def run_patient_specific_permutations(n_tot, n_all, n_perm=1000):
# perms
n_spec = {pid: [] for pid in n_tot}
for i in range(n_perm):
this = []
for pid in n_tot:
ix = np.random.permutation(n_all)[:n_tot[pid]]
this.append(ix)
this_spec = setops.specific_features(*this)
for pid, s in zip(n_tot.keys(), this_spec):
n_spec[pid].append(len(s))
return n_spec
dmr_res.to_pickle(fn, include_annotation=False)
logger.info("Saved DMR results to %s", fn)
dmr_res_all = dmr_res.results_significant
# 1. check the phenomenon is still observed in our syngeneic comparisons
# full list
for_plot = dict([(pid, dmr_res_all['syngeneic_%s' % pid]) for pid in pids])
plt_dict = bar_plot(for_plot, pids)
plt_dict['fig'].tight_layout()
plt_dict['fig'].savefig(os.path.join(outdir,
"syngeneic_full_list_directions.png"),
dpi=200)
# specific list
spec_ix = setops.specific_features(*[for_plot[pid].keys() for pid in pids])
for_plot = dict([(pid, dict([(k, for_plot[pid][k]) for k in s]))
for pid, s in zip(pids, spec_ix)])
plt_dict = bar_plot(for_plot, pids)
plt_dict['fig'].tight_layout()
plt_dict['fig'].savefig(os.path.join(
outdir, "syngeneic_specific_list_directions.png"),
dpi=200)
# 2. check that we see the same phenomenon when we switch to the validation cohort comparator
# full list
for_plot = dict([(pid, dmr_res_all['consistency_%s' % pid])
for pid in pids])
plt_dict = bar_plot(for_plot, pids)
plt_dict['fig'].tight_layout()
plt_dict['fig'].savefig(os.path.join(
dm_res = pd.read_excel(fn)
# total number of (DM cluster, gene) pairs
# n_all_de_dm = dm_res.shape[0]
# reported elsewhere, so use that here (it's approximate)
n_all_de_dm = 17000
fn = os.path.join(HGIC_LOCAL_DIR, 'current', 'core_pipeline',
'rnaseq_methylation_combined',
'de_dmr_concordant_syngeneic_only.xlsx')
de_dm_res = pd.read_excel(fn)
# number of patient specific DE/DMRs
de_dm_per_pat = {
pid: de_dm_res.index[de_dm_res[pid] == 'Y']
for pid in pids
}
n_tot_de_dm = {pid: de_dm_per_pat[pid].size for pid in pids}
tt = setops.specific_features(*[de_dm_per_pat[pid] for pid in pids])
n_ps_de_dmr = {pid: len(t) for pid, t in zip(pids, tt)}
n_spec_perm_de_dm = run_patient_specific_permutations(n_tot_de_dm,
n_all_de_dm,
n_perm=n_perm)
fig, axs = plot_perms_kde_vs_obs(
n_spec_perm_de_dm,
n_ps_de_dmr,
xlabel='Number of patient-specific DE/DMRs',
order=pids)
fig.savefig(os.path.join(outdir, "patient_specific_de_dmr.png"), dpi=200)
fig.savefig(os.path.join(outdir, "patient_specific_de_dmr.tiff"), dpi=200)
fig.savefig(os.path.join(outdir, "patient_specific_de_dmr.pdf"))
# set True to limit to the PIDs that are present in all comparisons
limit_to_common_comparisons = False
# compute specific DMRs and plot direction
common_pids = sorted(setops.reduce_intersection(*pids_included.values()))
dmrs_specific = {}
for k in pids_included:
if limit_to_common_comparisons:
this_pids = common_pids
else:
this_pids = pids_included[k]
dmrs_specific[k] = {}
this = [all_results[k].results_significant[p] for p in this_pids]
this_specific = setops.specific_features(*this)
for p, cids, spec_dict in zip(this_pids, this_specific, this):
if p not in dmrs_specific[k]:
dmrs_specific[k][p] = {}
for cid in cids:
dmrs_specific[k][p][cid] = spec_dict[cid]
for k in pids_included:
if limit_to_common_comparisons:
this_pids = common_pids
else:
this_pids = pids_included[k]
# scale figure width according to number of PIDs
fig_width = 5.5 + 0.24 * (len(this_pids) - 4)
de_by_direction = count_de_by_direction(de_res)
plt_dict = bar_plot(de_res, pids)
plt.setp(plt_dict["axs"][0].yaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][0].yaxis.get_label(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].yaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].yaxis.get_label(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].xaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].xaxis.get_label(), fontsize=fontsize)
plt_dict['fig'].tight_layout()
plt_dict['fig'].savefig(os.path.join(outdir,
"syngeneic_full_list_directions.png"),
dpi=200)
# specific list
spec_ix = setops.specific_features(*[de_res[pid].index for pid in pids])
for_plot = dict([(pid, de_res[pid].loc[s])
for pid, s in zip(pids, spec_ix)])
plt_dict = bar_plot(for_plot, pids)
plt.setp(plt_dict["axs"][0].yaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][0].yaxis.get_label(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].yaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].yaxis.get_label(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].xaxis.get_ticklabels(), fontsize=fontsize)
plt.setp(plt_dict["axs"][1].xaxis.get_label(), fontsize=fontsize)
plt_dict['fig'].tight_layout()
plt_dict['fig'].savefig(os.path.join(
outdir, "syngeneic_specific_list_directions.png"),
dpi=200)
# We don't see the phenotype
chrom_length[new_k] = cl[k]
for pid in pids:
fn = os.path.join(outdir, "%s_dmrs.bw" % pid)
write_bigwig(
dmr_res_s1[pid].results_significant,
clusters,
chrom_length,
fn,
chr_prefix=chr_prefix
)
# repeat for patient-specific DMRs
patient_specific_cids = dict(zip(
pids,
setops.specific_features(*[dmr_res_s1[pid].results_significant for pid in pids])
))
for pid in pids:
fn = os.path.join(outdir, "%s_specific_dmrs.bw" % pid)
this_res = dict([
(cid, dmr_res_s1[pid].results[cid]) for cid in patient_specific_cids[pid]
])
write_bigwig(
this_res,
clusters,
chrom_length,
fn,
chr_prefix=chr_prefix
)
dmr_res_s1 = dmr.DmrResultCollection.from_pickle(fn, anno=anno)
else:
raise IOError(
"We require a pre-computed file, %s, which could not be found." %
fn)
# extract full (all significant) results
dmr_res_all = dmr_res_s1.results_significant
clusters = dmr_res_s1[pids[0]].clusters
n_by_patient = dict([(pid, len(dmr_res_all[pid])) for pid in pids])
specific_dmrs = dict(
zip(pids,
setops.specific_features(*[dmr_res_all[pid] for pid in pids])))
n_by_patient_specific = dict([(pid, len(specific_dmrs[pid]))
for pid in pids])
ntot = sum(n_by_patient.values())
# 1) Null: DMRs are picked uniformly randomly from the pool with variable marginal totals for each patient.
# Marginal totals are given by the (real) number of DMRs in each patient.
rvs = dict([(pid, [
np.random.choice(range(ntot), replace=False, size=n_by_patient[pid])
for i in range(n_iter)
]) for pid in pids])
inters_1 = [[
len(x)
for x in setops.specific_features(*[rvs[pid][i] for pid in pids])
'dmr_%s' % t for t in relations_tss
]].any(axis=1)].gene)]) for pid in pids])
de_by_direction = same_de.count_de_by_direction(de_linked)
plt_dict = same_de.bar_plot(de_linked, pids, figsize=(3, 4))
plt_dict['fig'].tight_layout()
plt.setp(common.get_children_recursive(plt_dict['fig'], plt.Text),
fontsize=12)
plt_dict['fig'].savefig(os.path.join(
outdir, "de_linked_syngeneic_full_list_tss_directions.png"),
dpi=200)
# patient-specific DMRs linked to DE genes
spec_ix = setops.specific_features(
*[dmr_res_all[pid].keys() for pid in pids])
dm_specific = dict([(pid, dict([(k, dmr_res_all[pid][k]) for k in s]))
for pid, s in zip(pids, spec_ix)])
# manually link these
dm_specific_genes = {}
for pid in pids:
cl_ids = dm_specific[pid].keys()
dm_specific_genes[pid] = setops.reduce_union(
*[[t[0] for t in dmr_res_s1.clusters[c].genes] for c in cl_ids])
de_linked_spec = dict([
(pid, de_res_s1[pid].loc[de_res_s1[pid]['Gene Symbol'].isin(
dm_specific_genes[pid])]) for pid in pids
])
