diff --git a/notebooks/structure_prediction.ipynb b/notebooks/structure_prediction.ipynb
index ae68613..7fd067d 100644
--- a/notebooks/structure_prediction.ipynb
+++ b/notebooks/structure_prediction.ipynb
@@ -80,7 +80,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "7fd300b3f2364146bc2b1066cf3e3f06",
+       "model_id": "675085da034f4139976b9f860b5670e8",
        "version_major": 2,
        "version_minor": 0
       },
@@ -255,31 +255,31 @@
        "      <th>0</th>\n",
        "      <td>x00011-1</td>\n",
        "      <td>NS(=O)(=O)C=1C=CC=2CCCC2C1</td>\n",
-       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7c7374ab4740&gt;</td>\n",
+       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7b69c12ac430&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>1</th>\n",
        "      <td>x00035-1</td>\n",
        "      <td>NC1=NC=2C=CC=CC2S1</td>\n",
-       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7c7374ab47b0&gt;</td>\n",
+       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7b69c12ac2e0&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>2</th>\n",
        "      <td>x00046-1</td>\n",
        "      <td>NC1=NC=2C=C(Cl)C=CC2O1</td>\n",
-       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7c7374ab4820&gt;</td>\n",
+       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7b69c12ac740&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>3</th>\n",
        "      <td>x00052-1</td>\n",
        "      <td>NC(=O)C=1C=CC=C(Cl)C1</td>\n",
-       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7c7374ab4890&gt;</td>\n",
+       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7b69c12ac7b0&gt;</td>\n",
        "    </tr>\n",
        "    <tr>\n",
        "      <th>4</th>\n",
        "      <td>x00086-1</td>\n",
        "      <td>ClC=1C=CC=2OC(=O)NC2C1</td>\n",
-       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7c7374ab4900&gt;</td>\n",
+       "      <td>&lt;rdkit.Chem.rdchem.Mol object at 0x7b69c12ac820&gt;</td>\n",
        "    </tr>\n",
        "  </tbody>\n",
        "</table>\n",
@@ -294,11 +294,11 @@
        "4  x00086-1      ClC=1C=CC=2OC(=O)NC2C1   \n",
        "\n",
        "                                                mol  \n",
-       "0  <rdkit.Chem.rdchem.Mol object at 0x7c7374ab4740>  \n",
-       "1  <rdkit.Chem.rdchem.Mol object at 0x7c7374ab47b0>  \n",
-       "2  <rdkit.Chem.rdchem.Mol object at 0x7c7374ab4820>  \n",
-       "3  <rdkit.Chem.rdchem.Mol object at 0x7c7374ab4890>  \n",
-       "4  <rdkit.Chem.rdchem.Mol object at 0x7c7374ab4900>  "
+       "0  <rdkit.Chem.rdchem.Mol object at 0x7b69c12ac430>  \n",
+       "1  <rdkit.Chem.rdchem.Mol object at 0x7b69c12ac2e0>  \n",
+       "2  <rdkit.Chem.rdchem.Mol object at 0x7b69c12ac740>  \n",
+       "3  <rdkit.Chem.rdchem.Mol object at 0x7b69c12ac7b0>  \n",
+       "4  <rdkit.Chem.rdchem.Mol object at 0x7b69c12ac820>  "
       ]
      },
      "execution_count": 5,
@@ -683,7 +683,7 @@
     {
      "data": {
       "application/vnd.jupyter.widget-view+json": {
-       "model_id": "71197becbdd541ef86a336453c28aac4",
+       "model_id": "067c76e2aa9842599d3e2d8b484e510b",
        "version_major": 2,
        "version_minor": 0
       },
@@ -870,9 +870,12 @@
     "\n",
     "print(f\"Created zip archive: {zip_path} ({len(pdb_files)} files)\")\n",
     "\n",
+    "expected_smiles_dict = dict(zip(structure_df[\"id\"], structure_df[\"SMILES\"]))\n",
+    "\n",
     "is_valid, validation_errors = validate_structure_submission(\n",
     "    zip_path,\n",
     "    expected_ids=set(structure_df[\"id\"]),\n",
+    "    expected_ligand_smiles=expected_smiles_dict,\n",
     "    require_lig_resname=True,\n",
     ")\n",
     "\n",
diff --git a/validation/structure_validation.py b/validation/structure_validation.py
index bdbf9cb..3fdd651 100644
--- a/validation/structure_validation.py
+++ b/validation/structure_validation.py
@@ -2,16 +2,26 @@
 
 import zipfile
 import tempfile
+from collections import Counter
 import MDAnalysis as mda
 from pathlib import Path
 from typing import Union
+from rdkit import Chem
 
 STRUCTURE_DATASET_SIZE = 78
 
+
+def _heavy_atom_counts_from_smiles(smiles: str) -> Counter | None:
+    mol = Chem.MolFromSmiles(smiles)
+    if mol is None:
+        return None
+    return Counter(atom.GetSymbol() for atom in mol.GetAtoms())
+
+
 def validate_structure_submission(
     structure_predictions_file: Union[str, Path],
     expected_ids: set[str] | None = None,
-    expected_ligand_smiles: set[str] | None = None,
+    expected_ligand_smiles: dict[str, str] | None = None,
     require_lig_resname: bool = True,
 ) -> tuple[bool, list[str]]:
 
@@ -68,6 +78,28 @@ def validate_structure_submission(
                             if len(u.segments) > 2:
                                 errors.append(f"{name}: Found {len(u.segments)} chains, expected 2 or fewer")
 
+                            # 4. Check ligand matches expected SMILES by heavy-atom composition
+                            if expected_ligand_smiles is not None:
+                                pdb_id = Path(name).stem
+                                expected_smi = expected_ligand_smiles.get(pdb_id)
+                                if expected_smi is not None:
+                                    expected_counts = _heavy_atom_counts_from_smiles(expected_smi)
+                                    if expected_counts is None:
+                                        errors.append(
+                                            f"{name}: Could not parse expected SMILES for '{pdb_id}'"
+                                        )
+                                    else:
+                                        obs_counts = Counter(
+                                            elem for elem in ligands.elements
+                                            if elem.strip() not in ("H", "h", "")
+                                        )
+                                        if obs_counts != expected_counts:
+                                            errors.append(
+                                                f"{name}: Ligand heavy-atom composition mismatch. "
+                                                f"Expected {dict(sorted(expected_counts.items()))}, "
+                                                f"got {dict(sorted(obs_counts.items()))}"
+                                            )
+
                         except Exception as e:
                             errors.append(f"{name}: MDAnalysis failed to parse file: {e}")