feat: implement ESOL and ClinTox evaluation datasets

src/modernmolbert/eval/contributed_datasets.py

@@ -103,6 +103,63 @@ def load_example_activity_dataset(*, root: str | Path) -> EvalDataset:
     )
 
 
+def load_esol(*, root: str | Path) -> EvalDataset:
+    root = Path(root)
+    task_names = ["target"]
+
+    train = pd.read_csv(root / "train.csv")
+    valid_path = root / "valid.csv"
+    valid = pd.read_csv(valid_path) if valid_path.exists() else None
+    test = pd.read_csv(root / "test.csv")
+
+    train = _drop_missing_labels(train, task_names=task_names)
+    test = _drop_missing_labels(test, task_names=task_names)
+    if valid is not None:
+        valid = _drop_missing_labels(valid, task_names=task_names)
+
+    return make_eval_dataset_from_splits(
+        name="esol",
+        task_type="regression",
+        task_names=task_names,
+        train=train,
+        valid=valid,
+        test=test,
+        smiles_column="smiles",
+        metadata={"source": "MoleculeNet ESOL", "missing_label_policy": "Dropped"},
+    )
+
+
+def load_clintox(*, root: str | Path) -> EvalDataset:
+    root = Path(root)
+    task_names = ["FDA_APPROVED"]
+
+    train = pd.read_csv(root / "train.csv")
+    valid_path = root / "valid.csv"
+    valid = pd.read_csv(valid_path) if valid_path.exists() else None
+    test = pd.read_csv(root / "test.csv")
+
+    train = _drop_missing_labels(train, task_names=task_names)
+    test = _drop_missing_labels(test, task_names=task_names)
+    if valid is not None:
+        valid = _drop_missing_labels(valid, task_names=task_names)
+
+    _validate_binary_labels(train, task_name="FDA_APPROVED", split_name="train")
+    _validate_binary_labels(test, task_name="FDA_APPROVED", split_name="test")
+    if valid is not None:
+        _validate_binary_labels(valid, task_name="FDA_APPROVED", split_name="valid")
+
+    return make_eval_dataset_from_splits(
+        name="clintox",
+        task_type="classification",
+        task_names=task_names,
+        train=train,
+        valid=valid,
+        test=test,
+        smiles_column="smiles",
+        metadata={"source": "MoleculeNet ClinTox", "missing_label_policy": "Dropped"},
+    )
+
+
 def register_contributed_datasets() -> None:
     """Register project-maintained contributed datasets.
 
@@ -126,4 +183,30 @@ def register_contributed_datasets() -> None:
     #     )
     # )
 
+    register_dataset(
+        DatasetSpec(
+            name="esol",
+            task_type="regression",
+            task_names=("target",),
+            loader=load_esol,
+            description="ESOL water solubility regression from MoleculeNet.",
+            source="https://moleculenet.org/datasets-1",
+            citation="Wu et al. MoleculeNet: a benchmark for molecular machine learning. Chemical Science (2018).",
+            license="MIT",
+        )
+    )
+
+    register_dataset(
+        DatasetSpec(
+            name="clintox",
+            task_type="classification",
+            task_names=("FDA_APPROVED",),
+            loader=load_clintox,
+            description="ClinTox FDA approval classification from MoleculeNet.",
+            source="https://moleculenet.org/datasets-1",
+            citation="Wu et al. MoleculeNet: a benchmark for molecular machine learning. Chemical Science (2018).",
+            license="MIT",
+        )
+    )
+
     return None

tests/test_eval_clintox.py

@@ -0,0 +1,21 @@
+from pathlib import Path
+import pandas as pd
+from modernmolbert.eval.contributed_datasets import load_clintox
+
+
+def test_load_clintox(tmp_path: Path) -> None:
+    root = tmp_path / "clintox"
+    root.mkdir()
+
+    pd.DataFrame({"smiles": ["CCO", "CCN", "CCC"], "FDA_APPROVED": [0, 1, None]}).to_csv(
+        root / "train.csv", index=False
+    )
+    pd.DataFrame({"smiles": ["CCCl", "CCBr"], "FDA_APPROVED": [0, 1]}).to_csv(
+        root / "test.csv", index=False
+    )
+
+    dataset = load_clintox(root=root)
+    dataset.check()
+
+    assert dataset.name == "clintox"
+    assert len(dataset.train) == 2

tests/test_eval_esol.py

@@ -0,0 +1,21 @@
+from pathlib import Path
+import pandas as pd
+from modernmolbert.eval.contributed_datasets import load_esol
+
+
+def test_load_esol(tmp_path: Path) -> None:
+    root = tmp_path / "esol"
+    root.mkdir()
+
+    pd.DataFrame({"smiles": ["CCO", "CCN", "CCC"], "target": [1.2, -0.5, None]}).to_csv(
+        root / "train.csv", index=False
+    )
+    pd.DataFrame({"smiles": ["CCCl", "CCBr"], "target": [0.1, 0.2]}).to_csv(
+        root / "test.csv", index=False
+    )
+
+    dataset = load_esol(root=root)
+    dataset.check()
+
+    assert dataset.name == "esol"
+    assert len(dataset.train) == 2