diff --git a/models/rfd3/src/rfd3/utils/inference.py b/models/rfd3/src/rfd3/utils/inference.py
index 2d04e149..406f2196 100644
--- a/models/rfd3/src/rfd3/utils/inference.py
+++ b/models/rfd3/src/rfd3/utils/inference.py
@@ -405,21 +405,27 @@ def infer_ori_from_hotspots(atom_array: struc.AtomArray):
 
     # We can only perform distance computations on atoms with non-NaN coordinates
     nan_coords_mask = np.any(np.isnan(atom_array.coord), axis=1)
-    non_nan_atom_array = atom_array[~nan_coords_mask]
+    motif_mask = atom_array.is_motif_atom_with_fixed_coord.astype(bool)
+    non_nan_motif_atom_array = atom_array[~nan_coords_mask & motif_mask]
+    if non_nan_motif_atom_array.array_length() == 0:
+        raise ValueError(
+            "infer_ori_from_hotspots requires at least one fixed motif atom "
+            "(is_motif_atom_with_fixed_coord=True) to compute nearby atoms COM."
+        )
 
     # Perform the distance computation
     # RFD2 used 10 Angstroms instead of 12, but was for residue-level hotspots
     DISTANCE_CUTOFF = 12.0
-    cell_list = struc.CellList(non_nan_atom_array, cell_size=DISTANCE_CUTOFF)
+    cell_list = struc.CellList(non_nan_motif_atom_array, cell_size=DISTANCE_CUTOFF)
     nearby_atoms_mask = get_atom_mask_from_cell_list(
         hotspot_atom_array.coord,
         cell_list,
-        len(non_nan_atom_array),
+        len(non_nan_motif_atom_array),
         cutoff=DISTANCE_CUTOFF,
     )  # (n_query, n_cell_list)
 
     nearby_atoms_mask = np.any(nearby_atoms_mask, axis=0)  # (n_cell_list,)
-    nearby_atoms_com = non_nan_atom_array.coord[nearby_atoms_mask].mean(axis=0)
+    nearby_atoms_com = non_nan_motif_atom_array.coord[nearby_atoms_mask].mean(axis=0)
 
     vector_from_core_to_hotspot = hotspot_com - nearby_atoms_com
     vector_from_core_to_hotspot = vector_from_core_to_hotspot / np.linalg.norm(
diff --git a/models/rfd3na/src/rfd3na/utils/inference.py b/models/rfd3na/src/rfd3na/utils/inference.py
index 61f82927..1838aedb 100644
--- a/models/rfd3na/src/rfd3na/utils/inference.py
+++ b/models/rfd3na/src/rfd3na/utils/inference.py
@@ -405,21 +405,26 @@ def infer_ori_from_hotspots(atom_array: struc.AtomArray):
 
     # We can only perform distance computations on atoms with non-NaN coordinates
     nan_coords_mask = np.any(np.isnan(atom_array.coord), axis=1)
-    non_nan_atom_array = atom_array[~nan_coords_mask]
-
+    motif_mask = atom_array.is_motif_atom_with_fixed_coord.astype(bool)
+    non_nan_motif_atom_array = atom_array[~nan_coords_mask & motif_mask]
+    if non_nan_motif_atom_array.array_length() == 0:
+        raise ValueError(
+            "infer_ori_from_hotspots requires at least one fixed motif atom "
+            "(is_motif_atom_with_fixed_coord=True) to compute nearby atoms COM."
+        )
     # Perform the distance computation
     # RFD2 used 10 Angstroms instead of 12, but was for residue-level hotspots
     DISTANCE_CUTOFF = 12.0
-    cell_list = struc.CellList(non_nan_atom_array, cell_size=DISTANCE_CUTOFF)
+    cell_list = struc.CellList(non_nan_motif_atom_array, cell_size=DISTANCE_CUTOFF)
     nearby_atoms_mask = get_atom_mask_from_cell_list(
         hotspot_atom_array.coord,
         cell_list,
-        len(non_nan_atom_array),
+        len(non_nan_motif_atom_array),
         cutoff=DISTANCE_CUTOFF,
     )  # (n_query, n_cell_list)
 
     nearby_atoms_mask = np.any(nearby_atoms_mask, axis=0)  # (n_cell_list,)
-    nearby_atoms_com = non_nan_atom_array.coord[nearby_atoms_mask].mean(axis=0)
+    nearby_atoms_com = non_nan_motif_atom_array.coord[nearby_atoms_mask].mean(axis=0)
 
     vector_from_core_to_hotspot = hotspot_com - nearby_atoms_com
     vector_from_core_to_hotspot = vector_from_core_to_hotspot / np.linalg.norm(