Retrosynthesis
A planning strategy in organic chemistry that works backward from a target molecule to identify feasible synthetic routes — AI tools now automate the search for these routes using reaction databases.
Retrosynthesis is the process of planning a chemical synthesis by working backward from the target molecule to simpler, commercially available starting materials. Instead of asking “what can I make from these reagents?”, retrosynthetic analysis asks “how could this target molecule have been made, and from what precursors?”
The approach, formalized by E.J. Corey (Nobel Prize 1990), involves identifying disconnections — bonds in the target molecule that could have been formed in a key reaction step — and recursively applying this logic until you reach commercially available starting materials.
AI and retrosynthesis:
Retrosynthesis is well-suited to AI because it can be framed as a search problem over a space of known reactions. Models trained on patent and literature reaction databases learn to suggest disconnections and predict which reactions are likely to work.
Key tools:
- IBM RXN for Chemistry — web-based retrosynthesis and forward reaction prediction
- Molecule.one — commercial platform that ranks routes by synthetic difficulty and commercial availability of intermediates
Limitations of AI retrosynthesis:
- Training data is biased toward well-documented reaction classes in the patent literature — novel or exotic chemistry is underrepresented
- Suggestions are probability-ranked, not guaranteed to work
- Stereochemical considerations (which face of a molecule a reagent attacks) are handled less reliably than connectivity
Related terms: SMILES Notation
Related guide: Chemistry
Related tools: IBM RXN for Chemistry, Molecule.one