> ML_DATASET // OGBG-MOLPCBA-MOLECULAR-GRAPHS_v1.0
OGB: ogbg-molpcba (PubChem BioAssay Molecular Graphs)
Open Graph Benchmark / Stanford University / MIT · Computational Chemistry & Drug Discovery · 437,929 small molecules represented as 2D molecular graphs with 128 multi-task biological targets
Computational Chemistry & Drug DiscoveryMIT437,929 small molecules represented as 2D molecular graphs with 128 multi-task biological targetsopen
Dataset Profile & Characteristics
Label Type:128 binary multi-task bioassay activity outcomes (extremely sparse, imbalanced)
Languages:en
License Tier:permissive-open-source
Modalities:graph
Intended Use
- Benchmarking graph neural networks for virtual screening and multi-task bioactivity prediction
Prohibited / Discouraged Use
- Direct clinical drug dosage synthesis without wet lab validation
Bias, Leakage & Privacy Risk Analysis
Privacy / Sensitive Data Risks:
Chemical compounds and bioassays; zero personal data.
Known Bias:
Extreme label sparsity (only 1.4% of compound-assay pairs tested); heavy negative class skew.
Known Benchmark Leakage:
Scaffold split by molecular Murcko framework prevents structurally similar train/test memorization.