From 6edd9dc1da7a3dd3351b51ec7eb485878ddcaba5 Mon Sep 17 00:00:00 2001
From: Rocio Riveros <r.riverosmaidana@gmail.com>
Date: Tue, 29 Sep 2026 13:44:45 -0300
Subject: [PATCH] Fix FixedLocator/labels mismatch in create_heatmap for large
 matrices

sns.clustermap's tick locator only reliably matches the cell count once
the matrix is large enough (~40+ categories); below that it happens to
work. Explicitly fix the x/y tick positions to cell centers (offset by
0.5, matching clustermap's own convention) before assigning labels, so
set_xticklabels/set_yticklabels always see a matching tick count.

Confirmed the crash empirically with 44 and 48 structures; 21 worked
fine before this fix.

Also scale figsize with the number of structures and drop the per-cell
numeric annotations past 50 structures: with the crash fixed, larger
matrices (100+) still rendered with illegible overlapping labels and
cell text, since the figure size was previously fixed regardless of
matrix size.
---
 MHCXGraph/scripts/create_heatmaps.py | 65 +++++++++++++++++++---------
 1 file changed, 44 insertions(+), 21 deletions(-)

diff --git a/MHCXGraph/scripts/create_heatmaps.py b/MHCXGraph/scripts/create_heatmaps.py
index 3b4b538..410790a 100644
--- a/MHCXGraph/scripts/create_heatmaps.py
+++ b/MHCXGraph/scripts/create_heatmaps.py
@@ -471,6 +471,19 @@ def create_heatmap(args):
     comp_df_full = build_component_matrix(comp_df)
     ratio_df_full = build_ratio_matrix(comp_df)
 
+    # A fixed figure size makes tick labels and cell annotations illegible
+    # once there are many structures, so scale it with the matrix size.
+    # 0.25 in/structure keeps 11pt labels readable; small matrices keep
+    # the original 10x8 minimum.
+    structure_count = len(labels)
+    figure_width = max(10, structure_count * 0.25)
+    figure_height = max(8, structure_count * 0.25)
+
+    # Cell-value annotations turn into unreadable overlapping text once
+    # there are too many cells to fit them individually, so skip them
+    # past this size and let the color alone convey the value.
+    max_annotated_structures = 50
+
     # Seaborn receives a precomputed SciPy linkage matrix, so it does not
     # recalculate pairwise distances. It still uses SciPy to interpret and
     # render the dendrogram layout.
@@ -479,7 +492,7 @@ def create_heatmap(args):
         cmap="viridis",
         row_linkage=linkage_matrix,
         col_linkage=linkage_matrix,
-        figsize=(10, 8),
+        figsize=(figure_width, figure_height),
         dendrogram_ratio=0.15,
         cbar_kws={"label": "Similarity index"},
     )
@@ -492,36 +505,46 @@ def create_heatmap(args):
         columns=ordered_labels,
     )
 
-    cluster_grid.ax_heatmap.set_xticklabels(
+    heatmap_axis = cluster_grid.ax_heatmap
+
+    # Fix the tick positions before assigning labels. sns.clustermap's
+    # FixedLocator only reliably matches the cell count once the matrix
+    # is large (~40+ categories); below that it works by chance. Ticks
+    # are offset by 0.5 to land on cell centers, matching the convention
+    # clustermap already uses for its own small-matrix ticks.
+    n = len(ordered_labels)
+    heatmap_axis.set_xticks([i + 0.5 for i in range(n)])
+    heatmap_axis.set_yticks([i + 0.5 for i in range(n)])
+
+    heatmap_axis.set_xticklabels(
         ordered_labels,
         fontsize=11,
         rotation=90,
     )
 
-    cluster_grid.ax_heatmap.set_yticklabels(
+    heatmap_axis.set_yticklabels(
         ordered_labels,
         fontsize=11,
         rotation=0,
     )
 
-    heatmap_axis = cluster_grid.ax_heatmap
-
-    for row_index in range(len(ordered_labels)):
-        for column_index in range(len(ordered_labels)):
-            value = component_ordered.iloc[
-                row_index,
-                column_index,
-            ]
-
-            heatmap_axis.text(
-                column_index + 0.5,
-                row_index + 0.5,
-                f"{int(value)}",
-                horizontalalignment="center",
-                verticalalignment="center",
-                fontsize=11,
-                color="black",
-            )
+    if structure_count <= max_annotated_structures:
+        for row_index in range(len(ordered_labels)):
+            for column_index in range(len(ordered_labels)):
+                value = component_ordered.iloc[
+                    row_index,
+                    column_index,
+                ]
+
+                heatmap_axis.text(
+                    column_index + 0.5,
+                    row_index + 0.5,
+                    f"{int(value)}",
+                    horizontalalignment="center",
+                    verticalalignment="center",
+                    fontsize=11,
+                    color="black",
+                )
 
     output_file = output_directory / args.name