from itertools import combinations
from collections import OrderedDict
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from scipy.stats import ks_2samp, ttest_ind, mannwhitneyu
from ..visualization import *
[docs]class PairwiseComparison:
"""
Pairwise statistical comparison between measurements of two populations.
Attributes:
x (pd.DataFrame) - measurements of first population
y (pd.DataFrame) - measurements of second population
basis (str) - attribute on which populations are compared
"""
def __init__(self, x, y, basis):
"""
Instantiate pairwise comparison of two populations.
Attributes:
x (pd.DataFrame) - measurements of first population
y (pd.DataFrame) - measurements of second population
basis (str) - attribute on which populations are compared
"""
self.x = x
self.y = y
self.basis = basis
@property
def is_greater_than(self):
""" True if first population mean is greater than the second."""
return self.x[self.basis].mean() > self.y[self.basis].mean()
[docs] def compare(self, test='MW', **kwargs):
"""
Run statistical test.
Args:
test (str) - name of test used, one of ('KS', 't', 'MW')
Returns:
p (float) - p-value
is_greater_than (bool) - True if first population mean is greater
kwargs: keyword arguments for statistical test
"""
# extract compared values for each population
x, y = self.x[self.basis], self.y[self.basis]
# perform statistical test
if test.lower() == 'ks':
k, p = ks_2samp(x, y, **kwargs)
elif test.lower() == 't':
k, p = ttest_ind(x, y, **kwargs)
elif test.lower() == 'mw':
k, p = mannwhitneyu(x, y, **kwargs)
else:
raise ValueError('Test {:s} not recognized.'.format(test))
return p, self.is_greater_than
[docs]class PairwiseCelltypeComparison(PairwiseComparison):
"""
Pairwise statistical comparison between two concurrent cell types.
Attributes:
label (str) - attribute used to stratify populations
type1 (str) - first label
type2 (str) - second label
Inherited attributes:
x (pd.DataFrame) - measurements of first population
y (pd.DataFrame) - measurements of second population
basis (str) - attribute on which populations are compared
"""
def __init__(self, measurements, type1, type2, basis,
label='celltype',
concurrent_only=True):
"""
Instantiate comparison between two concurrent cell types.
Args:
measurements (pd.DataFrame) - measurement data
type1 (str or int) - first label
type2 (str or int) - second label
basis (str) - attribute on which populations are compared
label (str) - attribute used to define population labels
concurrent_only (bool) - if True, only compare concurrent cells
"""
# store labels
self.label = label
self.type1 = type1
self.type2 = type2
# select concurrent cells of each type
if concurrent_only:
k1 = 'concurrent_'+str(self.type1)
k2 = 'concurrent_'+str(self.type2)
measurements = measurements[measurements[k1] & measurements[k2]]
# split into two populations
x = measurements[measurements[label] == self.type1]
y = measurements[measurements[label] == self.type2]
# instantiate comparison
super().__init__(x, y, basis=basis)
[docs] def plot(self,
ax=None,
colors=None,
mode='violin',
ylabel=None,
**kwargs):
"""
Visualize comparison using seaborn box or violinplot.
Args:
ax (matplotlib.axes.AxesSubplot)
colors (dict) - color for each box/violin keyed by label
mode (str) - type of comparison, either 'box', 'violin', or 'strip'
ylabel (str) - label for yaxis
kwargs: keyword arguments for seaborn plotting function
"""
# create figure if none provided
if ax is None:
fig, ax = plt.subplots(figsize=(1.25, 1.5))
# plot boxplot
if mode == 'box':
sns.boxplot(ax=ax,
x=self.label,
y=self.basis,
data=pd.concat((self.x, self.y)),
order=(self.type1, self.type2),
notch=True,
width=0.8,
**kwargs)
# plot violinplot
elif mode == 'violin':
sns.violinplot(ax=ax,
x=self.label,
y=self.basis,
data=pd.concat((self.x, self.y)),
order=(self.type1, self.type2),
scale='width',
linewidth=0.5,
**kwargs)
# plot stripplot
elif mode == 'strip':
sns.stripplot(ax=ax,
x=self.label,
y=self.basis,
data=pd.concat((self.x, self.y)),
order=(self.type1, self.type2),
dodge=True,
**kwargs)
# format axis
self.format_axis(ax, colors=colors, mode=mode, ylabel=ylabel)
[docs]class CelltypeComparison:
"""
Summary of comparisons between all labeled celltypes.
Attributes:
measurements (pd.DataFrame) - measurement data
basis (str) - attribute on which populations are compared
label (str) - attribute used to define population labels
test (str) - statistical test used, one of ('KS', 't', 'MW')
"""
def __init__(self, measurements, basis,
label='celltype', test='MW', **kwargs):
"""
Instantiate summary of comparisons between all labeled cell types.
Args:
measurements (pd.DataFrame) - measurement data
basis (str) - attribute on which populations are compared
label (str) - attribute used to define population labels
test (str) - name of test used, one of ('KS', 't', 'MW')
kwargs: keyword arguments for statistical test
"""
self.measurements = measurements
self.basis = basis
self.label = label
self.test = test
# compute and report pvalues
pvals = self.run(**kwargs)
self.report(pvals)
@property
def pairs(self):
""" Unique pairs of labels. """
label_values = self.measurements[self.label].unique()
return list(sorted([sorted(x) for x in combinations(label_values, 2)]))
[docs] def compare_celltype(self, type1, type2, **kwargs):
"""
Args:
type1 (str) - first cell type
type2 (str) - second cell type
kwargs: keyword arguments for statistical test
Returns:
p (float) - p value for comparison statistic
is_greater_than (bool) - True if first population mean is greater
"""
comparison = PairwiseCelltypeComparison(self.measurements, type1, type2, self.basis, label=self.label)
return comparison.compare(self.test, **kwargs)
[docs] def run(self, **kwargs):
"""
Compare all pairwise combinations of cell types.
kwargs: keyword arguments for statistical test
Returns:
pvals (dict) - {comparison: pvalue} pairs
"""
signs = {True: ' > ', False: ' < '}
pvals = OrderedDict()
for pair in self.pairs:
pval, greater_than = self.compare_celltype(*pair, **kwargs)
comparison_name = signs[greater_than].join([str(x) for x in pair])
pvals[comparison_name] = pval
return pvals
[docs] def report(self, pvals):
"""
Print summary of statistical comparisons for each condition.
Args:
pvals (dict) - {comparison: pvalue} pairs
"""
print('Statistical test: {}'.format(self.test))
for test, pval in pvals.items():
print(test + ': p = {:0.4f}'.format(pval))
# # define labels and corresponding fill colors
# axis_labels = {'m': '−/−', 'h': '−/+', 'w': '+/+',
# '0': '−/−', '1': '−/+', '2': '+/+'}
# if colors is None:
# colors = {'m': '−/−', 'h': '−/+', 'w': '+/+',
# '0': 'y', '1': 'm', '2': 'c'}
