler. "It turns out that algorithms that people have been using are way too slow to do this."
Today, armed with these faster algorithms, Schmidlerís lab has begun to compute binding probabilities, starting with two of the previously discovered antibodies that broadly inhibit HIV. The algorithms run on a large cluster of computers and each will likely take weeks to months to complete, because of the complexity of the problem, says Schmidler.
In the end, Schmidler and his collaborators hope to use the algorithm's binding predictions to design an antigen for a vaccine that elicits an antibody that truly neutralizes HIV. "Once we make the computational predictions, then people at the Vaccine Institute can go in and test them in the lab," says Schmidler. "It's not an easy problem, but weíll get there."