diff --git a/src/DockQ/DockQ.py b/src/DockQ/DockQ.py
index 98dc38f..44f1316 100755
--- a/src/DockQ/DockQ.py
+++ b/src/DockQ/DockQ.py
@@ -199,13 +199,8 @@ def calc_sym_corrected_lrmsd(
     aligned_sample_receptor, aligned_ref_receptor = get_aligned_residues(
         sample_receptor, ref_receptor, receptor_alignment
     )
-    # get a copy of each structure, then only keep backbone atoms
-    sample_interface_atoms, ref_interface_atoms = get_interface_atoms(
-        (receptor_interface, ()),
-        (aligned_sample_receptor, ()),
-        (aligned_ref_receptor, ()),
-        atom_types=BACKBONE_ATOMS,
-    )
+
+    sample_interface_atoms, ref_interface_atoms = subset_atoms(aligned_sample_receptor, aligned_ref_receptor, atom_types=BACKBONE_ATOMS, residue_subset=receptor_interface)
 
     sample_ligand_atoms_ids = [atom.id for atom in sample_ligand.get_atoms()]
     sample_ligand_atoms_ele = [atom.element for atom in sample_ligand.get_atoms()]
@@ -230,7 +225,7 @@ def calc_sym_corrected_lrmsd(
 
     # Set to align on receptor interface
     super_imposer = SVDSuperimposer()
-    super_imposer.set(ref_interface_atoms, sample_interface_atoms)
+    super_imposer.set(np.asarray(ref_interface_atoms), np.asarray(sample_interface_atoms))
     super_imposer.run()
     rot, tran = super_imposer.get_rotran()
 
@@ -323,13 +318,13 @@ def calc_DockQ(
         ref_res_distances,
         threshold=interface_threshold ** 2,
     )
-    # get a copy of each structure, then only keep backbone atoms
-    sample_interface_atoms, ref_interface_atoms = get_interface_atoms(
-        interacting_pairs,
-        (aligned_sample_1, aligned_sample_2),
-        (aligned_ref_1, aligned_ref_2),
-        atom_types=BACKBONE_ATOMS,
-    )
+    
+    sample_interface_atoms1, ref_interface_atoms1 = subset_atoms(aligned_sample_1, aligned_ref_1, atom_types=BACKBONE_ATOMS, residue_subset=interacting_pairs[0])
+    sample_interface_atoms2, ref_interface_atoms2 = subset_atoms(aligned_sample_2, aligned_ref_2, atom_types=BACKBONE_ATOMS, residue_subset=interacting_pairs[1])
+    
+    sample_interface_atoms = np.asarray(sample_interface_atoms1 + sample_interface_atoms2)
+    ref_interface_atoms = np.asarray(ref_interface_atoms1 + ref_interface_atoms2)
+
     super_imposer = SVDSuperimposer()
     super_imposer.set(sample_interface_atoms, ref_interface_atoms)
     super_imposer.run()
@@ -354,19 +349,19 @@ def calc_DockQ(
         else ("ligand", "receptor")
     )
 
-    receptor_atoms_native, receptor_atoms_sample = get_atoms_per_residue(
-            receptor_chains, what="receptor", atom_types=BACKBONE_ATOMS
+    receptor_atoms_native, receptor_atoms_sample = subset_atoms(
+            receptor_chains[0], receptor_chains[1], atom_types=BACKBONE_ATOMS
         )
-    ligand_atoms_native, ligand_atoms_sample = get_atoms_per_residue(ligand_chains, what="ligand", atom_types=BACKBONE_ATOMS)
+    ligand_atoms_native, ligand_atoms_sample = subset_atoms(ligand_chains[0], ligand_chains[1], atom_types=BACKBONE_ATOMS)
     # Set to align on receptor
-    super_imposer.set(receptor_atoms_native, receptor_atoms_sample)
+    super_imposer.set(np.asarray(receptor_atoms_native), np.asarray(receptor_atoms_sample))
     super_imposer.run()
 
     rot, tran = super_imposer.get_rotran()
-    rotated_sample_atoms = np.dot(ligand_atoms_sample, rot) + tran
+    rotated_sample_atoms = np.dot(np.asarray(ligand_atoms_sample), rot) + tran
 
     lrms = super_imposer._rms(
-        ligand_atoms_native, rotated_sample_atoms
+        np.asarray(ligand_atoms_native), rotated_sample_atoms
     )  # using the private _rms function which does not superimpose
 
     info = {}
@@ -563,6 +558,40 @@ def get_interface_atoms(
     return np.asarray(mod_interface), np.asarray(ref_interface)
 
 
+@lru_cache
+def subset_atoms(
+    mod_chain,
+    ref_chain,
+    atom_types,
+    residue_subset=None,
+):
+    mod_atoms = []
+    ref_atoms = []
+
+    mod_residues = [res for res in mod_chain]
+    ref_residues = [res for res in ref_chain]
+    
+    # remove duplicate residues
+    residue_subset = set(residue_subset) if residue_subset else range(len(mod_residues))
+
+    for i in residue_subset:
+        mod_res_atoms = list(mod_residues[i].get_atoms())
+        ref_res_atoms = list(ref_residues[i].get_atoms())
+        mod_res_atoms_ids = [atom.id for atom in mod_res_atoms]
+        ref_res_atoms_ids = [atom.id for atom in ref_res_atoms]
+
+        for atom_type in atom_types:
+            try:
+                mod_i = mod_res_atoms_ids.index(atom_type)
+                ref_i = ref_res_atoms_ids.index(atom_type)
+                mod_atoms += [mod_res_atoms[mod_i].coord]
+                ref_atoms += [ref_res_atoms[ref_i].coord]
+            except:
+                continue
+
+    return mod_atoms, ref_atoms
+
+
 @lru_cache
 def run_on_chains(
     model_chains,