[Dataset] Add BigSolDB solubility dataset

[levakrasnovs]

Description

Adds the BigSolDB dataset (torch_geometric.datasets.BigSolDB), a large-scale experimental solubility database covering organic compounds in diverse solvents across a wide temperature range.

Paper: BigSolDB 2.0, Scientific Data 2025
Zenodo: https://doi.org/10.5281/zenodo.18552681

Stats

• 112,465 experimental solubility values
• 1,525 unique organic compounds
• 218 solvents (water + organic)
• Temperature range: 243-425 K
• Extracted from 1,687 peer-reviewed articles

Why BigSolDB?

BigSolDB is significantly larger and more diverse than the existing AQSOL dataset
(9,800 aqueous-only datapoints). It covers 218 solvents vs. water-only, making it
suitable for multi-solvent solubility prediction - an important task in drug
formulation, synthesis, and crystallization.

New files

• torch_geometric/datasets/bigsoldb.py — dataset class
• test/datasets/test_bigsoldb.py — unit tests (8 tests)

Notes

• No official train/val/test split is provided (unlike AQSOL). Users are encouraged
  to apply scaffold-based or cold-solvent splits depending on their evaluation protocol.
• Solvent graph is stored as x_solvent, edge_index_solvent, edge_attr_solvent.
• Use follow_batch=['x_solvent'] in DataLoader for correct solvent pooling.

Impact

• 3,800+ downloads on Zenodo across all versions
• Cited in peer-reviewed works across top venues:

1. Attia et al., Nature Communications 2025 (https://doi.org/10.1038/s41467-025-62717-7)
2. Al Ibrahim et al., JACS 2025 (https://doi.org/10.1021/jacs.5c13746)
3. Ramani et al., JCTC 2024 (https://doi.org/10.1021/acs.jctc.4c00382)
4. Krzyżanowski et al., Digital Discovery 2025 (https://doi.org/10.1039/D5DD00134J)
5. Guo et al., JCIM 2025 (https://doi.org/10.1021/acs.jcim.5c00781)
6. Gopichand et al., ACS Omega 2025 (https://doi.org/10.1021/acsomega.5c13630)
7. Broadbent et al., ICLR 2026 Workshop (https://openreview.net/forum?id=znY51c4s04)
8. Pimonova et al., Digital Discovery 2026 (https://doi.org/10.1039/D5DD00443H)
9. Fan et al., Digital Discovery 2026 (https://doi.org/10.1039/D5DD00407A)
10. Da Vià et al., Org. Process Res. Dev. 2025 (https://doi.org/10.1021/acs.oprd.4c00384)
11. Gomes et al., JCIM 2026 (https://doi.org/10.1021/acs.jcim.6c00606)
12. Wang et al., Digital Discovery 2026 (https://doi.org/10.1039/D5DD00456J)