"""CNVkit's core data structure, a copy number array."""
from __future__ import print_function, absolute_import, division
from builtins import map
from past.builtins import basestring
import logging
import numpy as np
import pandas as pd
from scipy.stats import mannwhitneyu, rankdata
from . import core, gary, descriptives, params, smoothing
from .metrics import segment_mean
[docs]class CopyNumArray(gary.GenomicArray):
"""An array of genomic intervals, treated like aCGH probes.
Required columns: chromosome, start, end, gene, log2
Optional columns: gc, rmask, spread, weight, probes
"""
_required_columns = ("chromosome", "start", "end", "gene", "log2")
_required_dtypes = (str, int, int, str, float)
# ENH: make gene optional
# Extra columns, in order:
# "depth", "gc", "rmask", "spread", "weight", "probes"
def __init__(self, data_table, meta_dict=None):
gary.GenomicArray.__init__(self, data_table, meta_dict)
@property
def log2(self):
return self.data["log2"]
@log2.setter
def log2(self, value):
self.data["log2"] = value
@property
def _chr_x_label(self):
if 'chr_x' in self.meta:
return self.meta['chr_x']
chr_x = ('chrX' if self.chromosome.iat[0].startswith('chr') else 'X')
self.meta['chr_x'] = chr_x
return chr_x
@property
def _chr_y_label(self):
if 'chr_y' in self.meta:
return self.meta['chr_y']
chr_y = ('chrY' if self._chr_x_label.startswith('chr') else 'Y')
self.meta['chr_y'] = chr_y
return chr_y
# More meta to store:
# is_sample_male = bool
# is_reference_male = bool
# * invalidate 'chr_x' if .chromosome/['chromosome'] is set
# Traversal
[docs] def by_gene(self, ignore=params.IGNORE_GENE_NAMES):
"""Iterate over probes grouped by gene name.
Group each series of intergenic bins as a "Background" gene; any
"Background" bins within a gene are grouped with that gene.
Bins' gene names are split on commas to accommodate overlapping genes
and bins that cover multiple genes.
Parameters
----------
ignore : list or tuple of str
Gene names to treat as "Background" bins instead of real genes,
grouping these bins with the surrounding gene or background region.
These bins will still retain their name in the output.
Yields
------
tuple
Pairs of: (gene name, CNA of rows with same name)
"""
start_idx = end_idx = None
for _chrom, subgary in self.by_chromosome():
prev_idx = 0
for gene, gene_idx in subgary._get_gene_map().items():
if not (gene == 'Background' or gene in ignore):
if not len(gene_idx):
logging.warn("Specified gene name somehow missing: %s",
gene)
continue
start_idx = gene_idx[0]
end_idx = gene_idx[-1] + 1
if prev_idx < start_idx:
# Include intergenic regions
yield "Background", subgary.as_dataframe(
subgary.data.iloc[prev_idx:start_idx])
yield gene, subgary.as_dataframe(
subgary.data.iloc[start_idx:end_idx])
prev_idx = end_idx
if prev_idx < len(subgary) - 1:
# Include the telomere
yield "Background", subgary.as_dataframe(
subgary.data.iloc[prev_idx:])
# Manipulation
[docs] def center_all(self, estimator=pd.Series.median, skip_low=False, by_chrom=True):
"""Re-center log2 values to the autosomes' average (in-place).
Parameters
----------
estimator : str or callable
Function to estimate central tendency. If a string, must be one of
'mean', 'median', 'mode', 'biweight' (for biweight location). Median
by default.
skip_low : bool
Whether to drop very-low-coverage bins (via `drop_low_coverage`)
before estimating the center value.
by_chrom : bool
If True, first apply `estimator` to each chromosome separately, then
apply `estimator` to the per-chromosome values, to reduce the impact
of uneven targeting or extreme aneuploidy. Otherwise, apply
`estimator` to all log2 values directly.
"""
est_funcs = {
"mean": pd.Series.mean,
"median": pd.Series.median,
"mode": descriptives.modal_location,
"biweight": descriptives.biweight_location,
}
if isinstance(estimator, basestring):
if estimator in est_funcs:
estimator = est_funcs[estimator]
else:
raise ValueError("Estimator must be a function or one of: %s"
% ", ".join(map(repr, est_funcs)))
cnarr = (self.drop_low_coverage() if skip_low else self).autosomes()
if cnarr:
if by_chrom:
values = pd.Series([estimator(subarr['log2'])
for _c, subarr in cnarr.by_chromosome()
if len(subarr)])
else:
values = cnarr['log2']
self.data['log2'] -= estimator(values)
[docs] def drop_low_coverage(self):
"""Drop bins with extremely low log2 coverage or copy ratio values.
These are generally bins that had no reads mapped due to sample-specific
issues. A very small log2 ratio or coverage value may have been
substituted to avoid domain or divide-by-zero errors.
"""
return self.as_dataframe(self.data[self.data['log2'] >
params.NULL_LOG2_COVERAGE -
params.MIN_REF_COVERAGE])
[docs] def squash_genes(self, summary_func=descriptives.biweight_location,
squash_background=False, ignore=params.IGNORE_GENE_NAMES):
"""Combine consecutive bins with the same targeted gene name.
Parameters
----------
summary_func : callable
Function to summarize an array of log2 values to produce a
new log2 value for a "squashed" (i.e. reduced) region. By default
this is the biweight location, but you might want median, mean, max,
min or something else in some cases.
squash_background : bool
If True, also reduce consecutive "Background" bins into a single
bin. Otherwise, keep "Background" and ignored bins as they are in
the output.
ignore : list or tuple of str
Bin names to be treated as "Background" instead of as unique genes.
Return
------
CopyNumArray
Another, usually smaller, copy of `self` with each gene's bins
reduced to a single bin with appropriate values.
"""
def squash_rows(name, rows):
"""Combine multiple rows (for the same gene) into one row."""
if len(rows) == 1:
return tuple(rows.iloc[0])
chrom = core.check_unique(rows.chromosome, 'chromosome')
start = rows.start.iat[0]
end = rows.end.iat[-1]
cvg = summary_func(rows.log2)
outrow = [chrom, start, end, name, cvg]
# Handle extra fields
# ENH - no coverage stat; do weighted average as appropriate
for xfield in ('gc', 'rmask', 'spread', 'weight'):
if xfield in self:
outrow.append(summary_func(rows[xfield]))
if 'probes' in self:
outrow.append(sum(rows['probes']))
return tuple(outrow)
outrows = []
for name, subarr in self.by_gene(ignore):
if name == 'Background' and not squash_background:
outrows.extend(subarr.data.itertuples(index=False))
else:
outrows.append(squash_rows(name, subarr.data))
return self.as_rows(outrows)
# Chromosomal sex (gender)
[docs] def shift_xx(self, male_reference=False, is_xx=None):
"""Adjust chrX coverages (divide in half) for apparent female samples."""
outprobes = self.copy()
if is_xx is None:
is_xx = self.guess_xx(male_reference=male_reference)
if is_xx and male_reference:
# Female: divide X coverages by 2 (in log2: subtract 1)
outprobes[outprobes.chromosome == self._chr_x_label, 'log2'] -= 1.0
# Male: no change
elif not is_xx and not male_reference:
# Male: multiply X coverages by 2 (in log2: add 1)
outprobes[outprobes.chromosome == self._chr_x_label, 'log2'] += 1.0
# Female: no change
return outprobes
[docs] def guess_xx(self, male_reference=False, verbose=True):
"""Detect chromosomal sex; return True if a sample is probably female.
Uses `compare_sex_chromosomes` to calculate coverage ratios of the X and
Y chromosomes versus autosomes.
Parameters
----------
male_reference : bool
Was this sample normalized to a male reference copy number profile?
verbose : bool
If True, print (i.e. log to console) the ratios of the log2
coverages of the X and Y chromosomes versus autosomes, the
"maleness" ratio of male vs. female expectations for each sex
chromosome, and the inferred chromosomal sex.
Returns
-------
bool
True if the coverage ratios indicate the sample is female.
"""
is_xy, stats = self.compare_sex_chromosomes(male_reference)
if is_xy is None:
return None
elif verbose:
logging.info("Relative log2 coverage of %s=%.3g, %s=%.3g "
"(maleness=%.3g x %.3g = %.3g) --> assuming %s",
self._chr_x_label, stats['chrx_ratio'],
self._chr_y_label, stats['chry_ratio'],
stats['chrx_male_lr'], stats['chry_male_lr'],
stats['chrx_male_lr'] * stats['chry_male_lr'],
('female', 'male')[is_xy])
return ~is_xy
[docs] def compare_sex_chromosomes(self, male_reference=False, skip_low=False):
"""Compare coverage ratios of sex chromosomes versus autosomes.
Perform 4 Mann-Whitney tests of the log2 coverages on chromosomes X and
Y, separately shifting for assumed male and female chromosomal sex.
Compare the U values obtained to infer whether the male or female
assumption fits the data better.
Parameters
----------
male_reference : bool
Whether a male reference copy number profile was used to normalize
the data. If so, a male sample should have log2 values of 0 on X and
Y, and female +1 on X, deep negative (below -3) on Y. Otherwise, a
male sample should have log2 values of -1 on X and 0 on
Y, and female 0 on X, deep negative (below -3) on Y.
skip_low : bool
If True, drop very-low-coverage bins (via `drop_low_coverage`)
before comparing log2 coverage ratios. Included for completeness,
but shouldn't affect the result much since the M-W test is
nonparametric and p-values are not used here.
Returns
-------
bool
True if the sample appears male.
dict
Calculated values used for the inference: relative log2 ratios of
chromosomes X and Y versus the autosomes; the Mann-Whitney U values
from each test; and ratios of U values for male vs. female
assumption on chromosomes X and Y.
"""
cnarr = self.drop_low_coverage() if skip_low else self
chrx = cnarr[cnarr.chromosome == self._chr_x_label]
if not len(chrx):
logging.warn("*WARNING* No %s found in probes; check the input",
self._chr_x_label)
return None, {}
chrx_l = chrx['log2'].values
auto = cnarr.autosomes()
auto_l = auto['log2'].values
# ENH: use scipy.stats.ttest_ind(a, b, equal_var=False)
def compare_to_auto(vals):
# u = mannwhitneyu(auto_l, vals, alternative='two-sided')[0]
# From scipy.stats.mannwhitneyu -- skip the p-value calculation
# (the tiebreaker there can raise an irrelevant error)
n1 = len(auto_l)
n2 = len(vals)
ranked = rankdata(np.concatenate((auto_l, vals)))
rankx = ranked[0:n1] # get the x-ranks
u1 = n1*n2 + (n1*(n1+1))/2.0 - np.sum(rankx, axis=0) # calc U for x
return n1*n2 - u1 # remainder is U for y
if male_reference:
female_chrx_u = compare_to_auto(chrx_l - 1)
male_chrx_u = compare_to_auto(chrx_l)
else:
female_chrx_u = compare_to_auto(chrx_l)
male_chrx_u = compare_to_auto(chrx_l + 1)
# Mann-Whitney U score is greater for similar-mean sets
chrx_male_lr = male_chrx_u / female_chrx_u
# Similar for chrY if it's present
chry = cnarr[cnarr.chromosome == self._chr_y_label]
if len(chry):
chry_l = chry['log2']
male_chry_u = compare_to_auto(chry_l)
female_chry_u = compare_to_auto(chry_l + 3)
chry_male_lr = male_chry_u / female_chry_u
else:
# If chrY is missing, don't sabotage the inference
male_chry_u = female_chry_u = np.nan
chry_male_lr = 1.0
# Relative log2 values, for convenient reporting
auto_mean = segment_mean(auto, skip_low=skip_low)
chrx_mean = segment_mean(chrx, skip_low=skip_low)
chry_mean = segment_mean(chry, skip_low=skip_low)
return (chrx_male_lr * chry_male_lr > 1.0,
dict(chrx_ratio=chrx_mean - auto_mean,
chry_ratio=chry_mean - auto_mean,
# For debugging, mainly
chrx_male_lr=chrx_male_lr,
chry_male_lr=chry_male_lr,
female_chrx_u=female_chrx_u,
male_chrx_u=male_chrx_u,
female_chry_u=female_chry_u,
male_chry_u=male_chry_u,
))
[docs] def expect_flat_log2(self, is_male_reference=None):
"""Get the uninformed expected copy ratios of each bin.
Create an array of log2 coverages like a "flat" reference.
This is a neutral copy ratio at each autosome (log2 = 0.0) and sex
chromosomes based on whether the reference is male (XX or XY).
"""
if is_male_reference is None:
is_male_reference = not self.guess_xx(verbose=False)
cvg = np.zeros(len(self), dtype=np.float_)
if is_male_reference:
# Single-copy X, Y
idx = np.asarray((self.chromosome == self._chr_x_label) |
(self.chromosome == self._chr_y_label))
else:
# Y will be all noise, so replace with 1 "flat" copy
idx = np.asarray(self.chromosome == self._chr_y_label)
cvg[idx] = -1.0
return cvg
# Reporting
[docs] def residuals(self, segments=None):
"""Difference in log2 value of each bin from its segment mean.
Parameters
----------
segments : GenomicArray, CopyNumArray, or None
Determines the "mean" value to which `self` log2 values are relative:
- If CopyNumArray, use the log2 values as the segment means to
subtract.
- If GenomicArray with no log2 values, group `self` by these ranges
and subtract each group's median log2 value.
- If None, subtract each chromosome's median.
Returns
-------
array
Residual log2 values from `self` relative to `segments`; same length
as `self`.
"""
if not segments:
resids = [subcna.log2 - subcna.log2.median()
for _chrom, subcna in self.by_chromosome()]
elif "log2" in segments:
resids = [subcna.log2 - seg.log2
for seg, subcna in self.by_ranges(segments)]
else:
resids = [subcna.log2 - subcna.log2.median()
for _seg, subcna in self.by_ranges(segments)]
return np.concatenate(resids) if resids else np.array([])
def _guess_average_depth(self, segments=None, window=100):
"""Estimate the effective average read depth from variance.
Assume read depths are Poisson distributed, converting log2 values to
absolute counts. Then the mean depth equals the variance , and the average
read depth is the estimated mean divided by the estimated variance.
Use robust estimators (Tukey's biweight location and midvariance) to
compensate for outliers and overdispersion.
With `segments`, take the residuals of this array's log2 values from
those of the segments to remove the confounding effect of real CNVs.
If `window` is an integer, calculate and subtract a smoothed trendline
to remove the effect of CNVs without segmentation (skipped if `segments`
are given).
See: http://www.evanmiller.org/how-to-read-an-unlabeled-sales-chart.html
"""
# Try to drop allosomes
cnarr = self.autosomes()
if not len(cnarr):
cnarr = self
# Remove variations due to real/likely CNVs
y_log2 = cnarr.residuals(segments)
if segments is None and window:
y_log2 -= smoothing.smoothed(y_log2, window)
# Guess Poisson parameter from absolute-scale values
y = np.exp2(y_log2)
# ENH: use weight argument to these stats
loc = descriptives.biweight_location(y)
spread = descriptives.biweight_midvariance(y, loc)
if spread > 0:
return loc / spread**2
return loc