Altering the specificity of an enzyme requires precise positioning of side-chain functional groups that interact with the modified groups of the new substrate. This requires not only sequence changes that introduce the new functional groups but also sequence changes that remodel the structure of the protein backbone so that the functional groups are properly positioned. We describe a computational design method for introducing specific enzyme-substrate interactions by directed remodeling of loops near the active site. Benchmark tests on 8 native protein-ligand complexes show that the method can recover native loop lengths and, often, native loop conformations. We then use the method to redesign a critical loop in human guanine deaminase such that a key side-chain interaction is made with the substrate ammelide. The redesigned enzyme is 100-fold more active on ammelide and 2.5e4-fold less active on guanine than wild-type enzyme: The net change in specificity is 2.5e6-fold. The structure of the designed protein was confirmed by X-ray crystallographic analysis: The remodeled loop adopts a conformation that is within 1-Å C␣ RMSD of the computational model. computational protein design ͉ loop modeling C omputational protein design methodology has been used to optimize properties such as protein stability (1, 2) and to introduce functions such as binding of small molecules (3), proteins (4), and nucleic acids (5), as well as enzymatic activity (6, 7). In most of these studies, the implicit assumption that the structure of the polypeptide backbone would remain largely fixed despite mutations of amino acid side chains was made for the sake of computational tractability.Explicit remodeling of the polypeptide backbone makes possible further optimization of these and other structural or functional properties. The set of combinations of protein sequences and structures is considerably larger when backbone flexibility is allowed and is likely to contain conformations that optimize a desired property to a greater degree than the original scaffold. This is illustrated by the backbone shifts that accompany functional divergence in natural protein evolution. Previous studies have achieved functional changes by backbone alteration, but relied on grafting methods that are restricted to sequences of known structure and function (8-10), which may be suboptimal with respect to the desired property.De novo protein structure prediction methods are well suited for sampling novel backbone conformations (11). These methods have recently been extended to focus sampling on conformations that satisfy specific positional constraints (12, 13). Computational design algorithms that iterate between sequence design and backbone optimization using structure-prediction methods have been used to design previously unobserved protein fold and loop conformations (2, 14), but have not yet been applied to achieving functional changes such as alteration of an enzyme's substrate specificity.We have developed a computational design algorithm that uses constr...