Medicinal chemists have advanced enzyme inhibitors all the way to clinical trials and FDA approval by using fragment-based approaches—that is, by identifying chemical fragments as small as a single ring and piecing them together into high-affinity binders. That success makes researchers keen to use similar strategies to predict substrates for the scores of proteins in the human genome that have unknown functions. However, a new report concludes that fragments are unlikely to be reliable for that application (J. Am. Chem. Soc. 2014, DOI: 10.1021/ja501354q). Prior substrate discovery studies using the fragment approach had mixed results. So Brian K. Shoichet of the University of California, San Francisco, teamed with Karen N. Allen of Boston University, Frank M. Raushel of Texas A&M University, and colleagues to probe the idea further. They systematically fragmented known substrates for enzymes from three different families then measured how well enzymes acted on each fragment. To the group’s surprise, minuscule changes obliterated enzyme activity by orders of magnitude. Even though fragments have helped researchers discover a few enzyme substrates previously, the team thinks that a library of fully elaborated metabolites will be better suited to the task.