def test_annotate(self):
vds = self.get_vds()
vds = vds.annotate_globals(foo=5)
self.assertEqual(vds.globals.dtype, hl.tstruct(foo=hl.tint32))
vds = vds.annotate_rows(x1=agg.count(),
x2=agg.fraction(False),
x3=agg.count_where(True),
x4=vds.info.AC + vds.foo)
vds = vds.annotate_cols(apple=6)
vds = vds.annotate_cols(y1=agg.count(),
y2=agg.fraction(False),
y3=agg.count_where(True),
y4=vds.foo + vds.apple)
expected_schema = hl.tstruct(s=hl.tstr, apple=hl.tint32, y1=hl.tint64, y2=hl.tfloat64, y3=hl.tint64,
y4=hl.tint32)
self.assertTrue(schema_eq(vds.col.dtype, expected_schema),
"expected: " + str(vds.col.dtype) + "\nactual: " + str(expected_schema))
vds = vds.select_entries(z1=vds.x1 + vds.foo,
z2=vds.x1 + vds.y1 + vds.foo)
self.assertTrue(schema_eq(vds.entry.dtype, hl.tstruct(z1=hl.tint64, z2=hl.tint64)))
def test_annotate(self):
vds = self.get_vds()
vds = vds.annotate_globals(foo=5)
self.assertEqual(vds.globals.dtype, hl.tstruct(foo=hl.tint32))
vds = vds.annotate_rows(x1=agg.count(),
x2=agg.fraction(False),
x3=agg.count_where(True),
x4=vds.info.AC + vds.foo)
vds = vds.annotate_cols(apple=6)
vds = vds.annotate_cols(y1=agg.count(),
y2=agg.fraction(False),
y3=agg.count_where(True),
y4=vds.foo + vds.apple)
expected_schema = hl.tstruct(s=hl.tstr,
apple=hl.tint32,
y1=hl.tint64,
y2=hl.tfloat64,
y3=hl.tint64,
y4=hl.tint32)
self.assertTrue(
schema_eq(vds.col.dtype, expected_schema), "expected: " +
str(vds.col.dtype) + "\nactual: " + str(expected_schema))
vds = vds.select_entries(z1=vds.x1 + vds.foo,
z2=vds.x1 + vds.y1 + vds.foo)
self.assertTrue(
schema_eq(vds.entry.dtype, hl.tstruct(z1=hl.tint64, z2=hl.tint64)))
def test_aggregate2(self):
schema = hl.tstruct(status=hl.tint32, GT=hl.tcall, qPheno=hl.tint32)
rows = [{'status': 0, 'GT': hl.Call([0, 0]), 'qPheno': 3},
{'status': 0, 'GT': hl.Call([0, 1]), 'qPheno': 13}]
kt = hl.Table.parallelize(rows, schema)
result = convert_struct_to_dict(
kt.group_by(status=kt.status)
.aggregate(
x1=agg.collect(kt.qPheno * 2),
x2=agg.explode(lambda elt: agg.collect(elt), [kt.qPheno, kt.qPheno + 1]),
x3=agg.min(kt.qPheno),
x4=agg.max(kt.qPheno),
x5=agg.sum(kt.qPheno),
x6=agg.product(hl.int64(kt.qPheno)),
x7=agg.count(),
x8=agg.count_where(kt.qPheno == 3),
x9=agg.fraction(kt.qPheno == 1),
x10=agg.stats(hl.float64(kt.qPheno)),
x11=agg.hardy_weinberg_test(kt.GT),
x13=agg.inbreeding(kt.GT, 0.1),
x14=agg.call_stats(kt.GT, ["A", "T"]),
x15=agg.collect(hl.Struct(a=5, b="foo", c=hl.Struct(banana='apple')))[0],
x16=agg.collect(hl.Struct(a=5, b="foo", c=hl.Struct(banana='apple')).c.banana)[0],
x17=agg.explode(lambda elt: agg.collect(elt), hl.null(hl.tarray(hl.tint32))),
x18=agg.explode(lambda elt: agg.collect(elt), hl.null(hl.tset(hl.tint32))),
x19=agg.take(kt.GT, 1, ordering=-kt.qPheno)
).take(1)[0])
expected = {u'status': 0,
u'x13': {u'n_called': 2, u'expected_homs': 1.64, u'f_stat': -1.777777777777777,
u'observed_homs': 1},
u'x14': {u'AC': [3, 1], u'AF': [0.75, 0.25], u'AN': 4, u'homozygote_count': [1, 0]},
u'x15': {u'a': 5, u'c': {u'banana': u'apple'}, u'b': u'foo'},
u'x10': {u'min': 3.0, u'max': 13.0, u'sum': 16.0, u'stdev': 5.0, u'n': 2, u'mean': 8.0},
u'x8': 1, u'x9': 0.0, u'x16': u'apple',
u'x11': {u'het_freq_hwe': 0.5, u'p_value': 0.5},
u'x2': [3, 4, 13, 14], u'x3': 3, u'x1': [6, 26], u'x6': 39, u'x7': 2, u'x4': 13, u'x5': 16,
u'x17': [],
u'x18': [],
u'x19': [hl.Call([0, 1])]}
self.maxDiff = None
self.assertDictEqual(result, expected)
def test_aggregate2(self):
schema = hl.tstruct(status=hl.tint32, GT=hl.tcall, qPheno=hl.tint32)
rows = [{'status': 0, 'GT': hl.Call([0, 0]), 'qPheno': 3},
{'status': 0, 'GT': hl.Call([0, 1]), 'qPheno': 13}]
kt = hl.Table.parallelize(rows, schema)
result = convert_struct_to_dict(
kt.group_by(status=kt.status)
.aggregate(
x1=agg.collect(kt.qPheno * 2),
x2=agg.explode(lambda elt: agg.collect(elt), [kt.qPheno, kt.qPheno + 1]),
x3=agg.min(kt.qPheno),
x4=agg.max(kt.qPheno),
x5=agg.sum(kt.qPheno),
x6=agg.product(hl.int64(kt.qPheno)),
x7=agg.count(),
x8=agg.count_where(kt.qPheno == 3),
x9=agg.fraction(kt.qPheno == 1),
x10=agg.stats(hl.float64(kt.qPheno)),
x11=agg.hardy_weinberg_test(kt.GT),
x13=agg.inbreeding(kt.GT, 0.1),
x14=agg.call_stats(kt.GT, ["A", "T"]),
x15=agg.collect(hl.Struct(a=5, b="foo", c=hl.Struct(banana='apple')))[0],
x16=agg.collect(hl.Struct(a=5, b="foo", c=hl.Struct(banana='apple')).c.banana)[0],
x17=agg.explode(lambda elt: agg.collect(elt), hl.null(hl.tarray(hl.tint32))),
x18=agg.explode(lambda elt: agg.collect(elt), hl.null(hl.tset(hl.tint32))),
x19=agg.take(kt.GT, 1, ordering=-kt.qPheno)
).take(1)[0])
expected = {u'status': 0,
u'x13': {u'n_called': 2, u'expected_homs': 1.64, u'f_stat': -1.777777777777777,
u'observed_homs': 1},
u'x14': {u'AC': [3, 1], u'AF': [0.75, 0.25], u'AN': 4, u'homozygote_count': [1, 0]},
u'x15': {u'a': 5, u'c': {u'banana': u'apple'}, u'b': u'foo'},
u'x10': {u'min': 3.0, u'max': 13.0, u'sum': 16.0, u'stdev': 5.0, u'n': 2, u'mean': 8.0},
u'x8': 1, u'x9': 0.0, u'x16': u'apple',
u'x11': {u'het_freq_hwe': 0.5, u'p_value': 0.5},
u'x2': [3, 4, 13, 14], u'x3': 3, u'x1': [6, 26], u'x6': 39, u'x7': 2, u'x4': 13, u'x5': 16,
u'x17': [],
u'x18': [],
u'x19': [hl.Call([0, 1])]}
self.maxDiff = None
self.assertDictEqual(result, expected)
# COMMAND ----------
mt = mt.filter_cols((mt.sample_qc.dp_stats.mean >= 4)
& (mt.sample_qc.call_rate >= 0.97))
print('After filter, %d/284 samples remain.' % mt.count_cols())
# COMMAND ----------
# MAGIC %md Next is genotype QC. To start, we'll print the post-sample-QC call rate. It's actually gone _up_ since the initial summary - dropping low-quality samples disproportionally removed missing genotypes.
# MAGIC
# MAGIC Import the `hail.expr.functions` module.
# COMMAND ----------
call_rate = mt.aggregate_entries(agg.fraction(hl.is_defined(mt.GT)))
print('pre QC call rate is %.3f' % call_rate)
# COMMAND ----------
# MAGIC %md It's a good idea to filter out genotypes where the reads aren't where they should be: if we find a genotype called homozygous reference with >10% alternate reads, a genotype called homozygous alternate with >10% reference reads, or a genotype called heterozygote without a ref / alt balance near 1:1, it is likely to be an error.
# COMMAND ----------
ab = mt.AD[1] / hl.sum(mt.AD)
filter_condition_ab = ((mt.GT.is_hom_ref() & (ab <= 0.1)) |
(mt.GT.is_het() & (ab >= 0.25) &
(ab <= 0.75)) | (mt.GT.is_hom_var() & (ab >= 0.9)))
mt = mt.filter_entries(filter_condition_ab)
