def plot(self, filename, panel_list, ancestries_df):
dataset_label, _ = self._unique_dataset_and_K_check(ancestries_df)
dataset = Dataset(dataset_label)
population_order = Dataset.used_populations()
rows, cols = 1, len(panel_list)
width, height = self.PLOT_SIZE
fig = plt.figure(figsize=(cols * width, rows * height), dpi=30)
fig.set_size_inches((cols*width), (rows*height))
ax_ids = (np.arange(rows * cols) + 1).tolist()[::-1]
# One subplot per panel
for panel in panel_list:
df_lite = ancestries_df.xs(panel.label, level="panel")
df_lite = df_lite.reset_index(drop=True).set_index("population")
plot_title = "Dataset: {}\n{}".format(dataset.name, panel.name)
ax = fig.add_subplot(rows, cols, ax_ids.pop())
fig, tax = ternary.figure(scale=1, ax=ax)
df_lite = df_lite.loc[population_order]
df_lite = df_lite[["EUR", "AFR", "AMR"]].dropna()
df_grouped = df_lite.groupby(level="population", sort=False)
for population, df_pop_group in df_grouped:
tax.scatter(
df_pop_group.values, label=population, s=45,
alpha=0.75, color=population_colors(population),
marker=population_markers(population)
)
self._ternary_plot_aesthetics(tax, plot_title, df_lite)
makedirs(self.PLOTS_DIR, exist_ok=True)
plt.savefig(join(self.PLOTS_DIR, filename), bbox_inches="tight")
def draw_ax(self, ax, components_to_compare, components_df,
explained_variance, reference_population, title):
ylabel_prefix = ""
xlabel_prefix = ""
for pop_code, components in components_df.groupby(level="population"):
marker = plot_helpers.population_markers(pop_code)
color = plot_helpers.population_colors(pop_code)
filled_markers = ['o', '.', 'D', 's', '^', '<', '>', '*']
lw = 0 if marker in filled_markers else 1 # linewidth
z = 1 if marker == 'o' else 0
# ^ americans are 'o' and appear on top
x = components[components_to_compare[0]]
y = components[components_to_compare[1]]
ax.scatter(x, y, lw=lw, label=pop_code, marker=marker,
c=color, zorder=z, s=30, alpha=0.65)
# Define inversion of axis to align components across plots
# Keep the reference population in the upper left
if pop_code == reference_population:
xaxis_mean = np.mean(ax.get_xlim())
yaxis_mean = np.mean(ax.get_ylim())
# The median determines where most the scatter cloud is
reference_in_the_left = np.median(x) < xaxis_mean
reference_in_the_top = np.median(y) > yaxis_mean
if not reference_in_the_left:
ax.invert_xaxis()
xlabel_prefix = "–"
if not reference_in_the_top:
ax.invert_yaxis()
ylabel_prefix = "–"
xcomponent, ycomponent = components_to_compare
xvariance = explained_variance.ix[xcomponent]
xlabel = "{}{}: {}%".format(xlabel_prefix, xcomponent, xvariance)
ax.set_xlabel(xlabel)
yvariance = explained_variance.ix[ycomponent]
ylabel = "{}{}: {}%".format(ylabel_prefix, ycomponent, yvariance)
ax.set_ylabel(ylabel)
if title:
ax.set_title(title, fontweight="bold")
self._pca_plot_aesthetics(ax)
return ax
