def run_feature_matrix(df, test, fp_cutoff=.5):
df = df.ix[df.apply(test.check_feature, 1)]
if hasattr(test, 'first_pass'):
fp = df.apply(test.first_pass, 1)
df = df[fp.p < fp_cutoff]
full = df.apply(test.full_test, 1)
res = pd.concat([full[['LR', 'fmla']], fp], keys=['Full', 'Univariate'], axis=1)
if type(res.index[0]) == tuple: # pandas bug
res.index = pd.MultiIndex.from_tuples(res.index, names=df.index.names)
res = res.join(pd.Series(bhCorrection(res[('Full', 'LR')], n=len(fp)),
name=('Full', 'LR_q')))
res = res.join(pd.Series(bhCorrection(res[('Univariate', 'p')], n=len(fp)),
name=('Univariate', 'q')))
return res.sort_index(axis=1).sort(columns=[('Full', 'LR')])
def run_feature_matrix(df, test, fp_cutoff=.5):
df = df.ix[df.apply(test.check_feature, 1)]
if hasattr(test, 'first_pass'):
fp = df.apply(test.first_pass, 1)
df = df[fp.p < fp_cutoff]
full = df.apply(test.full_test, 1)
res = pd.concat([full[['LR', 'fmla']], fp],
keys=['Full', 'Univariate'],
axis=1)
if type(res.index[0]) == tuple: # pandas bug
res.index = pd.MultiIndex.from_tuples(res.index, names=df.index.names)
res = res.join(
pd.Series(bhCorrection(res[('Full', 'LR')], n=len(fp)),
name=('Full', 'LR_q')))
res = res.join(
pd.Series(bhCorrection(res[('Univariate', 'p')], n=len(fp)),
name=('Univariate', 'q')))
return res.sort_index(axis=1).sort(columns=[('Full', 'LR')])
def filter_bad_pathways(self, gene_lookup):
for clin_type, p_vals in self.p_genes.iteritems():
for gene in set(p_vals.index).intersection(set(gene_lookup)):
for pathway in gene_lookup[gene]:
if ((pathway in self.q_pathways[clin_type]) and
(p_vals[gene] < self.p_pathways[clin_type][pathway]) and
(self.hit_matrix.ix[gene].sum() >
self.meta_matrix.ix[pathway].sum()*.5)):
self.p_pathways[clin_type][pathway] = nan
self.q_pathways[clin_type] = bhCorrection(self.p_pathways[clin_type])
def cox_screen(df, surv, axis=1):
if axis == 0:
df = df.T
c = df.apply(pd.value_counts, axis=1).count(1)
df = df.ix[c[c > 1].index]
rr = df.apply(lambda s: cox(s.dropna(), surv), axis=1)
rr[('LR', 'q')] = bhCorrection(rr['LR']['p'])
rr = rr.sort([('LR', 'q')])
rr = rr.sortlevel(0, axis=1)
return rr
def cox_screen(df, surv, axis=1):
if axis == 0:
df = df.T
c = df.apply(pd.value_counts, axis=1).count(1)
df = df.ix[c[c > 1].index]
rr = df.apply(lambda s: cox(s.dropna(), surv), axis=1)
rr[('LR', 'q')] = bhCorrection(rr['LR']['p'])
rr = rr.sort([('LR', 'q')])
rr = rr.sortlevel(0, axis=1)
return rr
def rna_filter(cn, val, rna):
'''
Filter copy number events with rna expression data.
Here we test whether the event is associated with a subsequent
change in expression in those patients.
cn: copy number matrix, should have a MultiIndex, with the gene name
in the last level
val: value of the copy number to test in [-2, -1, 1, 2]
'''
assert val in [-2, -1, 1, 2]
change = pd.DataFrame({g: kruskal_pandas(vec == val, rna.ix[g[-1]])
for g, vec in cn.iterrows()
if g[-1] in rna.index}).T
q_vals = bhCorrection(change.p)
filtered = cn.ix[true_index(q_vals < .1)]
return filtered
def rna_filter(cn, val, rna):
'''
Filter copy number events with rna expression data.
Here we test whether the event is associated with a subsequent
change in expression in those patients.
cn: copy number matrix, should have a MultiIndex, with the gene name
in the last level
val: value of the copy number to test in [-2, -1, 1, 2]
'''
assert val in [-2, -1, 1, 2]
change = pd.DataFrame({
g: kruskal_pandas(vec == val, rna.ix[g[-1]])
for g, vec in cn.iterrows() if g[-1] in rna.index
}).T
q_vals = bhCorrection(change.p)
filtered = cn.ix[true_index(q_vals < .1)]
return filtered
def lr_screen(df, surv):
rr = df.astype(float).apply(log_rank, args=(surv,), axis=1)
rr['q'] = bhCorrection(rr.p)
rr = rr.sort('p')
return rr
def lr_screen(df, surv):
rr = df.astype(float).apply(log_rank, args=(surv, ), axis=1)
rr['q'] = bhCorrection(rr.p)
rr = rr.sort('p')
return rr
def screen_feature(vec, test, df):
s = pd.DataFrame({f: test(vec, feature) for f,feature in df.iterrows()}).T
s['q'] = bhCorrection(s.p)
s = s.sort(columns='p')
return s