A new computational method can predict chemical reactions that could generate compounds with desirable pharmaceutical properties. The approach could help drug companies design molecules of interest, before spending money on a process that might not work.