The synthesis of a complex molecule begins from an initial design stage1–4 in which possible routes are triaged by strategy and feasibility, on the basis of analogy to similar reactions2,3. However, as molecular complexity increases, predictability decreases5; inevitably, even experienced chemists resort to trial and error to identify viable intermediates en route to the target molecule. We encountered such a problem in the synthesis of picrotoxane sesquiterpenes in which pattern-recognition methods anticipated success, but small variations in structure led to failure. Here, to solve this problem but avoid tedious guess-and-check experimentation, we built a virtual library of elusive late-stage intermediate analogues that were triaged by reactivity and altered the synthesis pathway. The efficiency of this method led to concise routes to 25 naturally occurring picrotoxanes. Costly density-functional-theory transition-state calculations were replaced with faster reactant parameterizations to increase scalability and, in this case, inform the mechanism. This approach can serve as an add-on search to human or computer-assisted synthesis planning applicable to high-complexity targets and/or steps with little representation in the literature or reaction databases.
