MoleculeACE
Benchmark testing model robustness on activity cliffs across 30 ChEMBL targets.
Composite
83.3
Experimental validation
Retrospective
Stages
Lead ID / ADMET
Modalities
small-molecule
Task types
regressionactivity-cliff
Size
targets: 30
molecules: 48,000
molecules: 48,000
License
MIT
First release
2022-11
Last updated
2024-05
Official site
Leaderboard
→ leaderboard
Dataset
→ dataset
Code / GitHub
HuggingFace
→ HF
Paper
Exposing the Limitations of Molecular Machine Learning with Activity Cliffs · van Tilborg D, Alenicheva A, Grisoni F · 2022 · paper · doi:10.1021/acs.jcim.2c01073 · 180 citations
Flags
none
Experts
Groups
Hosted by
—
Related benchmarks
Rubric (7-criterion)
rigor
5
coverage
3
maintenance
3
adoption
4
quality
5
accessibility
5
industry_relevance
4
Notes
Critical stress-test for generalization; exposed GNN weaknesses.