Ivelin Georgiev remembers the first time he was asked to program life—artificial life, that is. During his sophomore year at Eckerd College in St. Petersburg, Florida, Georgiev was assigned a project programming small organisms to compete in a virtual Darwinian battlefield. “It was amazing how we could use computers to show what nature does, to simulate it” says Georgiev, a native of Bulgaria. Georgiev’s cyber-pet won the competition, and he continued to study artificial organism programs for his undergraduate thesis, but his interests began to shift toward a more applied field—computational protein and drug design.
After receiving his BS in computer science and math, Georgiev began his PhD at Dartmouth College with Professor Bruce Donald. “He gave me a project I’m still working on today,” says Georgiev. “I’ve been very fortunate.” The assignment was to develop an algorithm able to search through all hypothetical protein sequences and structures and emerge with a selected few most likely to have desired target properties.
Once a protein’s basic structure is determined, scientists can begin to imagine redesigning it. Mutating a protein can alter such properties as stability, binding preference, and function. But there are countless possible mutations for any single protein, and it would be prohibitively expensive to test them all in a wet lab, says Georgiev. That’s where computer science can help. Based on a desired outcome, protein design algorithms search the combinatorial space of all possible protein sequences and produce a narrowed set of results most likely to produce that outcome. That subset can then be physically created and tested in the lab.
While working in Donald’s lab, which moved to Duke in 2006, Georgiev, in collaboration with peers, developed a number of algorithms, including K*, MinDEE, and BD, each of which identify protein mutations in unique ways. MinDEE and BD take into account flexible protein backbones and rotamers, factors traditionally left out because of increased complexity, while K* transcends standard single-conformation models by identifying an ensemble of conformations.
“Ivelin has done a fantastic job,” says Donald. He not only attacked the bioinformatics and algorithmic problems of protein design, says Donald, but has already confirmed the feasibility of his approach. Georgiev’s algorithms were used to predict mutations of gramicidin synthetase A (GrsA), an enzyme involved in the synthesis of gramicidin, an antibiotic. Another graduate student in Donald’s lab, Cheng-Yu Chen, experimentally created the protein mutants. The resulting products worked as hoped, preferentially binding a different substrate than the natural GrsA. Since similar enzymes make a variety of other antibiotics, their redesign could be a step toward creating novel antibiotics. “It could be a new way to cope with antibiotic resistance,” says Georgiev.
So what’s Georgiev up to now? “A few things,” he laughs, including a new, more efficient algorithm and several collaborative projects. Winner of the 2007-2008 Outstanding PhD Proposal Award and Best Poster Award at the Departmental Retreat in September, Georgiev values the time he’s spent in the Department of Computer Science. “It’s a great place for me, very collaborative,” he says.