def run_dmr_analyses(data, comparisons, anno, dmr_params, verbose=True):
"""
Compute DMRs for paired GBM-iNSC comparisons (defined in that order) for all patients
:param me_data: Pandas dataframe containing M values, columns are samples and rows are probes.
:param comparisons: Dictionary, each key is a title, each value is a 2-element iterable containing lists of
sample names. The comparison is run as group 1 - group 2 in each case.
:param anno:
:param pids:
:param dmr_params:
:return:
"""
dmr_res_obj = dmr.DmrResults(anno=anno)
dmr_res_obj.identify_clusters(**dmr_params)
dmr_res = {}
for the_ttl, the_samples in comparisons.items():
logger.info(
"Comparison %s. Group 1: %s. Group 2: %s.",
the_ttl,
','.join(the_samples[0]),
','.join(the_samples[1]),
)
the_obj = dmr_res_obj.copy()
the_obj.test_clusters(data,
samples=the_samples,
n_jobs=dmr_params['n_jobs'],
min_median_change=dmr_params['delta_m_min'],
method=dmr_params['dmr_test_method'],
alpha=dmr_params['alpha'],
**dmr_params['test_kwargs'])
dmr_res[the_ttl] = the_obj
return dmr.DmrResultCollection(**dmr_res)
def compute_dmr_clusters(anno, dmr_params):
clusters = []
cid = 0
for cc in anno.CHR.unique():
coords = anno.loc[anno.CHR == cc, 'MAPINFO'].sort_values()
this_clust = dmr.identify_cluster(coords, dmr_params['n_min'],
dmr_params['d_max'])
for cl in this_clust:
clusters.append(dmr.ProbeCluster(cl, anno, cluster_id=cid, chr=cc))
cid += 1
return dmr.DmrResults(clusters=clusters, anno=anno)
def compute_cross_dmr(me_data, me_meta, anno, pids, dmr_params, external_references=(('GIBCO', 'NSC'),)):
obj = dmr.DmrResults(anno=anno)
obj.identify_clusters(**dmr_params)
res = {}
# loop over GBM groups
for pid1 in pids:
res.setdefault(pid1, {})
the_idx1 = me_meta.index.str.contains(pid1) & (me_meta.loc[:, 'type'] == 'GBM')
# loop over iNSC groups
for pid2 in pids:
the_idx2 = me_meta.index.str.contains(pid2) & (me_meta.loc[:, 'type'] == 'iNSC')
the_idx = the_idx1 | the_idx2
the_groups = me_meta.loc[the_idx, 'type'].values
the_samples = me_meta.index[the_idx].groupby(the_groups).values()
the_obj = obj.copy()
the_obj.test_clusters(me_data,
samples=the_samples,
n_jobs=dmr_params['n_jobs'],
min_median_change=dmr_params['delta_m_min'],
method=dmr_params['dmr_test_method'],
**dmr_params['test_kwargs']
)
res[pid1][pid2] = the_obj
# loop over external reference NSC groups
for er, er_type in external_references:
the_idx2 = me_meta.index.str.contains(er) & (me_meta.loc[:, 'type'] == er_type)
the_idx = the_idx1 | the_idx2
the_groups = me_meta.loc[the_idx, 'type'].values
the_samples = me_meta.index[the_idx].groupby(the_groups).values()
the_obj = obj.copy()
the_obj.test_clusters(me_data,
samples=the_samples,
n_jobs=dmr_params['n_jobs'],
min_median_change=dmr_params['delta_m_min'],
method=dmr_params['dmr_test_method'],
**dmr_params['test_kwargs']
)
res[pid1][er] = the_obj
return dmr.DmrResultCollection(**res)
def paired_dmr(me_data, me_meta, anno, pids, dmr_params):
"""
Compute DMRs for paired GBM-iNSC comparisons (defined in that order) for all patients
:param me_data:
:param me_meta:
:param anno:
:param pids:
:param dmr_params:
:return:
"""
dmr_res_obj = dmr.DmrResults(anno=anno)
dmr_res_obj.identify_clusters(**dmr_params)
dmr_res = {}
for pid in pids:
the_idx1 = me_meta.index.str.contains(pid) & (me_meta.loc[:, 'type'] == 'GBM')
the_idx2 = me_meta.index.str.contains(pid) & (me_meta.loc[:, 'type'] == 'iNSC')
# control comparison order
the_samples = [
me_meta.index[the_idx1],
me_meta.index[the_idx2],
]
# the_idx = the_idx1 | the_idx2
# the_groups = me_meta.loc[the_idx, 'type'].values
# the_samples = me_meta.index[the_idx].groupby(the_groups).values()
the_obj = dmr_res_obj.copy()
the_obj.test_clusters(me_data,
samples=the_samples,
n_jobs=dmr_params['n_jobs'],
min_median_change=dmr_params['delta_m_min'],
method=dmr_params['dmr_test_method'],
alpha=dmr_params['alpha'],
**dmr_params['test_kwargs']
)
dmr_res[pid] = the_obj
return dmr.DmrResultCollection(**dmr_res)
facecolor='w',
framealpha=0.5)
ax.set_ylim([-0.01, 1.01])
ax.set_xlabel("M value")
ax.set_ylabel("ECDF")
fig.tight_layout()
fig.savefig(os.path.join(outdir, "methylation_ecdf.png"), dpi=200)
## TODO: linear interp along y axis to identify regions that could be found DMR due to norming differences?
## TODO: apply this to the hGIC project
# 1) DMR: All shBMI1 vs all scramble, etc... (aggregating cell lines)
dmr_res_obj = dmr.DmrResults(anno=anno)
dmr_res_obj.identify_clusters(**dmr_params)
comparisons = [
collections.OrderedDict([
('shCHD7', ['3021_1_shC', 'C', 'ICb1299_shCHD7', 'p62_3_shChd7']),
('scramble', ['3021_1_Scr', 'S', 'ICb1299_Scr', 'p62_3_Scr'])
]),
collections.OrderedDict([
('shBMI1', ['3021_1_shB', 'B', 'ICb1299_shBMI1', 'p62_3_shBmi1']),
('scramble', ['3021_1_Scr', 'S', 'ICb1299_Scr', 'p62_3_Scr'])
]),
collections.OrderedDict([
('shCHD7shBMI1',
['3021_1_shB+C', 'B+C', 'ICb1299_shBMI1CHD7', 'p62_3_shB+C']),
('scramble', ['3021_1_Scr', 'S', 'ICb1299_Scr', 'p62_3_Scr'])
mdat_019_e = mdat['GBM019_P4']
mdat_019_l = mdat['GBM019Luc_P12']
mdat_019_exvivo = mdat.loc[:, ['GBM019Luc_P3_PDX1', 'GBM019Luc_P2_PDX2']]
e_minus_l = pd.Series(mdat_019_e.values - mdat_019_l.values,
index=mdat.index)
exvivo_minus_l = mdat_019_exvivo.subtract(mdat_019_l.values, axis=0)
raise StopIteration
# DMR analysis
dmr_params = consts.DMR_PARAMS
dmr_params['n_jobs'] = mp.cpu_count()
anno = loader.load_illumina_methylationepic_annotation()
anno = anno.loc[mdat.index]
ffpe_gic_dmrs = dmr.DmrResults(anno=anno)
ffpe_gic_dmrs.identify_clusters(**dmr_params)
# use only GIC and FFPE here
this_mdat = mdat.loc[:,
obj.meta.descriptor.isin(['In vitro GIC', 'Bulk GBM']
)]
this_samples = this_mdat.columns.groupby(obj.meta.loc[this_mdat.columns,
'descriptor'])
samples = [
this_samples['In vitro GIC'],
this_samples['Bulk GBM'],
]
ffpe_gic_dmrs.test_clusters(this_mdat,
samples=samples,
n_jobs=dmr_params['n_jobs'],