class JointModelRunner(ModelRunner):
def run(self, args):
self.reader = JointCountsReader(args.jcnt_file_name)
self.writer = JointSnvMixWriter(args.jsm_file_name)
ModelRunner.run(self, args)
def _train(self, args):
if args.subsample_size > 0:
counts = self._subsample(args.subsample_size)
else:
counts = self.reader.get_counts()
self.priors_parser.load_from_file(args.priors_file)
self.priors = self.priors_parser.to_dict()
self._write_priors()
data = JointData(counts)
self.parameters = self.model.train(data, self.priors,
args.max_iters, args.convergence_threshold)
def _classify_chromosome(self, chr_name):
counts = self.reader.get_counts(chr_name)
jcnt_rows = self.reader.get_rows(chr_name)
end = self.reader.get_chr_size(chr_name)
n = int(1e5)
start = 0
stop = min(n, end)
while start < end:
sub_counts = counts[start:stop]
sub_rows = jcnt_rows[start:stop]
data = JointData(sub_counts)
resp = self.model.classify(data, self.parameters)
self.writer.write_data(chr_name, sub_rows, resp)
start = stop
stop = min(stop + n, end)
def run(self, args):
self.reader = JointCountsReader(args.jcnt_file_name)
self.writer = JointSnvMixWriter(args.jsm_file_name)
ModelRunner.run(self, args)
class ChromosomeModelRunner(ModelRunner):
def run(self, args):
self.reader = JointCountsReader(args.jcnt_file_name)
self.writer = JointSnvMixWriter(args.jsm_file_name)
ModelRunner.run(self, args)
def _train(self, args):
self.priors_parser.load_from_file(args.priors_file)
self.priors = self.priors_parser.to_dict()
self._write_priors()
chr_list = self.reader.get_chr_list()
self.parameters = {}
for chr_name in sorted(chr_list):
print chr_name
if args.subsample_size > 0:
counts = self._chrom_subsample(chr_name, args.subsample_size)
else:
counts = self.reader.get_counts(chr_name)
data = self.data_class(counts)
self.parameters[chr_name] = self.model.train(data, self.priors,
args.max_iters, args.convergence_threshold)
def _classify_chromosome(self, chr_name):
counts = self.reader.get_counts(chr_name)
jcnt_rows = self.reader.get_rows(chr_name)
end = self.reader.get_chr_size(chr_name)
n = int(1e5)
start = 0
stop = min(n, end)
while start < end:
sub_counts = counts[start:stop]
sub_rows = jcnt_rows[start:stop]
data = self.data_class(sub_counts)
resp = self.model.classify(data, self.parameters[chr_name])
self.writer.write_data(chr_name, sub_rows, resp)
start = stop
stop = min(stop + n, end)
def _chrom_subsample(self, chr_name, sample_size):
chr_size = self.reader.get_chr_size(chr_name=chr_name)
sample_size = min(chr_size, sample_size)
chr_sample_indices = random.sample(xrange(chr_size), sample_size)
chr_counts = self.reader.get_counts(chr_name)
chr_sample = chr_counts[chr_sample_indices]
return chr_sample
class IndependentModelRunner(ModelRunner):
def run(self, args):
self.reader = JointCountsReader(args.jcnt_file_name)
self.writer = JointSnvMixWriter(args.jsm_file_name)
ModelRunner.run(self, args)
def _train(self, args):
if args.subsample_size > 0:
counts = self._subsample(args.subsample_size)
else:
counts = self.reader.get_counts()
self.priors_parser.load_from_file(args.priors_file)
self.priors = self.priors_parser.to_dict()
self._write_priors()
self.parameters = {}
for genome in constants.genomes:
data = IndependentData(counts, genome)
self.parameters[genome] = self.model.train(data, self.priors[genome],
args.max_iters, args.convergence_threshold)
def _classify_chromosome(self, chr_name):
counts = self.reader.get_counts(chr_name)
jcnt_rows = self.reader.get_rows(chr_name)
end = self.reader.get_chr_size(chr_name)
n = int(1e5)
start = 0
stop = min(n, end)
while start < end:
sub_counts = counts[start:stop]
sub_rows = jcnt_rows[start:stop]
indep_resp = {}
for genome in constants.genomes:
data = IndependentData(sub_counts, genome)
indep_resp[genome] = self.model.classify(data, self.parameters[genome])
joint_resp = self._get_joint_responsibilities(indep_resp)
self.writer.write_data(chr_name, sub_rows, joint_resp)
start = stop
stop = min(stop + n, end)
def _get_joint_responsibilities(self, resp):
normal_resp = np.log(resp['normal'])
tumour_resp = np.log(resp['tumour'])
n = normal_resp.shape[0]
nclass_normal = normal_resp.shape[1]
column_shape = (n, 1)
log_resp = []
for i in range(nclass_normal):
log_resp.append(normal_resp[:, i].reshape(column_shape) + tumour_resp)
log_resp = np.hstack(log_resp)
resp = np.exp(log_resp)
return resp