USPTO-50K / USPTO-MIT (Retrosynthesis)

Reactions extracted from USPTO patents; standard retrosynthesis/forward-reaction benchmark.

Composite

78.0

Experimental validation

Retrospective

Stages

Lead ID / ADMETDevelopmental Candidate

Modalities

small-molecule

Task types

retrosynthesisreaction-prediction

Size

reactions: 1,800,000
canonical_50k: 50,037

License

Public

First release

2017

Last updated

2023

Official site

→ project page

Leaderboard

→ leaderboard

Dataset

→ dataset

Code / GitHub

→ repository

HuggingFace

→ HF

Paper

Neural Sequence-to-Sequence Models for Retrosynthesis Prediction · Liu B, Ramsundar B, Kawthekar P, et al. · 2017 · paper · doi:10.1021/acscentsci.7b00303 · 520 citations

Flags

data-leakage-known

Experts

Bharath Ramsundar, Connor Coley

Groups

MIT CSAIL / Jameel Clinic / Coley Lab, Coley Lab (MIT)

Hosted by

Therapeutics Data Commons (TDC), Papers With Code — Drug Discovery

Related benchmarks

—

Rubric (7-criterion)

rigor

coverage

maintenance

adoption

quality

accessibility

industry_relevance

Notes

Known leakage across canonical splits; use time-split or ORD for fairer eval.

← Back to all benchmarks

Compare:

Open comparison →