target_loci = None
if o.fn_target_loci != None:
target_loci = []
for l in open(o.fn_target_loci):
c, s, e = l.rstrip().split()
s, e = int(s), int(e)
if c == contig: target_loci.append([c, s, e])
tbx_dups = pysam.Tabixfile(o.fn_dup_tabix)
callset_clust = cluster.cluster_callsets(o.fn_call_table, contig)
g = gt.genotyper(contig,
gglob_dir=o.gglob_dir,
plot_dir=o.out_viz_dir,
subset_indivs=subset_indivs,
fn_fa=o.fn_fa,
dup_tabix=tbx_dups,
GC_inf=GC_inf)
fn_sunk_gt_out = o.fn_gt_out.replace(".genotypes", ".sunk_genotypes")
F_gt = open(o.fn_gt_out, 'w')
F_sunk_gt = open(fn_sunk_gt_out, 'w')
F_VCF = open(o.fn_vcf_out, 'w')
F_call = open(o.fn_call_out, 'w')
FINF = open("%s.info" % o.fn_call_out, 'w')
info_ob = info_io.info_io(FINF)
g.setup_output(F_gt, F_sunk_gt, F_VCF, F_call, info_ob)
if o.fn_target_loci != None:
target_loci = []
for l in open(o.fn_target_loci):
c, s, e = l.rstrip().split()
s, e = int(s), int(e)
if c == contig:
target_loci.append([c, s, e])
tbx_dups = pysam.Tabixfile(o.fn_dup_tabix)
callset_clust = cluster.cluster_callsets(o.fn_call_table, contig)
g = gt.genotyper(
contig,
gglob_dir=o.gglob_dir,
plot_dir=o.out_viz_dir,
subset_indivs=subset_indivs,
fn_fa=o.fn_fa,
dup_tabix=tbx_dups,
GC_inf=GC_inf,
)
fn_sunk_gt_out = o.fn_gt_out.replace(".genotypes", ".sunk_genotypes")
F_gt = open(o.fn_gt_out, "w")
F_sunk_gt = open(fn_sunk_gt_out, "w")
F_VCF = open(o.fn_vcf_out, "w")
F_call = open(o.fn_call_out, "w")
FINF = open("%s.info" % o.fn_call_out, "w")
info_ob = info_io.info_io(FINF)
def resolve_overlapping_clusters(self,
ll_cutoff,
tbx_dups,
indiv_id,
indiv_DTS,
ref_DTSs,
dCGHs,
gglob_dir,
out_viz_dir,
verbose=False,
min_overlapping=2,
subset_indivs=None,
min_d=0):
"""
resolve overlapping clusters into individual calls
let the calls have likelihoods
1. do the recip overlap cluster - clusters very similar calls w/ similar break-points
2. make sure there are at least 3 calls in there (w/ similar breakpoints)
3. make sure those calls sum to a log likelihood of <3
4. collapse overlaps
5. get the best breakpoints
"""
print("resolving breakpoints...", file=stderr)
final_calls = []
for chr, overlapping_calls in self.overlapping_calls_by_chr.items():
print(chr,
"%d calls in this chr" % (len(overlapping_calls)),
file=stderr)
indiv_cps = indiv_DTS.get_cps_by_chr(chr)
ref_cps = {}
for ref, refDTS in ref_DTSs.items():
ref_cps[ref] = refDTS.get_cps_by_chr(chr)
g = genotyper(chr,
gglob_dir=gglob_dir,
plot_dir=out_viz_dir,
subset_indivs=subset_indivs)
curr_dCGHs = self.get_curr_chr_dCGHs(chr, dCGHs)
wnd_starts, wnd_ends = indiv_DTS.get_wnds_by_chr(chr)
t = time.time()
n_assessed = -1
for overlap_cluster in overlapping_calls:
overlap_cutoff = 0.85
#overlap_cutoff = 0.75
n_assessed += 1
resolved_calls = overlap_cluster.overlap_resolve(
overlap_cutoff,
ll_cutoff,
tbx_dups,
min_overlapping=min_overlapping)
if len(resolved_calls) == 0: continue
variable_clusts = []
for clust in resolved_calls:
"""
now take all these resolved calls,
and genotype to ensure this seg is var
"""
d = self.get_delta(clust, wnd_starts, wnd_ends, curr_dCGHs,
indiv_cps, ref_cps)
if clust.size == 1:
print("skipping single call cluster")
continue
if d > 1.0:
variable_clusts.append(clust)
elif d > min_d:
X, idx_s, idx_e = g.get_gt_matrix(
chr, clust.get_med_start(), clust.get_med_end())
print(X)
gX = g.GMM_genotype(X)
if gX.gmm.n_components > 1 and gX.is_var(indiv_id, g):
variable_clusts.append(clust)
overlapping_call_clusts = get_overlapping_call_clusts(
variable_clusts)
for clust in overlapping_call_clusts:
final_call = self.get_final_call(clust)
final_calls.append(final_call)
"""
if min(overlap_cluster.all_starts)>14454554 and min(overlap_cluster.all_starts)<14933135:
self.plot_call(clust, wnd_starts, wnd_ends, curr_dCGHs, indiv_cps, ref_cps)
raw_input()
"""
#overlapping_call_clusts = get_overlapping_call_clusts(resolved_calls, flatten = True)
print("n-final_calls:%d n_assessed:%d time_elapsed:%fs" %
(len(final_calls), n_assessed, time.time() - t))
print("done", file=stderr)
print("%d calls with likelihood <%f" % (len(final_calls), ll_cutoff),
file=stderr)
return final_calls
#colors = ["red", "chocolate", "darkgreen", "gold", "darkcyan", "darkblue", "darkorchid"] + list(plt.cm.Greys(np.linspace(0.5, 1, 2)))
elif args.pop_color_scheme == "primate":
label = "species_label"
n_colors = len(pop_data[label].unique().tolist())
colors = ["royalblue"] + list(plt.cm.Greys(np.linspace(1, 0.25, 4)))
else:
if args.custom_label is None:
print "Must specify custom label for custom color scheme"
sys.exit(1)
label = args.custom_label
n_colors = len(pop_data[label].unique().tolist())
colors = list(plt.cm.hsv(np.linspace(0, 0.9, n_colors-2))) + list(plt.cm.Greys(np.linspace(0.5, 1, 2)))
indivs = pop_data.sample.unique().tolist()
g = gt.genotyper(args.contig, gglob_dir = args.gglob_dir, plot_dir = args.plot_dir, subset_indivs = indivs, fn_fa=args.fn_fa, dup_tabix = args.fn_dup_tabix, GC_inf = args.fn_GC_DTS)
X, idx_s, idx_e = g.get_gt_matrix(args.contig, args.start - args.pad, args.end + args.pad)
starts = g.wnd_starts[idx_s:idx_e]
data = pd.DataFrame(X, index = g.indivs, columns = map(str, starts))
pop_data = pop_data.loc[pop_data["sample"].isin(g.indivs)]
unique_labels = pop_data[label].unique().tolist()
pop_data["color"] = pop_data[label].map(lambda x: colors[unique_labels.index(x)])
pop_data["label"] = pop_data[label]
alpha = 0.6
fig = plt.figure(figsize=(24,15))
if args.genes is not None:
genes_h = 0.1
GC_inf = GC_data(args.fn_GC_DTS, args.contig, args.fn_DTS_contigs)
if args.subset_indivs is not None:
indivs = args.subset_indivs
elif args.manifest is not None:
indivs = pd.read_table(args.manifest,
header=0).sample.unique().tolist()
else:
indivs = list(pd.read_json("%s/gglob.idx" % args.gglob_dir).indivs)
# GENOTYPE TIME!
g = gt.genotyper(args.contig,
gglob_dir=args.gglob_dir,
plot_dir=args.plot_dir,
subset_indivs=indivs,
fn_fa=args.fn_fa,
dup_tabix=tbx_dups,
GC_inf=GC_inf)
regions = pd.read_csv(args.fn_regions,
header=None,
delimiter="\t",
index_col=None)
regions.columns = ["chr", "start", "end", "name"]
regions_by_contig = regions[regions['chr'] == args.contig]
nregions = regions_by_contig.shape[0]
FOUT = open(args.fn_out, 'w')
if args.contig == args.header_chr and subset == 0:
FOUT.write("chr\tstart\tend\tname\t%s\n" % ("\t".join(indivs)))
def resolve_overlapping_clusters(self, ll_cutoff,
tbx_dups,
indiv_id,
indiv_DTS,
ref_DTSs,
dCGHs,
gglob_dir,
out_viz_dir,
verbose=False,
min_overlapping=2,
subset_indivs=None,
min_d=0):
"""
resolve overlapping clusters into individual calls
let the calls have likelihoods
1. do the recip overlap cluster - clusters very similar calls w/ similar break-points
2. make sure there are at least 3 calls in there (w/ similar breakpoints)
3. make sure those calls sum to a log likelihood of <3
4. collapse overlaps
5. get the best breakpoints
"""
print >>stderr, "resolving breakpoints..."
final_calls = []
for chr, overlapping_calls in self.overlapping_calls_by_chr.iteritems():
print >>stderr, chr, "%d calls in this chr"%(len(overlapping_calls))
indiv_cps = indiv_DTS.get_cps_by_chr(chr)
ref_cps = {}
for ref, refDTS in ref_DTSs.iteritems():
ref_cps[ref] = refDTS.get_cps_by_chr(chr)
g = genotyper(chr, gglob_dir=gglob_dir, plot_dir=out_viz_dir, subset_indivs = subset_indivs)
curr_dCGHs = self.get_curr_chr_dCGHs(chr, dCGHs)
wnd_starts, wnd_ends = indiv_DTS.get_wnds_by_chr(chr)
t=time.time()
n_assessed=-1
for overlap_cluster in overlapping_calls:
overlap_cutoff = 0.85
#overlap_cutoff = 0.75
n_assessed+=1
resolved_calls = overlap_cluster.overlap_resolve(overlap_cutoff,
ll_cutoff,
tbx_dups,
min_overlapping=min_overlapping)
if len(resolved_calls) == 0: continue
variable_clusts = []
for clust in resolved_calls:
"""
now take all these resolved calls,
and genotype to ensure this seg is var
"""
d = self.get_delta(clust, wnd_starts, wnd_ends, curr_dCGHs, indiv_cps, ref_cps)
if clust.size == 1:
print "skipping single call cluster"
continue
if d > 1.0:
variable_clusts.append(clust)
elif d>min_d:
X, idx_s, idx_e = g.get_gt_matrix(chr, clust.get_med_start(), clust.get_med_end())
gX = g.GMM_genotype(X)
if gX.gmm.n_components>1 and gX.is_var(indiv_id, g):
variable_clusts.append(clust)
overlapping_call_clusts = get_overlapping_call_clusts(variable_clusts)
for clust in overlapping_call_clusts:
final_call = self.get_final_call(clust)
final_calls.append(final_call)
"""
if min(overlap_cluster.all_starts)>14454554 and min(overlap_cluster.all_starts)<14933135:
self.plot_call(clust, wnd_starts, wnd_ends, curr_dCGHs, indiv_cps, ref_cps)
raw_input()
"""
#overlapping_call_clusts = get_overlapping_call_clusts(resolved_calls, flatten = True)
print "n-final_calls:%d n_assessed:%d time_elapsed:%fs"%(len(final_calls), n_assessed, time.time()-t)
print >>stderr, "done"
print >>stderr, "%d calls with likelihood <%f"%(len(final_calls), ll_cutoff)
return final_calls