def test_get_allele_counts(self):
''' check that counting alleles works correctly
'''
# check the counts when the variants are aggregated for a gene
expected = {'gene_alt': 3, 'gene_ref': 195}
self.assertEqual(get_allele_counts(self.counts, gene=True), expected)
# check the counts when the variants are aggregated for a site
expected = {'ref_F': 135, 'ref_R': 125, 'alt_F': 45, 'alt_R': 43,
'parent_alt': 3, 'parent_ref': 195}
self.assertEqual(get_allele_counts(self.counts, gene=False), expected)
def test_get_allele_counts(self):
''' check that counting alleles works correctly
'''
# check the counts when the variants are aggregated for a gene
expected = {'gene_alt': 3, 'gene_ref': 195}
self.assertEqual(get_allele_counts(self.counts, gene=True), expected)
# check the counts when the variants are aggregated for a site
expected = {
'ref_F': 135,
'ref_R': 125,
'alt_F': 45,
'alt_R': 43,
'parent_alt': 3,
'parent_ref': 195
}
self.assertEqual(get_allele_counts(self.counts, gene=False), expected)
def test_sites(de_novos, pass_status=None):
""" tests each site for deviation from expected behaviour
Args:
de_novos: dataframe of de novo variants
pass_status: whether the candidate passed prelimary filtering for MAF,
and segdups. Sometimes we pass in tables that include variaants that
failed MAF etc, since instead of filtering we want a column
indicating pass status. We need to exclude these variants from the
strand bias and parental alt checks.
Returns:
tuple of pandas Series, one of p-values from testing if the variants have
a strand bias, and the second of p-values from testing if the variants
have an excess of parental alts.
"""
de_novos["key"] = list(
zip(de_novos["chrom"], de_novos["pos"], de_novos["alt"]))
alleles = de_novos[[
"key", "child_ref_F", "child_ref_R", "child_alt_F", "child_alt_R",
"mother_ref_F", "mother_ref_R", "mother_alt_F", "mother_alt_R",
"father_ref_F", "father_ref_R", "father_alt_F", "father_alt_R"
]].copy()
if pass_status is not None:
alleles = alleles[pass_status]
variants = alleles.groupby("key")
# count the ref and alt alleles for each de novo site
counts = [get_allele_counts(site) for name, site in variants]
results = pandas.DataFrame(counts)
results["key"] = [name for name, site in variants]
# check for overabundance of parental alt alleles using binomial test
parent_counts = pandas.DataFrame({
"alt": results["parent_alt"],
"ref": results["parent_ref"]
})
parental_alt_p = parent_counts.apply(scipy.stats.binom_test,
axis=1,
p=ERROR_RATE)
recode = dict(zip(results['key'], parental_alt_p))
parental_bias = de_novos['key'].map(recode)
# check for strand bias by fishers exact test on the allele counts
strand_bias_p = results.apply(site_strand_bias, axis=1)
recode = dict(zip(results['key'], strand_bias_p))
strand_bias = de_novos['key'].map(recode)
return strand_bias, parental_bias
def test_genes(de_novos, strand_bias, pass_status=None):
""" checks if the variants in a gene have more parental ALTs than expected
Args:
de_novos: dataframe of de novo variants
Returns:
p-value for whether the forward or reverse are biased in the proportion
of ref and alt alleles within each gene.
"""
sites = de_novos.copy()
if pass_status is not None:
sites = sites[pass_status]
strand_bias = strand_bias.copy()[pass_status]
# exclude de novo SNVs that fail the strand bias filter, otherwise these
# skew the parental alts within genes
sites = sites[(strand_bias >= P_CUTOFF) & (sites["ref"].str.len() == 1) &
(sites["alt"].str.len() == 1)]
# cover the edge case where we don't have any sites for testing
if len(sites) == 0:
return [float('nan')] * len(de_novos)
sites = sites[[
"symbol", "mother_ref_F", "mother_ref_R", "mother_alt_F",
"mother_alt_R", "father_ref_F", "father_ref_R", "father_alt_F",
"father_alt_R"
]].copy()
# count the number of parental alleles within genes, and restructure data
# for testing
genes = sites.groupby("symbol")
counts = [get_allele_counts(site, gene=True) for name, site in genes]
results = pandas.DataFrame(counts)
results["symbol"] = [name for name, site in genes]
# check for overabundance of parental alt alleles using binomial test
parent_counts = pandas.DataFrame({
"alt": results["gene_alt"],
"ref": results["gene_ref"]
})
parental_alt_p = parent_counts.apply(scipy.stats.binom_test,
axis=1,
p=ERROR_RATE)
recode = dict(zip(results["symbol"], parental_alt_p))
return de_novos['symbol'].map(recode)
def test_sites(de_novos, pass_status=None):
""" tests each site for deviation from expected behaviour
Args:
de_novos: dataframe of de novo variants
pass_status: whether the candidate passed prelimary filtering for MAF,
and segdups. Sometimes we pass in tables that include variaants that
failed MAF etc, since instead of filtering we want a column
indicating pass status. We need to exclude these variants from the
strand bias and parental alt checks.
Returns:
tuple of pandas Series, one of p-values from testing if the variants have
a strand bias, and the second of p-values from testing if the variants
have an excess of parental alts.
"""
de_novos["key"] = list(zip(de_novos["chrom"], de_novos["pos"], de_novos["alt"]))
alleles = de_novos[["key",
"child_ref_F", "child_ref_R", "child_alt_F", "child_alt_R",
"mother_ref_F", "mother_ref_R", "mother_alt_F", "mother_alt_R",
"father_ref_F", "father_ref_R", "father_alt_F", "father_alt_R"]].copy()
if pass_status is not None:
alleles = alleles[pass_status]
alleles = alleles.convert_objects(convert_numeric=True)
variants = alleles.groupby("key")
# count the ref and alt alleles for each de novo site
counts = [ get_allele_counts(site) for name, site in variants ]
results = pandas.DataFrame(counts)
results["key"] = [ name for name, site in variants ]
# check for overabundance of parental alt alleles using binomial test
parent_counts = pandas.DataFrame({"alt": results["parent_alt"], "ref": results["parent_ref"]})
parental_alt_p = parent_counts.apply(scipy.stats.binom_test, axis=1, p=ERROR_RATE)
recode = dict(zip(results['key'], parental_alt_p))
parental_bias = de_novos['key'].map(recode)
# check for strand bias by fishers exact test on the allele counts
strand_bias_p = results.apply(site_strand_bias, axis=1)
recode = dict(zip(results['key'], strand_bias_p))
strand_bias = de_novos['key'].map(recode)
return strand_bias, parental_bias
def test_genes(de_novos, strand_bias, pass_status=None):
""" checks if the variants in a gene have more parental ALTs than expected
Args:
de_novos: dataframe of de novo variants
Returns:
p-value for whether the forward or reverse are biased in the proportion
of ref and alt alleles within each gene.
"""
sites = de_novos.copy()
if pass_status is not None:
sites = sites[pass_status]
strand_bias = strand_bias.copy()[pass_status]
# exclude de novo SNVs that fail the strand bias filter, otherwise these
# skew the parental alts within genes
sites = sites[(strand_bias >= P_CUTOFF) & (de_novos["ref"].str.len() == 1) &
(de_novos["alt"].str.len() == 1)]
# cover the edge case where we don't have any sites for testing
if len(sites) == 0:
return [float('nan')] * len(de_novos)
sites = sites[["symbol",
"mother_ref_F", "mother_ref_R", "mother_alt_F", "mother_alt_R",
"father_ref_F", "father_ref_R", "father_alt_F", "father_alt_R"]].copy()
sites = sites.convert_objects(convert_numeric=True)
# count the number of parental alleles within genes, and restructure data
# for testing
genes = sites.groupby("symbol")
counts = [ get_allele_counts(site, gene=True) for name, site in genes ]
results = pandas.DataFrame(counts)
results["symbol"] = [ name for name, site in genes ]
# check for overabundance of parental alt alleles using binomial test
parent_counts = pandas.DataFrame({"alt": results["gene_alt"], "ref": results["gene_ref"]})
parental_alt_p = parent_counts.apply(scipy.stats.binom_test, axis=1, p=ERROR_RATE)
recode = dict(zip(results["symbol"], parental_alt_p))
return de_novos['symbol'].map(recode)
