import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
import matplotlib as mpl
[docs]def plot_decomp_scatters(
adata,
n_components,
base_name="PC",
obsm_name="X_pca",
hue=None,
palette="viridis",
hue_quantile=(0.25, 0.75),
nrows=5,
ncols=5,
):
available_comps = adata.obsm[obsm_name].shape[1]
nrows = min(nrows, available_comps // 2 // ncols + 1)
fig, axes = plt.subplots(
figsize=(ncols * 3, nrows * 3), nrows=nrows, ncols=ncols, dpi=150
)
vmin = None
vmax = None
for i, ax in enumerate(axes.ravel()):
_x = i * 2
_y = i * 2 + 1
if _y > available_comps - 1:
ax.axis("off")
continue
_plot_data = pd.DataFrame(
{
f"{base_name}{_x + 1}": adata.obsm[obsm_name][:, _x],
f"{base_name}{_y + 1}": adata.obsm[obsm_name][:, _y],
},
index=adata.obs_names,
)
if hue is not None:
_plot_data[hue] = adata.obs[hue]
hue_dtype = str(_plot_data[hue].dtype)
if hue_dtype.startswith("float") or hue_dtype.startswith("int"):
vmin, vmax = _plot_data[hue].quantile(hue_quantile).tolist()
hue_norm = (vmin, vmax)
else:
hue_norm = None
else:
hue_norm = None
sns.scatterplot(
ax=ax,
data=_plot_data,
x=f"{base_name}{_x + 1}",
y=f"{base_name}{_y + 1}",
hue=hue,
hue_norm=hue_norm,
s=1,
linewidth=0,
palette=palette,
legend=None,
)
# adjust axis
xmin, xmax = _plot_data[f"{base_name}{_x + 1}"].quantile([0.01, 0.99]).tolist()
delta = (xmax - xmin) * 0.2
xmin -= delta
xmax += delta
ymin, ymax = _plot_data[f"{base_name}{_y + 1}"].quantile([0.01, 0.99]).tolist()
delta = (ymax - ymin) * 0.2
ymin -= delta
ymax += delta
ax.set(xticks=[], yticks=[], xlim=(xmin, xmax), ylim=(ymin, ymax))
if _x + 1 <= n_components:
ax.set_xlabel(f"{base_name}{_x + 1}", color="red")
if _y + 1 <= n_components:
ax.set_ylabel(f"{base_name}{_y + 1}", color="red")
print(f"Red axis labels are used {base_name}s")
if vmin is not None and vmax is not None:
cnorm = mpl.colors.Normalize(vmin, vmax)
fig.colorbar(
mpl.cm.ScalarMappable(norm=cnorm, cmap="viridis"),
ax=fig.axes,
shrink=0.6,
fraction=0.06,
label=hue,
)
return fig, axes
