"One of our research problems is to figure out what's missing from our designs that naturally occurring enzymes have figured out," says Baker. In follow-up studies, his group has taken two approaches to this problem: refining its computer algorithms, and asking nature to step in where the researchers left off. By using their minimally functional enzymes as evolutionary starting points, the researchers can use directed evolution to create more efficient catalysts.

In the past, directed evolution has been an alternative approach to creating desirable enzymes. But Baker believes that it can be used as a complement to computational approaches. Computational design gives researchers a way to build proteins from the ground up, allowing them to engineer enzymes for reactions that have no natural counterpart. "That way, we would be free from the tyranny of having to find something in nature to start from," says Arnold, whose work has focused heavily on directed evolution.

But directed evolution provides a means of making those structural tweaks that the computational designs algorithms aren't yet sophisticated enough to handle. "It's actually the wave of the future," says Baker, "because these directed-evolution experiments can capture much more subtle things than we can capture in the calculations."

Baker's is not the first group to tackle computational enzyme design. For example, Caltech biochemist Steve Mayo, a pioneer in computational protein design, reported the creation of enzymes from existing nonenzyme proteins in 2001. But Baker's approach differs in that it doesn't use existing proteins as starting points--it's true de novo design.

Arnold says that Baker's enzymes are also more powerful than Mayo's, but that it's hard to nail down precisely how much more. "It's a different kind of enzyme, so you can't really compare apples and oranges," she says. "But this is a pretty good apple."