Merge pull request #893 from uclahs-cds/czhu-fix-call-variant

Fix callVariant: variant bubble aligning issue

Author: Chenghao (Trevor) Zhu

CHANGELOG.md

@@ -18,6 +18,8 @@ This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.htm
 
 - Fixed issue that `callVariant` fails on transcripts with SEC very close to the start codon.
 
+- Fixed issue with extremely long run-time with complex alt splice events. #892
+
 ## [1.4.3] - 2025-01-18
 
 ### Fixed

docs/files/fuzz_test_history.tsv

@@ -32,3 +32,6 @@ v1.2.1	e7a4096	2023-11-28	comprehensive	522929	0	0	0:00:00.378236	15.50100807513
 v1.4.2	c2da21c	2025-01-19	snv	270	0	0	0:00:00.170374	0.37595620821363007	0:00:56.681483	105.92716856631384
 v1.4.2	c2da21c	2025-01-19	indel	275	0	0	0:00:00.170908	0.3764300020522951	0:00:40.836362	93.63957347416282
 v1.4.2	c2da21c	2025-01-19	comprehensive	471	0	0	0:00:00.352437	0.6980351919379612	0:00:38.006367	140.52343056870941
+v1.4.3	64a1cd7	2025-02-12	snv	3615	0	0	0:00:00.164877	0.5071353395324305	0:00:57.359060	118.20011598924994
+v1.4.3	64a1cd7	2025-02-12	indel	3623	0	0	0:00:00.228542	1.0644126008878305	0:00:39.258896	93.83931730856402
+v1.4.3	64a1cd7	2025-02-12	comprehensive	6985	0	0	0:00:00.372239	0.8090526920972508	0:00:35.994138	142.17524693622445

moPepGen/svgraph/ThreeFrameTVG.py

@@ -1418,17 +1418,19 @@ def is_candidate_out_node(x:TVGNode, y:TVGNode):
             # the input node itself.
             return node, set()
 
-        visited:Set[TVGNode] = {node}
+        visited:Dict[str, TVGNode] = {node.id: node}
         # When a new farthest node is found, the downstream nodes of the old
         # farthest node are put into this `exceptions` container, so they
         # can be visited again.
         exceptions:Set[TVGNode] = set()
+        non_members:Set[str] = set()
         while queue:
             cur:TVGNode = queue.popleft()
             if cur is None:
                 continue
 
             if cur.reading_frame_index != node.reading_frame_index:
+                non_members.add(cur.id)
                 continue
 
             if subgraph_checker:
@@ -1437,7 +1439,7 @@ def is_candidate_out_node(x:TVGNode, y:TVGNode):
                     subgraph_checker = False
 
             visited_len_before = len(visited)
-            visited.add(cur)
+            visited[cur.id] = cur
             visited_len_after = len(visited)
             if visited_len_before == visited_len_after:
                 if cur is farthest and cur is not node:
@@ -1497,7 +1499,8 @@ def is_candidate_out_node(x:TVGNode, y:TVGNode):
             queue.append(farthest)
             exceptions.add(cur)
             continue
-        return farthest, visited
+        members = {v for k,v in visited.items() if k not in non_members}
+        return farthest, members
 
     def first_node_is_smaller(self, first:TVGNode, second:TVGNode) -> bool:
         """ Check if the first node is larger """
@@ -1628,6 +1631,11 @@ def align_variants(self, node:TVGNode) -> Tuple[TVGNode, TVGNode]:
             new_bridge = bridge_in.copy()
             for edge in bridge_in.out_edges:
                 self.add_edge(new_bridge, edge.out_node, edge.type)
+                # Here we want to limit the process within the variant bubble.
+                # In-bridge nodes should not be merged with their outgoing nodes
+                # that do not belong to the bubble.
+                if edge.out_node not in members:
+                    end_nodes.add(edge.out_node)
             bridge_map[new_bridge] = bridge_in
             trash.add(bridge_in)
 
@@ -1670,7 +1678,7 @@ def align_variants(self, node:TVGNode) -> Tuple[TVGNode, TVGNode]:
             for out_edge in copy.copy(cur.out_edges):
                 out_node:TVGNode = out_edge.out_node
 
-                # So this is the case that some of the end_nodes are in end_nodes
+                # So this is the case that some of the out_nodes are in end_nodes
                 # but not the others.
                 if out_node in end_nodes:
                     new_node = cur.copy()