Encompassing molecular graphs for property prediction, adapted from MoleculeNet as part of the Open Graph Benchmark. It is designed for a binary classification task involving 128 distinct molecular properties.
Use Cases
- Train graph neural networks to predict molecular properties using the 128 binary classification targets.
- Benchmark multi-task learning models on the 128 property prediction tasks.
- Analyze the correlation between different molecular properties within the set of 128 targets.
Strengths
- Part of the standardized Open Graph Benchmark for consistent model evaluation.
- Provides 128 distinct molecular property prediction tasks for multi-task learning.
Limitations
- The dataset is described as 'small', suggesting a limited number of molecular graphs.
- Not all 128 properties are present for all molecular graphs, indicating missing labels.
Provenance
- Source
- Adapted from MoleculeNet by teams at Stanford.
- Collection Method
- null
- Time Range
- null
- Freshness
- null
- Geography
- null