Add pipeline dataset card

README.md

@@ -1,19 +1,47 @@
 ---
-dataset_info:
-  features:
-  - name: smiles
-    dtype: string
-  - name: y
-    list: float64
-  splits:
-  - name: train
-    num_bytes: 17863997
-    num_examples: 247449
-  download_size: 11373798
-  dataset_size: 17863997
-configs:
-- config_name: default
-  data_files:
-  - split: train
-    path: data/train-*
+license: other
+pretty_name: zinc_clean
+tags:
+- chemistry
+- molecules
+- graph-generation
+- flow-matching
+size_categories:
+- 100K<n<1M
 ---
+
+# nico8771/zinc_clean — cleaned ZINC-250k
+
+Each row is a molecule as **canonical SMILES** plus RDKit-recomputed targets. Charged
+groups (`N+`/`O-`) are preserved. 
+
+> Source: torch_molecule ZINC-250k (HuggingFace). Code:
+> <https://github.com/Nico-Conti/flow-matching-molecules> (`dataset/`).
+
+## Schema
+
+| column | type | description |
+|---|---|---|
+| `smiles` | string | canonical, kekulizable, single-fragment SMILES (post-sanitize) |
+| `y` | list[float] | RDKit targets, columns = `logP`, `qed`, `SAS` |
+
+## Pipeline
+
+1. **Parse** with RDKit; unparseable dropped.
+2. **Standardize** — remove stereochemistry, sanitize (charges preserved; `N+`/`O-` kept).
+3. **Kekulize** over atom vocab (`C`, `N`, `O`, `F`, `P`, `S`, `Cl`, `Br`, `I`, `N+`, `O-`); atoms outside the vocab dropped.
+4. **Round-trip check** — `smiles -> (X, E) -> mol -> smiles` must reproduce the
+   canonical molecule as a single fragment (**filter applied**).
+
+Bonds use 4 classes (none / single / double / triple). Targets
+(`logP`, `qed`, `SAS`) are recomputed from the canonical SMILES with RDKit.
+
+### Drop / keep counts (this build)
+
+| outcome | count |
+|---|---|
+| `drop_roundtrip` | 1 |
+| `drop_vocab` | 2,005 |
+| `kept` | 247,449 |
+
+Kept: **247,449** molecules.