By incorporating sequence homology and context associations, global probabilistic approaches to annotate genome-scale metabolic networks can substantially improve the accuracy of biochemical predictions, revealing potential functionality and directing experimental validation.