diff --git a/src/openqdc/datasets/base.py b/src/openqdc/datasets/base.py
index 6c939e7..16dc892 100644
--- a/src/openqdc/datasets/base.py
+++ b/src/openqdc/datasets/base.py
@@ -132,13 +132,16 @@ def _precompute_E(self):
         s = np.array(self.data["atomic_inputs"][:, :2], dtype=int)
         s[:, 1] += IsolatedAtomEnergyFactory.max_charge
         matrixs = [matrix[s[:, 0], s[:, 1]] for matrix in self.__isolated_atom_energies__]
-        matrixs = [np.split(matrix, splits_idx)[:-1] for matrix in matrixs]
+        # matrixs = [np.split(matrix, splits_idx)[:-1] for matrix in matrixs]
         converted_energy_data = self.convert_energy(self.data["energies"])
         # calculation per molecule formation energy statistics
-        e = []
-        for i in range(len(self.__energy_methods__)):
-            e.append(converted_energy_data[:, i] - np.array(list(map(lambda x: x.sum(), matrixs[i]))))
-        E = np.array(e).T
+        n = len(self.__energy_methods__)
+        E = []
+        for i, matrix in enumerate(matrixs):
+            c = np.cumsum(np.append([0], matrix))[splits_idx]
+            c[1:] = c[1:] - c[:-1]
+            E.append(converted_energy_data[:, i] -c)
+        E = np.array(E).T
         formation_E_mean = np.nanmean(E, axis=0)
         formation_E_std = np.nanstd(E, axis=0)
         total_E_mean = np.nanmean(converted_energy_data, axis=0)