import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from lib5c.util.plotting import plotter
[docs]@plotter
def plot_distance_bias(ob, bins, bin_labels=None, idx='disp', threshold=0.05,
colors=None, labels=None, xlabel='distance range',
legend_label='group', **kwargs):
"""
Plots a bar plot illustrating the degree to which p-values are biased among
different distance scales.
This method visualizes distance bias by computing a specified percentile of
all p-values called in an analysis, and then computing the proportion of
pixels with p-values below this percentile in each distance bin.
Parameters
----------
ob : hic3defdr.HiC3DeFDR object or list of hic3defdr.HiC3DeFDR objects
The analyis or analyses to inspect for distance bias.
bins : list of tuple of int
Each tuple represents a distance bin as a (min, max) pair, where min and
max are distances in bin units and the ranges are inclusive. If either
min or max is None, the distance bin will be considered unbounded on
that end.
bin_labels : list of str
Pass a list of labels to describe the distance bins.
idx : {'disp', 'loop'}
Pass 'disp' to use p-values for all points for which dispersion was
estimated. Pass 'loop' to only use p-values for points which are in
loops.
threshold : float
The percentile to use for the comparison.
colors : str or list of str, optional
If ``ob`` is a single object, pass a single color to color the bars in
the barplot. If ``ob`` is a list of objects, pass a list of colors. Pass
None to use automatic colors.
labels : list of str, optional
If ``ob`` is a list of objects, you must pass a list of strings to label
the objects. Otherwise, this kwarg does nothing.
xlabel : str
The label for the x-axis.
legend_label : str
If ``ob`` is a list of objects, the label to use for the legend title.
Otherwise, this kwarg does nothing.
kwargs : kwargs
Typical plotter kwargs.
Returns
-------
pyplot axis
The axis plotted on.
"""
# promote ob to list and resolve labels
if type(ob) not in [list, tuple]:
ob = [ob]
hue = None
color = 'k' if colors is None else colors
colors = None
if labels is None:
labels = ['group1']
else:
hue = legend_label
color = None
if labels is None:
raise ValueError('must pass labels if ob is a list or tuple')
# resolve bin_labels
if bin_labels is None:
bin_labels = []
for min_dist, max_dist in bins:
if min_dist is None and max_dist is not None:
label = '<= %s' % max_dist
elif min_dist is not None and max_dist is None:
label = '>= %s' % min_dist
elif min_dist is None and max_dist is None:
label = 'all'
else:
label = '%s to %s' % (min_dist, max_dist)
bin_labels.append(label)
data = []
for o, label in zip(ob, labels):
# load data
disp_idx, _ = o.load_data('disp_idx', 'all')
if idx == 'loop':
loop_idx, _ = o.load_data('loop_idx', 'all')
rc_idx = (disp_idx, loop_idx)
p_idx = loop_idx
else:
rc_idx = disp_idx
p_idx = None
disp_idx, _ = o.load_data('disp_idx', 'all')
loop_idx, _ = o.load_data('loop_idx', 'all')
row, _ = o.load_data('row', 'all', idx=rc_idx)
col, _ = o.load_data('col', 'all', idx=rc_idx)
dist = col - row
pvalues, _ = o.load_data('pvalues', 'all', idx=p_idx)
p_star = np.percentile(pvalues, 100*threshold)
# process each distance bin
for bin_label, (min_dist, max_dist) in zip(bin_labels, bins):
dist_idx = np.ones(len(dist), dtype=bool)
if min_dist is not None:
dist_idx[dist < min_dist] = False
if max_dist is not None:
dist_idx[dist > max_dist] = False
perc = np.mean(pvalues[dist_idx] < p_star)
data.append({legend_label: label, xlabel: bin_label,
'percentage significant': perc})
df = pd.DataFrame(data)
sns.barplot(data=df, x=xlabel, y='percentage significant', hue=hue,
color=color, palette=colors)
xlim = plt.xlim()
plt.hlines(threshold, *xlim, color='gray', linestyle='--')
plt.xlim(xlim)