USPTO-50K / USPTO-MIT (Retrosynthesis)
Reactions extracted from USPTO patents; standard retrosynthesis/forward-reaction benchmark.
Composite score: 78.0
Rubric (1–5 per criterion)
rigor
4/5
coverage
4/5
maintenance
2/5
adoption
5/5
quality
3/5
accessibility
5/5
industry_relevance
4/5
Metadata
Stages
Lead ID / ADMETDevelopmental Candidate
Modalities
small-molecule
Task types
retrosynthesisreaction-prediction
License
Public
First release
2017
Last updated
2023
Flags
data-leakage-known
Size & scope
- reactions: 1800000
- canonical_50k: 50037
Primary paper
Title
Neural Sequence-to-Sequence Models for Retrosynthesis Prediction
Authors
Liu B, Ramsundar B, Kawthekar P, et al.
Year
2017
DOI / arXiv
Citations
520
Links
- Official site: https://github.com/Hanjun-Dai/GLN
- GitHub: https://github.com/Hanjun-Dai/GLN
- Leaderboard: N/A
Hosted by (initiatives)
Experts (primary authors / maintainers)
Groups (host labs / companies / consortia)
Related benchmarks
- none
Notes (honest caveats)
Known leakage across canonical splits; use time-split or ORD for fairer eval.