changes min_allowed_nodes_pct in adaptive_core_expansion to 0.9

CHANGELOG.md

@@ -5,6 +5,11 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Changed
+- The default `min_allowed_nodes_pct` in `adaptive_core_expansion` has been changed from 0.8 to 0.9, meaning that a valid "high" core partition must include at least 90% of all nodes. Components that don't meet this criteria will not be denoised by ACE. 
+
 ## [0.29.0] - 2026-06-15
 
 ### Added

src/pixelator/pna/graph/adaptive_core_expansion.py

@@ -257,7 +257,7 @@ def adaptive_core_expansion(
     max_iter: int = 200,
     min_seed_pct: float = 0.1,
     nodes_to_move_threshold: int = 10,
-    min_allowed_nodes_pct: float = 0.8,
+    min_allowed_nodes_pct: float = 0.9,
     select_lcc: bool = True,
 ) -> Graph:
     """Perform Adaptive Core Expansion (ACE) graph partitioning.

tests/pna/denoise/test_denoise.py

@@ -513,7 +513,8 @@ def f(*args, **kwargs):
 
 
 REFERENCE_ACE_COMPONENT = "57129a8b0fff38c6"
-REFERENCE_ACE_LOW_NODE_COUNT = 5436
+REFERENCE_ACE_COMPONENT_HIGH_QUALITY = "c4c3ef9497b3746d"
+REFERENCE_ACE_LOW_NODE_COUNT = 32
 
 
 def test_denoise_pls_reference_component_runs_and_cleans_coreness(denoise_pxl_dataset):
@@ -588,13 +589,14 @@ def test_denoise_ace_reference_component(denoise_pxl_dataset):
         denoise_pxl_dataset: Denoise pxl dataset.
     """
     comp_graph = PNAGraph.from_edgelist(
-        denoise_pxl_dataset.filter(components=[REFERENCE_ACE_COMPONENT])
+        denoise_pxl_dataset.filter(components=[REFERENCE_ACE_COMPONENT_HIGH_QUALITY])
         .edgelist()
         .to_polars()
         .lazy()
     )
     removed = denoise_ace(comp_graph)
     assert removed != [None]
+    a = len(removed)
     assert len(removed) == REFERENCE_ACE_LOW_NODE_COUNT
     partitions = nx.get_node_attributes(comp_graph.raw, "partition")
     low_ids = {n for n, p in partitions.items() if p == "low"}
@@ -625,21 +627,23 @@ def test_denoise_ace_analysis(denoise_pxl_dataset, tmp_path):
         == obs["number_of_nodes_removed_in_denoise"]
     ).all()
     assert int(
-        obs.loc[REFERENCE_ACE_COMPONENT, "denoised_nodes_marked_only_by_ace"]
+        obs.loc[
+            REFERENCE_ACE_COMPONENT_HIGH_QUALITY, "denoised_nodes_marked_only_by_ace"
+        ]
     ) == (REFERENCE_ACE_LOW_NODE_COUNT)
 
     assert (
         obs.loc[:, "denoised_nodes_marked_stranded"].sum() == 0
     )  # ACE-only denoising with LCC seed should not produce stranded nodes
 
     orig_graph = PNAGraph.from_edgelist(
-        denoise_pxl_dataset.filter(components=[REFERENCE_ACE_COMPONENT])
+        denoise_pxl_dataset.filter(components=[REFERENCE_ACE_COMPONENT_HIGH_QUALITY])
         .edgelist()
         .to_polars()
         .lazy()
     )
     denoised_graph = PNAGraph.from_edgelist(
-        denoised_dataset.filter(components=[REFERENCE_ACE_COMPONENT])
+        denoised_dataset.filter(components=[REFERENCE_ACE_COMPONENT_HIGH_QUALITY])
         .edgelist()
         .to_polars()
         .lazy()

tests/pna/denoise/test_denoise_cli.py

@@ -72,8 +72,7 @@ def test_denoise_ace_cli_runs_ok(denoise_pxl_file):
         result = read(out_pxl)
         obs = result.adata().obs
         assert (
-            int(obs.loc["57129a8b0fff38c6", "denoised_nodes_marked_only_by_ace"])
-            == 5436
+            int(obs.loc["c4c3ef9497b3746d", "denoised_nodes_marked_only_by_ace"]) == 32
         )
         summary_cols = [
             "denoised_nodes_marked_only_by_ace",