diff --git a/chebai/preprocessing/datasets/chebi.py b/chebai/preprocessing/datasets/chebi.py
index 1df144d9..9fa1c1c7 100644
--- a/chebai/preprocessing/datasets/chebi.py
+++ b/chebai/preprocessing/datasets/chebi.py
@@ -301,9 +301,7 @@ def _graph_to_raw_dataset(self, g: nx.DiGraph) -> pd.DataFrame:
         data = pd.DataFrame(data)
         data = data[~data["SMILES"].isnull()]
         data = data[[name not in CHEBI_BLACKLIST for name, _ in data.iterrows()]]
-        # This filters the DataFrame to include only the rows where at least one value in the row from 4th column
-        # onwards is True/non-zero.
-        data = data[data.iloc[:, self._LABELS_START_IDX :].any(axis=1)]
+
         return data
 
     # ------------------------------ Phase: Setup data -----------------------------------
@@ -702,18 +700,24 @@ class ChEBIOverXPartial(ChEBIOverX):
         top_class_id (int): The ID of the top class from which to extract subclasses.
     """
 
-    def __init__(self, top_class_id: int, **kwargs):
+    def __init__(self, top_class_id: int, external_data_ratio: float, **kwargs):
         """
         Initializes the ChEBIOverXPartial dataset.
 
         Args:
             top_class_id (int): The ID of the top class from which to extract subclasses.
             **kwargs: Additional keyword arguments passed to the superclass initializer.
+            external_data_ratio (float): How much external data (i.e., samples where top_class_id
+            is no positive label) to include in the dataset. 0 means no external data, 1 means
+            the maximum amount (i.e., the complete ChEBI dataset).
         """
         if "top_class_id" not in kwargs:
             kwargs["top_class_id"] = top_class_id
+        if "external_data_ratio" not in kwargs:
+            kwargs["external_data_ratio"] = external_data_ratio
 
         self.top_class_id: int = top_class_id
+        self.external_data_ratio: float = external_data_ratio
         super().__init__(**kwargs)
 
     @property
@@ -727,7 +731,7 @@ def processed_dir_main(self) -> str:
         return os.path.join(
             self.base_dir,
             self._name,
-            f"partial_{self.top_class_id}",
+            f"partial_{self.top_class_id}_ext_ratio_{self.external_data_ratio:.2f}",
             "processed",
         )
 
@@ -746,9 +750,53 @@ def _extract_class_hierarchy(self, chebi_path: str) -> nx.DiGraph:
             descendants of the top class ID.
         """
         g = super()._extract_class_hierarchy(chebi_path)
-        g = g.subgraph(list(g.successors(self.top_class_id)) + [self.top_class_id])
+        top_class_successors = list(g.successors(self.top_class_id)) + [
+            self.top_class_id
+        ]
+        external_nodes = list(set(n for n in g.nodes if n not in top_class_successors))
+        if 0 < self.external_data_ratio < 1:
+            n_external_nodes = int(
+                len(top_class_successors)
+                * self.external_data_ratio
+                / (1 - self.external_data_ratio)
+            )
+            print(
+                f"Extracting {n_external_nodes} external nodes from the ChEBI dataset (ratio: {self.external_data_ratio:.2f})"
+            )
+            external_nodes = external_nodes[: int(n_external_nodes)]
+        elif self.external_data_ratio == 0:
+            external_nodes = []
+
+        g = g.subgraph(top_class_successors + external_nodes)
+        print(
+            f"Subgraph contains {len(g.nodes)} nodes, of which {len(top_class_successors)} are subclasses of the top class ID {self.top_class_id}."
+        )
         return g
 
+    def select_classes(self, g: nx.DiGraph, *args, **kwargs) -> List:
+        """Only selects classes that meet the threshold AND are subclasses of the top class ID (including itself)."""
+        smiles = nx.get_node_attributes(g, "smiles")
+        nodes = list(
+            sorted(
+                {
+                    node
+                    for node in g.nodes
+                    if sum(
+                        1 if smiles[s] is not None else 0 for s in g.successors(node)
+                    )
+                    >= self.THRESHOLD
+                    and (
+                        self.top_class_id in g.predecessors(node)
+                        or node == self.top_class_id
+                    )
+                }
+            )
+        )
+        filename = "classes.txt"
+        with open(os.path.join(self.processed_dir_main, filename), "wt") as fout:
+            fout.writelines(str(node) + "\n" for node in nodes)
+        return nodes
+
 
 class ChEBIOver50Partial(ChEBIOverXPartial, ChEBIOver50):
     """
@@ -842,7 +890,7 @@ def term_callback(doc: fastobo.term.TermFrame) -> Union[Dict, bool]:
 
 
 atom_index = (
-    "\*",
+    r"\*",
     "H",
     "He",
     "Li",
@@ -1473,3 +1521,15 @@ def term_callback(doc: fastobo.term.TermFrame) -> Union[Dict, bool]:
 ]
 
 JCI_500_COLUMNS_INT = [int(n.split(":")[-1]) for n in JCI_500_COLUMNS]
+
+if __name__ == "__main__":
+    data_module_05 = ChEBIOver50Partial(
+        chebi_version=241,
+        splits_file_path=os.path.join(
+            "data", "chebi_v241", "ChEBI50", "splits_80_10_10.csv"
+        ),
+        top_class_id=22712,
+        external_data_ratio=0.5,
+    )
+    data_module_05.prepare_data()
+    data_module_05.setup()
diff --git a/tests/unit/dataset_classes/testChebiOverXPartial.py b/tests/unit/dataset_classes/testChebiOverXPartial.py
index 76584ebf..0e263335 100644
--- a/tests/unit/dataset_classes/testChebiOverXPartial.py
+++ b/tests/unit/dataset_classes/testChebiOverXPartial.py
@@ -16,7 +16,9 @@ def setUpClass(cls, mock_makedirs) -> None:
         """
         Set up the ChEBIOverXPartial instance with a mock processed directory path and a test graph.
         """
-        cls.chebi_extractor = ChEBIOverXPartial(top_class_id=11111, chebi_version=231)
+        cls.chebi_extractor = ChEBIOverXPartial(
+            top_class_id=11111, external_data_ratio=0, chebi_version=231
+        )
         cls.test_graph = ChebiMockOntology.get_transitively_closed_graph()
 
     @patch(