def test_runs_reproducible(self):
"""Makes sure that generate_trained_model returns the expected proto."""
output_model_proto = tempfile.NamedTemporaryFile(mode='w',
dir=FLAGS.test_tmpdir,
delete=False)
fname = output_model_proto.name
generate_trained_model.generate_trained_model_runner(
truth_variants=testdata.TRUTH_VARIANTS_VCF,
reads=testdata.CHR20_BAM,
ref=testdata.CHR20_FASTA,
output_model_proto=fname,
output_model_pckl=None,
exclude_contig=None,
from_contig='chr20',
random_seed=42,
indel_weight=1)
with tf.gfile.GFile(fname) as f:
window_selector_model = text_format.Parse(
f.read(), realigner_pb2.WindowSelectorModel())
# Hardcoded value obtained from a "golden" run.
expected = realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.
ALLELE_COUNT_LINEAR,
allele_count_linear_model=(
realigner_pb2.WindowSelectorModel.AlleleCountLinearModel(
bias=0.0259438883513,
coeff_soft_clip=0.00196795910597,
coeff_substitution=-0.545202672482,
coeff_insertion=0.267441004515,
coeff_deletion=0.211069211364,
coeff_reference=0.191676750779,
decision_boundary=3.0)))
self.assertEqual(window_selector_model, expected)
def setUp(self):
window_selector_model = realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.VARIANT_READS,
variant_reads_model=realigner_pb2.WindowSelectorModel.
VariantReadsThresholdModel(min_num_supporting_reads=1,
max_num_supporting_reads=10))
self.config = realigner_pb2.WindowSelectorOptions(
min_mapq=20,
min_base_quality=20,
min_windows_distance=4,
region_expansion_in_bp=20,
window_selector_model=window_selector_model)
def model_to_proto(model):
"""Returns an allele count-based linear WindowSelectorModel."""
# AFAIK sklearn does not provide a way to extract the coefficients based on
# the columns of its input, we thus have to use the fact we know their order.
allele_count_linear_model = (
realigner_pb2.WindowSelectorModel.AlleleCountLinearModel(
bias=model.intercept_[0],
coeff_soft_clip=model.coef_[0][5],
coeff_substitution=model.coef_[0][2],
coeff_insertion=model.coef_[0][3],
coeff_deletion=model.coef_[0][4],
coeff_reference=model.coef_[0][1],
decision_boundary=_DEFAULT_THRESHOLD))
return realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.ALLELE_COUNT_LINEAR,
allele_count_linear_model=allele_count_linear_model)
def setUp(self):
window_selector_model = realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.ALLELE_COUNT_LINEAR,
allele_count_linear_model=realigner_pb2.WindowSelectorModel.
AlleleCountLinearModel(bias=0,
coeff_soft_clip=0,
coeff_substitution=-0.5,
coeff_insertion=1,
coeff_deletion=1,
coeff_reference=-0.5,
decision_boundary=0))
self.config = realigner_pb2.WindowSelectorOptions(
min_mapq=20,
min_base_quality=20,
min_windows_distance=4,
region_expansion_in_bp=20,
window_selector_model=window_selector_model)
def window_selector_config(flags_obj):
"""Creates a WindowSelectorOptions proto based on input and default settings.
Args:
flags_obj: configuration FLAGS.
Returns:
realigner_pb2.WindowSelector protobuf.
Raises:
ValueError: If either ws_{min,max}_supporting_reads are set and
ws_use_window_selector_model is True.
Or if ws_window_selector_model > ws_max_num_supporting_reads.
Or if ws_use_window_selector_model is False and
ws_window_selector_model is not None.
"""
if not flags_obj.ws_use_window_selector_model:
if flags_obj.ws_window_selector_model is not None:
raise ValueError('Cannot specify a ws_window_selector_model '
'if ws_use_window_selector_model is False.')
min_num_supporting_reads = (
_DEFAULT_MIN_SUPPORTING_READS
if flags_obj.ws_min_num_supporting_reads == _UNSET_WS_INT_FLAG else
flags_obj.ws_min_num_supporting_reads)
max_num_supporting_reads = (
_DEFAULT_MAX_SUPPORTING_READS
if flags_obj.ws_max_num_supporting_reads == _UNSET_WS_INT_FLAG else
flags_obj.ws_max_num_supporting_reads)
window_selector_model = realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.VARIANT_READS,
variant_reads_model=realigner_pb2.WindowSelectorModel.
VariantReadsThresholdModel(
min_num_supporting_reads=min_num_supporting_reads,
max_num_supporting_reads=max_num_supporting_reads))
else:
if flags_obj.ws_min_num_supporting_reads != _UNSET_WS_INT_FLAG:
raise ValueError('Cannot use both ws_min_num_supporting_reads and '
'ws_use_window_selector_model flags.')
if flags_obj.ws_max_num_supporting_reads != _UNSET_WS_INT_FLAG:
raise ValueError('Cannot use both ws_max_num_supporting_reads and '
'ws_use_window_selector_model flags.')
if flags_obj.ws_window_selector_model is None:
window_selector_model = _ALLELE_COUNT_LINEAR_MODEL_DEFAULT
else:
with tf.gfile.GFile(flags_obj.ws_window_selector_model) as f:
window_selector_model = text_format.Parse(
f.read(), realigner_pb2.WindowSelectorModel())
if (window_selector_model.model_type ==
realigner_pb2.WindowSelectorModel.VARIANT_READS):
model = window_selector_model.variant_reads_model
if model.max_num_supporting_reads < model.min_num_supporting_reads:
raise ValueError('ws_min_supporting_reads should be smaller than '
'ws_max_supporting_reads.')
ws_config = realigner_pb2.WindowSelectorOptions(
min_mapq=flags_obj.ws_min_mapq,
min_base_quality=flags_obj.ws_min_base_quality,
min_windows_distance=flags_obj.ws_min_windows_distance,
max_window_size=flags_obj.ws_max_window_size,
region_expansion_in_bp=flags_obj.ws_region_expansion_in_bp,
window_selector_model=window_selector_model)
return ws_config
'alignment.')
flags.DEFINE_integer('kmer_size', 32,
'K-mer size for fast pass alinger reads index.')
# Margin added to the reference sequence for the aligner module.
_REF_ALIGN_MARGIN = 20
_DEFAULT_MIN_SUPPORTING_READS = 2
_DEFAULT_MAX_SUPPORTING_READS = 300
_ALLELE_COUNT_LINEAR_MODEL_DEFAULT = realigner_pb2.WindowSelectorModel(
model_type=realigner_pb2.WindowSelectorModel.ALLELE_COUNT_LINEAR,
allele_count_linear_model=realigner_pb2.WindowSelectorModel.
AlleleCountLinearModel(
bias=-0.683379,
coeff_soft_clip=2.997000,
coeff_substitution=-0.086644,
coeff_insertion=2.493585,
coeff_deletion=1.795914,
coeff_reference=-0.059787,
decision_boundary=3))
# ---------------------------------------------------------------------------
# Set configuration settings.
# ---------------------------------------------------------------------------
def window_selector_config(flags_obj):
"""Creates a WindowSelectorOptions proto based on input and default settings.
Args:
