Zhao, S. et al. Nature doi:10.1038/nature12576 (22 September 2013).

Assigning functions to the millions of proteins discovered in genome sequencing projects remains a slow and tedious experimental process. Computational tools can expedite progress, but the results of computational predictions must be interpreted with care. Zhao et al. report an approach to predict enzyme substrate specificities using both structure knowledge and genome context, which they applied to annotate the known structure but unknown metabolic function of a marine bacterium protein, HpdD. They used homology modeling and in silico metabolite docking for several proteins in the genome neighborhood of HpdD to assign its function, and they confirmed their computational predictions with metabolomics and genetics experiments. The strategy can also be used to assign the functions of orthologs found in other organisms.