A new computational approach developed at MIT makes it easier to predict mutations that will lead to optimized proteins, based on a relatively small amount of data.