Update README.md

README.md

@@ -1,3 +1,78 @@
----
-license: cc-by-nc-4.0
----
+---
+tags:
+- molecular-property-prediction
+- out-of-distribution
+- QM9
+- graph-neural-networks
+- SCOPE-BENCH
+license: cc-by-nc-4.0
+pretty_name: SCOPE-BENCH
+task_categories:
+- regression
+- feature-extraction
+size_categories:
+- 100K<n<1M
+language:
+- en
+---
+
+# SCOPE-BENCH: Scaffold-Cluster Out-Of-Distribution Performance Evaluation Benchmark
+
+SCOPE-BENCH is a rigorous out-of-distribution (OOD) benchmark for molecular property prediction. Unlike conventional scaffold splits, SCOPE-BENCH creates structurally disjoint source and target domains by clustering molecules based on physicochemical descriptors, blocking shortcut learning, and revealing true extrapolation abilities.
+
+## Dataset Splits
+
+- Source (training): a0.csv, a1.csv, a6.csv, a8.csv, a9.csv, a11.csv (6 clusters, 94,562 samples)
+- Validation: a10.csv (1 cluster, 18,326 samples)
+- Target (test): a2.csv, a3.csv, a4.csv, a5.csv, a7.csv (5 clusters, 19,894 samples)
+- Fine-grained tasks: scaffold_datasets1/*.csv (16 independent zero-shot tasks, >=200 samples each)
+
+## Molecular Properties
+
+Each CSV file contains the following columns:
+- smiles: SMILES string
+- lumo: Lowest Unoccupied Molecular Orbital (eV)
+- homo: Highest Occupied Molecular Orbital (eV)
+- gap: HOMO-LUMO gap (eV)
+
+## Key Results from the Paper
+
+- SOTA models fail: ViSNet, ETNN, GotenNet suffer 4.1x average MAE increase on SCOPE-BENCH vs. standard scaffold split.
+- SOMA framework achieves 9.4% average improvement over supervised fine-tuning (ViSNet +7.5%, ETNN +11.2%, GotenNet +3.9-5.9%).
+
+## File Structure
+
+SCOPE-BENCH/
+├── a0.csv ... a11.csv          # 12 cluster files
+├── scaffold_datasets1/         # 16 independent target scaffolds
+└── README.md
+
+## Usage
+
+import pandas as pd
+
+# Load a source cluster
+df = pd.read_csv("https://hf-mirror.com/datasets/linzhuohao/SCOPE-BENCH/resolve/main/a0.csv")
+print(df.head())
+
+# Load a fine-grained task
+task_df = pd.read_csv("https://hf-mirror.com/datasets/linzhuohao/SCOPE-BENCH/resolve/main/scaffold_datasets1/a001.csv")
+
+## Citation
+
+If you use this dataset, please cite the NeurIPS 2026 paper:
+
+@article{lin2026scope,
+  title={Rethinking Molecular OOD Generalization via Target-Aware Source Selection},
+  author={Lin, Zhuohao and Hu, Wenbin},
+  journal={Advances in Neural Information Processing Systems (NeurIPS)},
+  year={2026}
+}
+
+## License
+
+This dataset is released under CC BY-NC 4.0.
+
+## Acknowledgements
+
+This dataset is built upon the QM9 dataset. We thank the original authors for making their data publicly available.