Unexpected grid potential caused by seaborn or scanpy

[DingWB]

When other packages are imported, such as seaborn and set some kind of style, there would be a unexpected grid in heatmap

import os,sys
%matplotlib inline
import matplotlib.pylab as plt
import pickle
plt.rcParams['figure.dpi'] = 100
plt.rcParams['savefig.dpi']=300
plt.rcParams['font.family']='sans serif'
plt.rcParams['font.sans-serif']='Arial'
plt.rcParams['pdf.fonttype']=42
import pandas as pd
from PyComplexHeatmap import *
import PyComplexHeatmap as pch
print(pch.__version__)
import scanpy
#import seaborn as sns
# sns.set_style("whitegrid")
# sns.set_style("white")

df_row=pd.read_csv("df_row.csv",index_col=0)
df_row.spatial_domain=df_row.spatial_domain.astype(str)
df_col=pd.read_csv("df_col.csv",index_col=0)
df_col.chromosome=df_col.chromosome.astype(str)
data=pd.read_csv("cnv_mtr_filtered.csv",index_col=0)
df_col=df_col.sample(500)
data=data.loc[:,df_col.index.tolist()]
data

plt.figure(figsize=(10, 5))


right_annotation = HeatmapAnnotation(
                **,
                       axis=0)

top_annotation = HeatmapAnnotation(
    
                label=anno_label(df_col['chromosome'],merge=True, rotation=90),
                chromosome = anno_simple(df_col['chromosome'], add_text=False, legend_kws={'frameon':False}, 
                                         colors= dict(zip(np.arange(1,23).astype(str),['#5daab4','#000000']*11)), 
                                         legend=True), axis=1)

cm = ClusterMapPlotter(data=data, 
                       vmax=1.02, vmin=0.98, 
                       top_annotation=top_annotation,   
                       left_annotation =right_annotation ,
                       cmap='RdBu_r',  
                       row_cluster=False, col_cluster=False, 
                        col_split = df_col['chromosome'], 
                       col_split_order = [str(i) for i in list(range(1,23))],
                       col_split_gap=0.4 
                      )

When use sns.set_style("whitegrid")
or change setting in scanpy:

import scanpy
sc.settings.verbosity = 3 
sc.settings.set_figure_params(dpi=100, dpi_save=300,
                              facecolor='white', figsize=(4,4))

An unexpected grid would appear:

When set sns.set_style("whitegrid") or in default, the heatmap looks good:

[XiaofengZhou16]

Hi @DingWB — I'd like to work on this if the fix is still wanted.

On the current main branch, both heatmapPlotter.plot() and the internal plot_heatmap() call pcolormesh() without explicitly disabling the Axes grid. A global style that sets axes.grid=True, such as a seaborn or Scanpy style, can therefore leak into the heatmap axes.

I have prepared a local minimal patch that disables the inherited axis grid in both paths while retaining cell borders controlled by linewidths and linecolor. It also adds a backend-independent regression test using matplotlib.rc_context({"axes.grid": True}), without adding seaborn or Scanpy as dependencies.

Would this scope be acceptable before I open a PR?

[DingWB]

Yes, please sync the latest version to your local laptop and then open a PR.
Thank you very much. @XiaofengZhou16

[DingWB]

Hi @XiaofengZhou16 ,

I added ax.grid(False) to each pcolormesh, could you please test it to see whether the issue is gone?
You need to install the latest version from GitHub.

[XiaofengZhou16]

Hi @DingWB — I tested the latest GitHub version at commit 02707c7.

Using matplotlib.rc_context({"axes.grid": True}) to reproduce the inherited-style condition, the unexpected axis grid is gone in both heatmap() and the internal plot_heatmap() path. I also verified that cell borders set through linewidths remain visible.

The regression tests passed on Ubuntu, macOS, and Windows with Python 3.11 and 3.12:
https://github.com/XiaofengZhou16/PyComplexHeatmap/actions/runs/32280845810

I opened #124 with the two regression tests only; it does not duplicate your code change or add seaborn/Scanpy dependencies. I did not run the original private/data-specific example, so this confirms the minimal reproducible style condition rather than that exact dataset.

Comment added by Codex