diff --git a/atom3d/datasets/__main__.py b/atom3d/datasets/__main__.py
index ad0688e..8ce7792 100644
--- a/atom3d/datasets/__main__.py
+++ b/atom3d/datasets/__main__.py
@@ -5,6 +5,7 @@
 import atom3d.datasets.datasets as da
 import atom3d.util.file as fi
 import atom3d.util.formats as fo
+import atom3d.util.rosetta as ro
 
 logger = logging.getLogger(__name__)
 
@@ -31,9 +32,13 @@ def main(input_dir, output_lmdb, filetype, score_path, serialization_format):
     else:
         fileext = filetype
     file_list = da.get_file_list(input_dir, fileext)
+    if score_path is not None:
+        transform = ro.Scores([score_path])
+    else:
+        transform = None
     logger.info(f'Found {len(file_list)} files.')
 
-    dataset = da.load_dataset(file_list, filetype)
+    dataset = da.load_dataset(file_list, filetype, transform=transform)
     da.make_lmdb_dataset(
         dataset, output_lmdb, serialization_format=serialization_format)
 
diff --git a/atom3d/util/rosetta.py b/atom3d/util/rosetta.py
index 156ceed..5f8cd43 100644
--- a/atom3d/util/rosetta.py
+++ b/atom3d/util/rosetta.py
@@ -23,12 +23,14 @@ def __init__(self, file_list):
         file_list = [Path(x).absolute() for x in file_list]
         for silent_file in file_list:
             key = self._key_from_silent_file(silent_file)
+            if len(file_list) == 1:
+                key = 'all'
             self._scores[key] = self._parse_scores(silent_file)
 
         self._scores = pd.concat(self._scores).sort_index()
 
     def _parse_scores(self, silent_file):
-        grep_cmd = f"grep ^SCORE: {silent_file}"
+        grep_cmd = f"grep ^SCORE: '{silent_file}'"
         out = subprocess.Popen(
             grep_cmd, stdout=subprocess.PIPE, stderr=subprocess.PIPE,
             cwd=os.getcwd(), shell=True)
@@ -48,7 +50,11 @@ def _lookup_helper(self, key):
         # If there are multiple rows matching key, return only the first one.
         # Sometime pandas return single row pd.DataFrame, so we use .squeeze()
         # to ensure it always return a pd.Series.
-        return self._scores.loc[key].head(1).astype(np.float64).squeeze().to_dict()
+        tmp = self._scores.loc[key]
+        if type(tmp) == pd.DataFrame:
+            tmp = tmp.head(1)
+
+        return tmp.astype(np.float64).squeeze().to_dict()
 
     def _lookup(self, file_path):
         file_path = Path(file_path)
@@ -61,9 +67,16 @@ def _lookup(self, file_path):
         key = (file_path.parent.parent.stem, file_path.stem)
         if key in self._scores.index:
             return key, self._lookup_helper(key)
+        if len(self._scores.index.get_level_values(0).unique()):
+            key = ('all', file_path.name)
+            result = self._lookup_helper(key)
+            key = (file_path.stem.split('_')[0],
+                   '_'.join(file_path.stem.split('_')[1:]))
+            return (key, result)
+
         return file_path.parent.stem, None
 
-    def __call__(self, x, error_if_missing=False):
+    def __call__(self, x, error_if_missing=True):
         key, x['scores'] = self._lookup(x['file_path'])
         x['id'] = str(key)
         if x['scores'] is None and error_if_missing: