def test_variants_overlap(self):
v1 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=10)
v2 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=20)
with mock.patch.object(ranges, 'ranges_overlap') as mock_overlap:
mock_overlap.return_value = 'SENTINEL'
self.assertEqual(variant_utils.variants_overlap(v1, v2), 'SENTINEL')
mock_overlap.assert_called_once_with(
variant_utils.variant_range(v1), variant_utils.variant_range(v2))
def test_variants_overlap(self):
v1 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=10)
v2 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=20)
with mock.patch.object(ranges, 'ranges_overlap') as mock_overlap:
mock_overlap.return_value = 'SENTINEL'
self.assertEqual(variant_utils.variants_overlap(v1, v2), 'SENTINEL')
mock_overlap.assert_called_once_with(
variant_utils.variant_range(v1), variant_utils.variant_range(v2))
def test_variant_position_and_range(self):
v1 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=10)
v2 = test_utils.make_variant(chrom='1', alleles=['AGCT', 'C'], start=10)
pos = ranges.make_range('1', 10, 11)
range_ = ranges.make_range('1', 10, 14)
v1_range_tuple = ('1', 10, 11)
v2_range_tuple = ('1', 10, 14)
self.assertEqual(pos, variant_utils.variant_position(v1))
self.assertEqual(pos, variant_utils.variant_position(v2))
self.assertEqual(pos, variant_utils.variant_range(v1))
self.assertEqual(range_, variant_utils.variant_range(v2))
self.assertEqual(v1_range_tuple, variant_utils.variant_range_tuple(v1))
self.assertEqual(v2_range_tuple, variant_utils.variant_range_tuple(v2))
def test_variant_position_and_range(self):
v1 = test_utils.make_variant(chrom='1', alleles=['A', 'C'], start=10)
v2 = test_utils.make_variant(chrom='1', alleles=['AGCT', 'C'], start=10)
pos = ranges.make_range('1', 10, 11)
range_ = ranges.make_range('1', 10, 14)
v1_range_tuple = ('1', 10, 11)
v2_range_tuple = ('1', 10, 14)
self.assertEqual(pos, variant_utils.variant_position(v1))
self.assertEqual(pos, variant_utils.variant_position(v2))
self.assertEqual(pos, variant_utils.variant_range(v1))
self.assertEqual(range_, variant_utils.variant_range(v2))
self.assertEqual(v1_range_tuple, variant_utils.variant_range_tuple(v1))
self.assertEqual(v2_range_tuple, variant_utils.variant_range_tuple(v2))
def _transform_call_variants_output_to_variants(
input_sorted_tfrecord_path, qual_filter, multi_allelic_qual_filter,
sample_name):
"""Yields Variant protos in sorted order from CallVariantsOutput protos.
Variants present in the input TFRecord are converted to Variant protos, with
the following filters applied: 1) variants are omitted if their quality is
lower than the `qual_filter` threshold. 2) multi-allelic variants omit
individual alleles whose qualities are lower than the
`multi_allelic_qual_filter` threshold.
Args:
input_sorted_tfrecord_path: str. TFRecord format file containing sorted
CallVariantsOutput protos.
qual_filter: double. The qual value below which to filter variants.
multi_allelic_qual_filter: double. The qual value below which to filter
multi-allelic variants.
sample_name: str. Sample name to write to VCF file.
Yields:
Variant protos in sorted order representing the CallVariantsOutput calls.
"""
for _, group in itertools.groupby(
io_utils.read_tfrecords(
input_sorted_tfrecord_path, proto=deepvariant_pb2.CallVariantsOutput),
lambda x: variant_utils.variant_range(x.variant)):
outputs = list(group)
canonical_variant, predictions = merge_predictions(
outputs, multi_allelic_qual_filter)
variant = add_call_to_variant(
canonical_variant,
predictions,
qual_filter=qual_filter,
sample_name=sample_name)
yield variant
def _transform_call_variants_output_to_variants(input_sorted_tfrecord_path,
qual_filter,
multi_allelic_qual_filter,
sample_name):
"""Yields Variant protos in sorted order from CallVariantsOutput protos.
Variants present in the input TFRecord are converted to Variant protos, with
the following filters applied: 1) variants are omitted if their quality is
lower than the `qual_filter` threshold. 2) multi-allelic variants omit
individual alleles whose qualities are lower than the
`multi_allelic_qual_filter` threshold.
Args:
input_sorted_tfrecord_path: str. TFRecord format file containing sorted
CallVariantsOutput protos.
qual_filter: double. The qual value below which to filter variants.
multi_allelic_qual_filter: double. The qual value below which to filter
multi-allelic variants.
sample_name: str. Sample name to write to VCF file.
Yields:
Variant protos in sorted order representing the CallVariantsOutput calls.
"""
for _, group in itertools.groupby(
io_utils.read_tfrecords(input_sorted_tfrecord_path,
proto=deepvariant_pb2.CallVariantsOutput),
lambda x: variant_utils.variant_range(x.variant)):
outputs = _sort_grouped_variants(group)
canonical_variant, predictions = merge_predictions(
outputs, multi_allelic_qual_filter)
variant = add_call_to_variant(canonical_variant,
predictions,
qual_filter=qual_filter,
sample_name=sample_name)
yield variant
def _transform_call_variants_output_to_variants(input_sorted_tfrecord_path,
qual_filter,
multi_allelic_qual_filter,
sample_name, group_variants,
use_multiallelic_model):
"""Yields Variant protos in sorted order from CallVariantsOutput protos.
Variants present in the input TFRecord are converted to Variant protos, with
the following filters applied: 1) variants are omitted if their quality is
lower than the `qual_filter` threshold. 2) multi-allelic variants omit
individual alleles whose qualities are lower than the
`multi_allelic_qual_filter` threshold.
Args:
input_sorted_tfrecord_path: str. TFRecord format file containing sorted
CallVariantsOutput protos.
qual_filter: double. The qual value below which to filter variants.
multi_allelic_qual_filter: double. The qual value below which to filter
multi-allelic variants.
sample_name: str. Sample name to write to VCF file.
group_variants: bool. If true, group variants that have same start and end
position.
use_multiallelic_model: if True, use a specialized model for genotype
resolution of multiallelic cases with two alts.
Yields:
Variant protos in sorted order representing the CallVariantsOutput calls.
"""
multiallelic_model = get_multiallelic_model(
use_multiallelic_model=use_multiallelic_model)
group_fn = None
if group_variants:
group_fn = lambda x: variant_utils.variant_range(x.variant)
for _, group in itertools.groupby(
tfrecord.read_tfrecords(
input_sorted_tfrecord_path, proto=deepvariant_pb2.CallVariantsOutput),
group_fn):
outputs = _sort_grouped_variants(group)
canonical_variant, predictions = merge_predictions(
outputs,
multi_allelic_qual_filter,
multiallelic_model=multiallelic_model)
variant = add_call_to_variant(
canonical_variant,
predictions,
qual_filter=qual_filter,
sample_name=sample_name)
yield variant
def _label_grouped_variants(self, variants):
# redacted
# redacted
# they should be computed in the grouping.
span = ranges.span([variant_utils.variant_range(v) for v in variants])
truths = list(
self._get_truth_variants(
ranges.expand(span,
_TRUTH_VARIANTS_QUERY_REGION_EXPANSION_IN_BP)))
if len(truths) > self.max_group_size:
logging.warning((
'Found a large number of variants to label (n_candidates=%d, '
'n_truth=%d) relative to candidate cap of %d. This may make the '
'algorithm very slow.'), len(variants), len(truths),
self.max_group_size)
# redacted
logging.warning(
'Returning all variants with not-confident markers.')
for variant in variants:
yield variant_labeler.VariantLabel(is_confident=False,
genotype=(-1, -1),
variant=variant)
return
ref = self.make_labeler_ref(variants, truths)
labeled_variants = label_variants(variants, truths, ref)
if not labeled_variants:
raise ValueError('Failed to assign labels for variants', variants)
else:
for labeled in labeled_variants:
yield variant_labeler.VariantLabel(
# redacted
# now. Rethink how we establish a variant is confident. Seems like
# it'd be confident if it has a non-ref genotype (as we only
# consider confident truth variants) or if it overlaps the confident
# regions.
is_confident=self._confident_regions.variant_overlaps(
labeled),
genotype=tuple(labeled.calls[0].genotype),
variant=labeled)
def _label_grouped_variants(self, variants):
# redacted
# redacted
# they should be computed in the grouping.
span = ranges.span([variant_utils.variant_range(v) for v in variants])
truths = list(
self._get_truth_variants(
ranges.expand(span, _TRUTH_VARIANTS_QUERY_REGION_EXPANSION_IN_BP)))
if len(truths) > self.max_group_size:
logging.warning(
('Found a large number of variants to label (n_candidates=%d, '
'n_truth=%d) relative to candidate cap of %d. This may make the '
'algorithm very slow.'), len(variants), len(truths),
self.max_group_size)
# redacted
logging.warning('Returning all variants with not-confident markers.')
for variant in variants:
yield variant_labeler.VariantLabel(
is_confident=False, genotype=(-1, -1), variant=variant)
return
ref = self.make_labeler_ref(variants, truths)
labeled_variants = label_variants(variants, truths, ref)
if not labeled_variants:
raise ValueError('Failed to assign labels for variants', variants)
else:
for labeled in labeled_variants:
yield variant_labeler.VariantLabel(
# redacted
# now. Rethink how we establish a variant is confident. Seems like
# it'd be confident if it has a non-ref genotype (as we only
# consider confident truth variants) or if it overlaps the confident
# regions.
is_confident=self._confident_regions.variant_overlaps(labeled),
genotype=tuple(labeled.calls[0].genotype),
variant=labeled)
def query(self, region):
return iter(variant for variant in self.variants
if ranges.ranges_overlap(
variant_utils.variant_range(variant), region))
def query(self, region):
return iter(
variant for variant in self.variants
if ranges.ranges_overlap(variant_utils.variant_range(variant), region)
)
