Report Biolink classes with their real case in /get_semantic_types

src/nodenorm/handlers/semantic_types.py

@@ -28,10 +28,20 @@ async def get(self) -> dict:
             )
             raise network_error from gen_exc
 
+        # The aggregation's bucket keys are the indexed keyword values of `type`, which the
+        # index mapping lowercases ("biolink:smallmolecule"), unlike the `_source` that
+        # /get_normalized_nodes reads. Don't add the keys directly: look each one up in the
+        # Biolink model (bmt matches case-insensitively) and report its `class_uri`
+        # ("biolink:SmallMolecule"). A key the model doesn't know is reported as is.
+        #
+        # These lookups don't need memoizing here: bmt's Toolkit wraps get_element() and
+        # get_ancestors() in functools.lru_cache (1024 entries, more than the model's
+        # classes), so after the first request the loop costs about a millisecond.
         semantic_types = set()
         for bucket in type_aggregation_result.body["aggregations"]["unique_types"]["buckets"]:
             biolink_type = bucket["key"]
-            semantic_types.add(biolink_type)
+            element = toolkit.get_element(biolink_type)
+            semantic_types.add(element["class_uri"] if element else biolink_type)
             for ancestor in toolkit.get_ancestors(biolink_type):
                 semantic_types.add(toolkit.get_element(ancestor)["class_uri"])
 

tests/test_openapi.py

@@ -271,7 +271,8 @@ def __init__(self):
 
     async def search(self, **kwargs):
         self.calls.append(kwargs)
-        return SimpleNamespace(body={"aggregations": {"unique_types": {"buckets": [{"key": "biolink:Disease"}]}}})
+        # Real indices lowercase the `type` keyword, so aggregation keys come back lowercased.
+        return SimpleNamespace(body={"aggregations": {"unique_types": {"buckets": [{"key": "biolink:disease"}]}}})
 
     async def msearch(self, **kwargs):
         self.calls.append(kwargs)
@@ -356,7 +357,8 @@ def test_status_matches_current_metadata_schema(self):
 
 
 class TestSemanticTypes(AsyncHTTPTestCase):
-    """Regression: the handler lower-cased every class ("biolink:namedthing")."""
+    """Regression: the handler lower-cased every class ("biolink:namedthing"), and then
+    reported the lowercased aggregation key next to its capitalised class_uri."""
 
     def get_app(self) -> tornado.web.Application:
         app = tornado.web.Application([(r"/get_semantic_types", SemanticTypeHandler)])
@@ -374,6 +376,7 @@ def test_types_are_biolink_class_curies(self):
         types = body["semantic_types"]["types"]
         assert "biolink:Disease" in types and "biolink:NamedThing" in types
         assert all(re.fullmatch(r"biolink:[A-Z][A-Za-z]*", t) for t in types), types
+        assert len({t.lower() for t in types}) == len(types), types
 
 
 class UnreachableElasticsearch: