MOSES

Molecular Sets — distribution learning benchmark with 8 reference metrics on ZINC subset.

Composite
72.4
Experimental validation
Retrospective
Stages
Lead ID / ADMETDevelopmental Candidate
Modalities
small-molecule
Task types
molecule-generation
Size
train_set: 1,936,962
metrics: 8
License
MIT
First release
2018-11
Last updated
2022-04
Official site
→ project page
Leaderboard
→ leaderboard
Dataset
→ dataset
Code / GitHub
→ repository
HuggingFace
→ HF
Paper
Molecular Sets (MOSES): A Benchmarking Platform for Molecular Generation Models · Polykovskiy D, Zhebrak A, Sanchez-Lengeling B, et al. · 2020 · paper · doi:10.3389/fphar.2020.565644 · 550 citations
Flags
none
Experts
Daniil Polykovskiy, Alex Zhavoronkov, Alán Aspuru-Guzik
Groups
Insilico Medicine, Matter Lab (UToronto)
Hosted by
Therapeutics Data Commons (TDC), MoleculeNet
Related benchmarks
GuacaMol, Practical Molecular Optimization (PMO)

Rubric (7-criterion)

rigor
4
coverage
3
maintenance
2
adoption
5
quality
3
accessibility
5
industry_relevance
3

Notes

Distribution-learning metrics known to saturate.

← Back to all benchmarks

Compare:
Open comparison →