def run(self):
total_llh = 0.0
for vjins in self.vjins_nts:
assert len(vjins) >= 2 # because it has the last v and the first j
for mobj in re.finditer(self.d_nts, vjins):
# check special case
v_hang_right = 0 if self.v_hang == 0 else 3 - self.v_hang
if len(self.d_nts) == 3:
if (mobj.start() - v_hang_right - 1) % 3 == 0:
continue
elif len(self.d_nts) == 4:
if (mobj.start() - v_hang_right - 1) != 1:
continue
vdins = vjins[: mobj.start()]
djins = vjins[mobj.end(): ]
cdr3_nt = self.v_cdr3_nt + vjins[1:-1] + self.j_cdr3_nt
assert lcommon.nt2aa(cdr3_nt) == self.cdr3_aa
event = lclone.Cdr3Clone(1, cdr3_nt, self.v, self.j, d=self.d,
aa=self.cdr3_aa, vdel=self.vdel, jdel=self.jdel,
d5del=self.d5del, d3del=self.d3del,
vdins=vdins, djins = djins)
llh = ntclone_likelihood(event, self.model)
# addjust empty D llh (TEMPORARY HACK):
if self.d_nts == '':
if self.d == 'TRBD1':
llh += log10(13)
elif self.d == 'TRBD2':
llh += log10(17)
total_llh += 10 ** llh
pickle.dump(total_llh, gzip.open(self.outfile, 'wb'))
def run(self):
llhs = []
jclones = pickle.load(gzip.open(self.infile, 'rb'))
for c in jclones:
clonellh = rcommon.ntclone_likelihood(c, self.model)
llhs.append(clonellh)
sumllh = sum([10 ** llh for llh in llhs])
pickle.dump(sumllh, gzip.open(self.outfile, 'wb'))
def run(self):
llhs = []
events = []
for sam in os.listdir(self.indir):
jfile = os.path.join(self.indir, sam, self.j)
if os.path.exists(jfile):
jclones = pickle.load(gzip.open(jfile, 'rb'))
for c in jclones:
if not rcommon.visited_event(events, c):
events.append(c)
clonellh = rcommon.ntclone_likelihood(c, self.model)
llhs.append(clonellh)
sumllh = sum([10 ** llh for llh in llhs])
pickle.dump(sumllh, gzip.open(self.outfile, 'wb'))
def aaclones_likelihood(clone2sams, model, db_dir, sam2total, group2sams,
outfile, ingroup, outgroup):
f = open(outfile, 'w')
f.write("sample\tnum_ntclones\tprob_observed\n")
for clone, (insams, outsams) in clone2sams.iteritems():
f.write("#%s\n" % clone)
events = []
event_llhs = []
for i, sams in enumerate([insams, outsams]):
if not sams:
continue
sam2ntclones = get_ntclones(clone, sams, db_dir)
f.write("#Group_%d\n" % (i + 1))
for sam, ntclones in sam2ntclones.iteritems():
total = sam2total[sam]
llhoods = []
for ntclone in ntclones:
clonellhood = rcommon.ntclone_likelihood(ntclone, model)
#prob_observed = clonellhood + log10(total)
logmu = log10(total) + clonellhood
prob_observed = poisson.logsf(1, 10 ** logmu) # prob. observing >=1 ntclone
llhoods.append(prob_observed)
if not rcommon.visited_event(events, ntclone):
events.append(ntclone)
event_llhs.append(clonellhood)
#if clonellhood != float(-inf):
# event_llhs.append(clonellhood)
llhoods_str = ",".join(["%f" % llh for llh in llhoods])
f.write("%s\t%d\t%s\n" % (sam, len(ntclones), llhoods_str))
# calc prob to observe the aa clones (sum of all nt events)
if sum([10**llh for llh in event_llhs]) > 0:
aa_llh = log10(sum([10**llh for llh in event_llhs]))
avr_total = (sam2total[ingroup] + sam2total[outgroup]) / 2
avr_logmu = aa_llh + log10(avr_total)
avr_aa_llh = poisson.logsf(1, 10 ** avr_logmu)
f.write("#Clone_log_likelihood: %f, %f\n" % (aa_llh, avr_aa_llh))
ingroup_llh = get_group_likelihood(aa_llh, group2sams[ingroup],
insams, sam2total)
outgroup_llh = get_group_likelihood(aa_llh, group2sams[outgroup],
outsams, sam2total)
f.write("#Ingrp vs Outgrp: %f vs %f\n#\n" % (ingroup_llh, outgroup_llh))
f.close()
def run(self):
#events = []
#event_file = "%s_events" % self.outfile
total_llh = 0.0
model = pickle.load(gzip.open(self.modelfile, 'rb'))
vd_ins_nts = pickle.load(gzip.open(self.vfile, 'rb'))
dj_ins_nts = pickle.load(gzip.open(self.jfile, 'rb'))
#self.logToMaster("vdel: %d; vd_ins_nts: %d" % (self.vdel, len(vd_ins_nts)))
#self.logToMaster("jdel: %d; dj_ins_nts: %d" % (self.jdel, len(dj_ins_nts)))
for vd_ins in vd_ins_nts:
for dj_ins in dj_ins_nts:
cdr3_nt = (self.v_cdr3_nt + vd_ins[1: ] + self.d_cdr3_nt +
dj_ins[: -1] + self.j_cdr3_nt)
assert lcommon.nt2aa(cdr3_nt) == self.cdr3_aa
event = lclone.Cdr3Clone(1, cdr3_nt, self.v, self.j, d=self.d,
aa=self.cdr3_aa, vdel=self.vdel, jdel=self.jdel,
d5del=self.d5del, d3del=self.d3del,
vdins=vd_ins, djins = dj_ins)
#events.append(event)
llh = ntclone_likelihood(event, model)
total_llh += 10 ** llh
pickle.dump(total_llh, gzip.open(self.outfile, 'wb'))
