SynPlanner-data
Data repository for
SynPlanner — an open-source tool for retrosynthetic planning.
Quick start
1pip install SynPlanner
2synplan download_preset --preset synplanner-article --save_to synplan_data
Repository structure
SynPlanner-data/
├── policy/ Reaction rules + policy network weights
│ └── {architecture}/
│ └── {rules_version}/
│ ├── reaction_rules.tsv
│ ├── pipeline.yaml
│ └── {weights_version}/
│ ├── ranking_policy.ckpt
│ └── filtering_policy.ckpt
│
├── value/ Value network weights
│ └── {architecture}/
│ └── {version}/
│ ├── value_network.ckpt
│ └── meta.yaml
│
├── building_blocks/ Building block sets
│ └── {name}/
│ ├── building_blocks.tsv
│ └── meta.yaml
│
├── reaction_data/ Reaction data pipeline (raw → standardized → filtered)
│ └── {source}/
│ ├── raw/
│ ├── standardized/{YYYY-MM-DD}/
│ └── filtered/{YYYY-MM-DD}/
│
├── training_data/ Per-network training inputs
│ ├── ranking_policy/{YYYY-MM-DD}/
│ ├── filtering_policy/{YYYY-MM-DD}/
│ └── value_network/{YYYY-MM-DD}/
│
├── presets/ Ready-to-use preset definitions
│ └── {name}.yaml
│
└── benchmarks/ Benchmark target sets
└── sascore/
Versioning
Model components (policy/, value/, building_blocks/) are versioned by name:
- Architecture family:
policy/supervised_gcn/, policy/transformer/
- Rules version:
policy/supervised_gcn/v1/ — new extraction = new version
- Weights version:
policy/supervised_gcn/v1/v1/ — retrained weights = new inner version
- Value network:
value/supervised_gcn/v1/ — retrained = v2/
Pipeline data (reaction_data/, training_data/) is versioned by date (YYYY-MM-DD).
New processing protocol = new date directory alongside old ones.
Presets
Presets are YAML files that map logical names to exact file paths:
1name: synplanner-article
2files:
3 reaction_rules: policy/supervised_gcn/v1/reaction_rules.tsv
4 ranking_policy: policy/supervised_gcn/v1/v1/ranking_policy.ckpt
5 filtering_policy: policy/supervised_gcn/v1/v1/filtering_policy.ckpt
6 value_network: value/supervised_gcn/v1/value_network.ckpt
7 building_blocks: building_blocks/emolecules-salt-ln/building_blocks.tsv
Download a preset (Python):
1from synplan.utils.loading import download_preset
2
3paths = download_preset("synplanner-article", save_to="synplan_data")
Available data
policy/supervised_gcn/v1
GCN-based policy trained on filtered USPTO with 24k extracted reaction rules.
| File | Description | Size |
|---|
reaction_rules.tsv | Reaction rules in SMARTS format | 8.6 MB |
v1/ranking_policy.ckpt | Ranking policy network | 157 MB |
v1/filtering_policy.ckpt | Filtering policy network | 298 MB |
value/supervised_gcn/v1
GCN-based value network trained via self-learning on ChEMBL targets.
| File | Description | Size |
|---|
value_network.ckpt | Value network | 15 MB |
building_blocks/emolecules-salt-ln
Standardized building blocks from eMolecules, Sigma Aldrich, and Lancaster.
| File | Description | Size |
|---|
building_blocks.tsv | 186k molecules (SMILES + price columns) | 6.9 MB |
reaction_data/uspto
USPTO reaction data pipeline.
| File | Description | Size |
|---|
raw/uspto_full_mapped.smi.zip | Original USPTO (1.48M reactions) | 165 MB |
standardized/2024-12-31/ | Standardized reactions + config | to be added |
filtered/2024-12-31/ | Filtered reactions + config | to be added |
training_data
| Directory | Description | Size |
|---|
filtering_policy/2024-12-31/ | 600k ChEMBL + COCONUT molecules | 17 MB |
value_network/2024-12-31/ | 28k ChEMBL + COCONUT targets | 720 KB |
ranking_policy/2024-12-31/ | Config + README (uses reaction_data) | — |
benchmarks/sascore
700 target molecules split into 7 SAScore bins (1.5–8.5). Downloaded separately from presets.
Citation
If you use this data, please cite:
Akhmetshin, T.; Zankov, D.; Gantzer, P.; Babadeev, D.; Pinigina, A.; Madzhidov, T.; Varnek, A.
SynPlanner: An End-to-End Tool for Synthesis Planning.
J. Chem. Inf. Model. 2025,
65 (1), 15–21.
DOI:
10.1021/acs.jcim.4c02004
Links
