diff --git a/torchmdnet/datasets/qm9.py b/torchmdnet/datasets/qm9.py
index 6b97787d..92c16511 100644
--- a/torchmdnet/datasets/qm9.py
+++ b/torchmdnet/datasets/qm9.py
@@ -6,6 +6,7 @@
 from torch_geometric.transforms import Compose
 from torch_geometric.datasets import QM9 as QM9_geometric
 from torch_geometric.nn.models.schnet import qm9_target_dict
+from torch_geometric.data import Data
 
 
 class QM9(QM9_geometric):
@@ -25,17 +26,27 @@ def __init__(self, root, transform=None, label=None):
         else:
             transform = Compose([transform, self._filter_label])
 
-        super(QM9, self).__init__(root, transform=transform)
+        # Keep only pos, z and y in each sample
+        def pre_transform(x):
+            return Data(
+                pos=x.pos,
+                z=x.z,
+                y=x.y,
+            )
+
+        super(QM9, self).__init__(
+            root, transform=transform, pre_transform=pre_transform
+        )
 
     def get_atomref(self, max_z=100):
         """Atomic energy reference values for the :py:mod:`torchmdnet.priors.Atomref` prior.
 
-	  Args:
-	      max_z (int): Maximum atomic number
+        Args:
+            max_z (int): Maximum atomic number
 
-	  Returns:
-	      torch.Tensor: Atomic energy reference values for each element in the dataset.
-	"""
+        Returns:
+            torch.Tensor: Atomic energy reference values for each element in the dataset.
+        """
         atomref = self.atomref(self.label_idx)
         if atomref is None:
             return None