def read_ancestry_files(self, only_optimal_Ks=False):
dataframes = []
datasets = Dataset.all_datasets()
Ks = self.available_Ks()
panels = Panel.all_panels() + Panel.all_control_panels()
for dataset, K, panel in product(datasets, Ks, panels):
if only_optimal_Ks and self.optimal_Ks()[dataset.label] != K:
continue
# Results are sorted in directories named like DATASET_PANEL
tag = "{}_{}".format(dataset.label, panel.label)
basedir = join(ADMIXTURE_DIR, tag)
if not isdir(basedir):
continue
# Read the .Q file for ratios of ancestry per sample
fname = "{}.{}.Q".format(tag, K)
ancestries_df = pd.read_csv(join(basedir, fname), sep="\s+",
names=list(range(K)))
# Read the .fam file for the sample IDs (they're in the same order)
fname = "{}.fam".format(tag)
samples = pd.read_csv(join(basedir, fname), sep="\s+", index_col=0,
usecols=[0], names=["sample"])
ancestries_df.index = samples.index
# Add population data to the sample IDs
samples_df = ThousandGenomes().all_samples()
ancestries_df = samples_df.join(ancestries_df).dropna()
continents_present = len(ancestries_df["superpopulation"].unique())
if continents_present >= 3:
self.infer_ancestral_components_from_samples_origin(ancestries_df)
self.infer_ancestral_components_from_reference_pop(ancestries_df)
# Arrange the hierarchical index
ancestries_df.reset_index(inplace=True)
ancestries_df["dataset"] = dataset.label
ancestries_df["K"] = K
ancestries_df["panel"] = panel.label
ancestries_df.set_index(["dataset", "K", "panel"], inplace=True)
dataframes.append(ancestries_df)
return pd.concat(dataframes)
def plot_(self, components_df, explained_variance, title, filename,
component_pairs=[("PC1", "PC2")], plot_size=None, legend_on=True):
# + 1 axes for the one with the legend, +1 because index starts at 1
ax_ids = list(np.arange(1, len(component_pairs) + 2))
nrows, ncols, figsize = self._fig_dimensions(len(ax_ids), plot_size)
fig = plt.figure(figsize=figsize)
for components_to_compare in component_pairs:
ax_id = ax_ids.pop(0)
ax = fig.add_subplot(nrows, ncols, ax_id)
ax = self.draw_ax(ax, components_to_compare, components_df,
explained_variance, "PEL", title)
if legend_on:
# Legend subplot. It will use the handles and labels of the last ax
handles, labels = ax.get_legend_handles_labels()
populations_df = ThousandGenomes.population_names()
descriptions = populations_df.ix[labels, "description"]
legend_labels = [" - ".join([code, desc])
for code, desc in descriptions.iteritems()]
ax = fig.add_subplot(nrows, ncols, ax_ids.pop(0))
ax = legend_subplot(ax, handles, legend_labels)
# plt.tight_layout()
fig.suptitle(title, fontsize=18, position=(0.12, 1.1), ha="left",
family="serif")
plt.subplots_adjust(wspace=0.05)
if filename is not None:
makedirs(self.FIGS_DIR, exist_ok=True)
plt.savefig(join(self.FIGS_DIR, filename), facecolor="w",
bbox_inches="tight")
def _generate_maf_long_df(self):
mafs = ThousandGenomes.mafs()
long_format_mafs = OrderedDict()
# population_level can be "population" or "superpopulation"
for population_level, dic in mafs.items():
names, frames = dic.keys(), dic.values()
merged_df = pd.concat(frames, axis=1, keys=names)
long_df = pd.melt(merged_df)
long_df.columns = ["panel", "population", "MAF"]
long_format_mafs[population_level] = long_df
return long_format_mafs