def get_gistic_genes(data_path, cancer, filter_with_rna=True,
collapse_on_bands=True, min_patients=5):
"""
Gets a matrix of events for high grade amplifications and homozygous
deletions.
We filter down this list by asserting that a copy number event corresponds
with a resultant expression change.
The final matrix merges gene-level events on the same band to combine
redundant events and reduce the test space.
"""
gistic = FH.get_gistic_gene_matrix(data_path, cancer, '01')
deletion = gistic[(gistic == -2).sum(1) > min_patients]
amp = gistic[(gistic == 2).sum(1) > min_patients]
ft = pd.MultiIndex.from_tuples # rediculously long pandas names
deletion.index = ft([('Deletion', s[0], s[2]) for s in deletion.index])
amp.index = ft([('Amplification', s[0], s[2]) for s in amp.index])
if filter_with_rna:
rna = FH.read_rnaSeq(data_path, cancer)
deletion = rna_filter(deletion, -2, rna)
amp = rna_filter(amp, 2, rna)
cna_genes = amp.append(deletion)
if collapse_on_bands == False:
return cna_genes
cna_genes = pd.DataFrame({(a[0], a[1], tuple(b.index.get_level_values(2))):
b.mean().round() for a, b in
cna_genes.groupby(level=[0, 1])}).T
cna_genes.index = pd.MultiIndex.from_tuples(cna_genes.index)
return cna_genes
def get_gistic_genes(data_path,
cancer,
filter_with_rna=True,
collapse_on_bands=True,
min_patients=5):
'''
Gets a matrix of events for high grade amplifications and homozygous
deletions.
We filter down this list by asserting that a copy number event corresponds
with a resultant expression change.
The final matrix merges gene-level events on the same band to combine
redundant events and reduce the test space.
'''
gistic = FH.get_gistic_gene_matrix(data_path, cancer, '01')
deletion = gistic[(gistic == -2).sum(1) > min_patients]
amp = gistic[(gistic == 2).sum(1) > min_patients]
ft = pd.MultiIndex.from_tuples # rediculously long pandas names
deletion.index = ft([('Deletion', s[0], s[2]) for s in deletion.index])
amp.index = ft([('Amplification', s[0], s[2]) for s in amp.index])
if filter_with_rna:
rna = FH.read_rnaSeq(data_path, cancer)
deletion = rna_filter(deletion, -2, rna)
amp = rna_filter(amp, 2, rna)
cna_genes = amp.append(deletion)
if collapse_on_bands == False:
return cna_genes
cna_genes = pd.DataFrame({(a[0], a[1], tuple(b.index.get_level_values(2))):
b.mean().round()
for a, b in cna_genes.groupby(level=[0, 1])}).T
cna_genes.index = pd.MultiIndex.from_tuples(cna_genes.index)
return cna_genes
def get_cna_rates(data_path, cancer, patients=None):
"""
Get copy-number aberration rates from GISTIC processing pipeline.
This function depends on the current Firehose output of this program
as of July 2013.
"""
gistic = FH.get_gistic_gene_matrix(data_path, cancer)
amp_gene_all = (gistic >= 1).astype(int).sum()
amp_gene_high = (gistic == 2).astype(int).sum()
del_gene_all = (gistic <= -1).astype(int).sum()
del_gene_homo = (gistic <= -2).astype(int).sum()
lesions = FH.get_gistic_lesions(data_path, cancer)
amp_lesion_all = (lesions.ix['Amplification'] >= 1).sum()
amp_lesion_high = (lesions.ix['Amplification'] == 2).sum()
del_lesion_all = (lesions.ix['Deletion'] <= -1).sum()
del_lesion_homo = (lesions.ix['Deletion'] == -2).sum()
arm_cn = FH.get_gistic_arm_values(data_path, cancer)
chromosomal_instability = arm_cn.abs().mean()
cna_df = {'gene_amp': amp_gene_all, 'gene_amp_high': amp_gene_high,
'gene_del': del_gene_all, 'gene_del_homo': del_gene_homo,
'lesion_amp': amp_lesion_all, 'lesion_amp_high': amp_lesion_high,
'lesion_del': del_lesion_all, 'lesion_del_homo': del_lesion_homo,
'chrom_instability': chromosomal_instability}
cna_df = pd.DataFrame(cna_df)
if patients is not None:
cna_df = cna_df.ix[patients].dropna()
return cna_df
def get_global_vars(data_path, cancer, patients=None):
"""
Get compiled DataFrame of global molecular variables from Firehose
data. Returns a feature by patient DataFrame with (data-type, variable)
on the columns and patient barcodes on the index.
"""
try:
data_matrix = FH.read_rnaSeq(data_path, cancer, patients)
U, S, vH = frame_svd(data_matrix)
exp_pc = pd.DataFrame({'pc1': vH[0], 'pc2': vH[1]})
except:
exp_pc = pd.DataFrame()
try:
data_matrix = read_methylation(data_path, cancer, patients)
U, S, vH = frame_svd(data_matrix)
meth_pc = pd.DataFrame({'pc1': vH[0], 'pc2': vH[1]})
except:
meth_pc = pd.DataFrame()
try:
meth_age, amar = 'FAIL', 'FAIL'
# meth_age, amar = get_age_signal(data_path, cancer)
meth_pc = meth_pc.join(meth_age).join(amar)
print 'Should probably check this out'
except:
pass
cna_rates = get_cna_rates(data_path, cancer, patients)
mutation_rates = get_mutation_rates(data_path, cancer, patients)
gv = pd.concat([exp_pc, meth_pc, cna_rates, mutation_rates],
keys=['mRNASeq', 'methylation', 'cna', 'mutation'], axis=1)
gv = gv.dropna(how='all', axis=1)
return gv
def get_gistic(data_path, cancer, filter_with_rna=True,
collapse_on_bands=True, min_patients=5):
"""
Get the combined GISTIC feature matrix for testing.
"""
lesions = FH.get_gistic_lesions(cancer, data_path)
cna_genes = get_gistic_genes(data_path, cancer, filter_with_rna,
collapse_on_bands, min_patients)
cna = cna_genes.append(lesions)
return cna
def __init__(self, run, cancer, cn_type, patients=None):
'''
'''
Dataset.__init__(self, cancer.path, cn_type, compressed=False)
min_pat = run.parameters['min_patients']
if cn_type == 'CN_broad':
self.df = FH.get_gistic(run.data_path, cancer.name,
min_patients=min_pat)
if patients is not None:
self.df = self.df.ix[:, patients].dropna(1, how='all')
self.features = self.df
def __init__(self, run, cancer, cn_type, patients=None):
'''
'''
Dataset.__init__(self, cancer.path, cn_type, compressed=False)
min_pat = run.parameters['min_patients']
if cn_type == 'CN_broad':
self.df = FH.get_gistic(run.data_path, cancer.name,
min_patients=min_pat)
if patients is not None:
self.df = self.df.ix[:, patients].dropna(1, how='all')
self.features = self.df
def get_gistic(data_path,
cancer,
filter_with_rna=True,
collapse_on_bands=True,
min_patients=5):
'''
Get the combined GISTIC feature matrix for testing.
'''
lesions = FH.get_gistic_lesions(cancer, data_path)
cna_genes = get_gistic_genes(data_path, cancer, filter_with_rna,
collapse_on_bands, min_patients)
cna = cna_genes.append(lesions)
return cna
def __init__(self, run, cancer, patients=None,
create_features=True, draw_figures=False):
"""
"""
Dataset.__init__(self, cancer.path, 'Mutation', compressed=False)
self.df = FH.get_mutation_matrix(run.data_path, cancer.name)
if patients is not None:
self.df = self.df.ix[:, patients].dropna(1, how='all')
if create_features is True:
min_pat = run.parameters['min_patients']
self._create_feature_matrix(run.gene_sets, min_pat)
if draw_figures is True:
self._create_pathway_figures(run.gene_sets)
def get_cna_rates(data_path, cancer, patients=None):
'''
Get copy-number aberration rates from GISTIC processing pipeline.
This function depends on the current Firehose output of this program
as of July 2013.
'''
gistic = FH.get_gistic_gene_matrix(data_path, cancer)
amp_gene_all = (gistic >= 1).astype(int).sum()
amp_gene_high = (gistic == 2).astype(int).sum()
del_gene_all = (gistic <= -1).astype(int).sum()
del_gene_homo = (gistic <= -2).astype(int).sum()
lesions = FH.get_gistic_lesions(data_path, cancer)
amp_lesion_all = (lesions.ix['Amplification'] >= 1).sum()
amp_lesion_high = (lesions.ix['Amplification'] == 2).sum()
del_lesion_all = (lesions.ix['Deletion'] <= -1).sum()
del_lesion_homo = (lesions.ix['Deletion'] == -2).sum()
arm_cn = FH.get_gistic_arm_values(data_path, cancer)
chromosomal_instability = arm_cn.abs().mean()
cna_df = {
'gene_amp': amp_gene_all,
'gene_amp_high': amp_gene_high,
'gene_del': del_gene_all,
'gene_del_homo': del_gene_homo,
'lesion_amp': amp_lesion_all,
'lesion_amp_high': amp_lesion_high,
'lesion_del': del_lesion_all,
'lesion_del_homo': del_lesion_homo,
'chrom_instability': chromosomal_instability
}
cna_df = pd.DataFrame(cna_df)
if patients is not None:
cna_df = cna_df.ix[patients].dropna()
return cna_df
def get_beta_values(data_path, cancer, patients=None, tissue_code='All'):
"""
Retrieve methylation beta-values from my pre-processed file.
TCGA has a lot more columns that eat up memory, so I parse out the
beta-values in preprocess_methylation.py.
This file still has all of the probes by pateints so it still eats my
kill a computer without a lot of memory (takes ~2GB for HNSC).
"""
path = '{}/ucsd_processing/{}/methylation450/'.format(data_path, cancer)
t = pd.read_table(path + 'beta_values.txt', skiprows=[1], index_col=[0])
t = t.rename(columns=lambda s: s if s != t.columns[0] else 'symbol')
t = t.set_index('symbol', append=True)
t = t.swaplevel(0, 1)
t = t.sort_index() # think this is a bug that it needs to be sorted
t = FH.fix_barcode_columns(t, patients, tissue_code)
return t
def get_beta_values(data_path, cancer, patients=None, tissue_code='All'):
'''
Retrieve methylation beta-values from my pre-processed file.
TCGA has a lot more columns that eat up memory, so I parse out the
beta-values in preprocess_methylation.py.
This file still has all of the probes by pateints so it still eats my
kill a computer without a lot of memory (takes ~2GB for HNSC).
'''
path = '{}/ucsd_processing/{}/methylation450/'.format(data_path, cancer)
t = pd.read_table(path + 'beta_values.txt', skiprows=[1], index_col=[0])
t = t.rename(columns=lambda s: s if s != t.columns[0] else 'symbol')
t = t.set_index('symbol', append=True)
t = t.swaplevel(0, 1)
t = t.sort_index() # think this is a bug that it needs to be sorted
t = FH.fix_barcode_columns(t, patients, tissue_code)
return t
def __init__(self,
run,
cancer,
patients=None,
create_features=True,
draw_figures=False):
"""
"""
Dataset.__init__(self, cancer.path, 'Mutation', compressed=False)
self.df = FH.get_mutation_matrix(run.data_path, cancer.name)
if patients is not None:
self.df = self.df.ix[:, patients].dropna(1, how='all')
if create_features is True:
min_pat = run.parameters['min_patients']
self._create_feature_matrix(run.gene_sets, min_pat)
if draw_figures is True:
self._create_pathway_figures(run.gene_sets)
def get_global_vars(data_path, cancer, patients=None):
'''
Get compiled DataFrame of global molecular variables from Firehose
data. Returns a feature by patient DataFrame with (data-type, variable)
on the columns and patient barcodes on the index.
'''
try:
data_matrix = FH.read_rnaSeq(data_path, cancer, patients)
U, S, vH = frame_svd(data_matrix)
exp_pc = pd.DataFrame({'pc1': vH[0], 'pc2': vH[1]})
except:
exp_pc = pd.DataFrame()
try:
data_matrix = read_methylation(data_path, cancer, patients)
U, S, vH = frame_svd(data_matrix)
meth_pc = pd.DataFrame({'pc1': vH[0], 'pc2': vH[1]})
except:
meth_pc = pd.DataFrame()
try:
meth_age, amar = 'FAIL', 'FAIL'
# meth_age, amar = get_age_signal(data_path, cancer)
meth_pc = meth_pc.join(meth_age).join(amar)
print 'Should probably check this out'
except:
pass
cna_rates = get_cna_rates(data_path, cancer, patients)
mutation_rates = get_mutation_rates(data_path, cancer, patients)
gv = pd.concat([exp_pc, meth_pc, cna_rates, mutation_rates],
keys=['mRNASeq', 'methylation', 'cna', 'mutation'],
axis=1)
gv = gv.dropna(how='all', axis=1)
return gv
