def test_aggregate_ir(self):
ds = (hl.utils.range_matrix_table(5, 5)
.annotate_globals(g1=5)
.annotate_entries(e1=3))
x = [("col_idx", lambda e: ds.aggregate_cols(e)),
("row_idx", lambda e: ds.aggregate_rows(e))]
for name, f in x:
r = f(hl.struct(x=agg.sum(ds[name]) + ds.g1,
y=agg.filter(ds[name] % 2 != 0, agg.sum(ds[name] + 2)) + ds.g1,
z=agg.sum(ds.g1 + ds[name]) + ds.g1,
mean=agg.mean(ds[name])))
self.assertEqual(convert_struct_to_dict(r), {u'x': 15, u'y': 13, u'z': 40, u'mean': 2.0})
r = f(5)
self.assertEqual(r, 5)
r = f(hl.null(hl.tint32))
self.assertEqual(r, None)
r = f(agg.filter(ds[name] % 2 != 0, agg.sum(ds[name] + 2)) + ds.g1)
self.assertEqual(r, 13)
r = ds.aggregate_entries(agg.filter((ds.row_idx % 2 != 0) & (ds.col_idx % 2 != 0),
agg.sum(ds.e1 + ds.g1 + ds.row_idx + ds.col_idx)) + ds.g1)
self.assertTrue(r, 48)
def test_aggregate_ir(self):
ds = (hl.utils.range_matrix_table(5, 5)
.annotate_globals(g1=5)
.annotate_entries(e1=3))
x = [("col_idx", lambda e: ds.aggregate_cols(e)),
("row_idx", lambda e: ds.aggregate_rows(e))]
for name, f in x:
r = f(hl.struct(x=agg.sum(ds[name]) + ds.g1,
y=agg.filter(ds[name] % 2 != 0, agg.sum(ds[name] + 2)) + ds.g1,
z=agg.sum(ds.g1 + ds[name]) + ds.g1,
mean=agg.mean(ds[name])))
self.assertEqual(convert_struct_to_dict(r), {u'x': 15, u'y': 13, u'z': 40, u'mean': 2.0})
r = f(5)
self.assertEqual(r, 5)
r = f(hl.null(hl.tint32))
self.assertEqual(r, None)
r = f(agg.filter(ds[name] % 2 != 0, agg.sum(ds[name] + 2)) + ds.g1)
self.assertEqual(r, 13)
r = ds.aggregate_entries(agg.filter((ds.row_idx % 2 != 0) & (ds.col_idx % 2 != 0),
agg.sum(ds.e1 + ds.g1 + ds.row_idx + ds.col_idx)) + ds.g1)
self.assertTrue(r, 48)
def test_aggregate_ir(self):
kt = hl.utils.range_table(10).annotate_globals(g1=5)
r = kt.aggregate(hl.struct(x=agg.sum(kt.idx) + kt.g1,
y=agg.filter(kt.idx % 2 != 0, agg.sum(kt.idx + 2)) + kt.g1,
z=agg.sum(kt.g1 + kt.idx) + kt.g1))
self.assertEqual(convert_struct_to_dict(r), {u'x': 50, u'y': 40, u'z': 100})
r = kt.aggregate(5)
self.assertEqual(r, 5)
r = kt.aggregate(hl.null(hl.tint32))
self.assertEqual(r, None)
r = kt.aggregate(agg.filter(kt.idx % 2 != 0, agg.sum(kt.idx + 2)) + kt.g1)
self.assertEqual(r, 40)
def test_aggregate_ir(self):
kt = hl.utils.range_table(10).annotate_globals(g1=5)
r = kt.aggregate(hl.struct(x=agg.sum(kt.idx) + kt.g1,
y=agg.filter(kt.idx % 2 != 0, agg.sum(kt.idx + 2)) + kt.g1,
z=agg.sum(kt.g1 + kt.idx) + kt.g1))
self.assertEqual(convert_struct_to_dict(r), {u'x': 50, u'y': 40, u'z': 100})
r = kt.aggregate(5)
self.assertEqual(r, 5)
r = kt.aggregate(hl.null(hl.tint32))
self.assertEqual(r, None)
r = kt.aggregate(agg.filter(kt.idx % 2 != 0, agg.sum(kt.idx + 2)) + kt.g1)
self.assertEqual(r, 40)
def hwe_normalize(call_expr):
mt = matrix_table_source('hwe_normalize/call_expr', call_expr)
mt = mt.select_entries(__gt=call_expr.n_alt_alleles())
mt = mt.annotate_rows(__AC=agg.sum(mt.__gt),
__n_called=agg.count_where(hl.is_defined(mt.__gt)))
mt = mt.filter_rows((mt.__AC > 0) & (mt.__AC < 2 * mt.__n_called))
n_variants = mt.count_rows()
if n_variants == 0:
raise FatalError(
"hwe_normalize: found 0 variants after filtering out monomorphic sites."
)
info(
f"hwe_normalize: found {n_variants} variants after filtering out monomorphic sites."
)
mt = mt.annotate_rows(__mean_gt=mt.__AC / mt.__n_called)
mt = mt.annotate_rows(__hwe_scaled_std_dev=hl.sqrt(mt.__mean_gt *
(2 - mt.__mean_gt) *
n_variants / 2))
mt = mt.unfilter_entries()
normalized_gt = hl.or_else(
(mt.__gt - mt.__mean_gt) / mt.__hwe_scaled_std_dev, 0.0)
return normalized_gt
def test_select_cols(self):
mt = hl.utils.range_matrix_table(3, 5, n_partitions=4)
mt = mt.annotate_entries(e=mt.col_idx * mt.row_idx)
mt = mt.annotate_globals(g=1)
mt = mt.annotate_cols(sum=agg.sum(mt.e + mt.col_idx + mt.row_idx + mt.g) + mt.col_idx + mt.g,
count=agg.count_where(mt.e % 2 == 0),
foo=agg.count())
result = convert_struct_to_dict(mt.cols().collect()[-2])
self.assertEqual(result, {'col_idx': 3, 'sum': 28, 'count': 2, 'foo': 3})
def test_select_cols(self):
mt = hl.utils.range_matrix_table(3, 5, n_partitions=4)
mt = mt.annotate_entries(e=mt.col_idx * mt.row_idx)
mt = mt.annotate_globals(g=1)
mt = mt.annotate_cols(sum=agg.sum(mt.e + mt.col_idx + mt.row_idx + mt.g) + mt.col_idx + mt.g,
count=agg.count_where(mt.e % 2 == 0),
foo=agg.count())
result = convert_struct_to_dict(mt.cols().collect()[-2])
self.assertEqual(result, {'col_idx': 3, 'sum': 28, 'count': 2, 'foo': 3})
def hwe_normalized_pca(dataset, k=10, compute_loadings=False, as_array=False):
"""Run principal component analysis (PCA) on the Hardy-Weinberg-normalized call matrix.
Examples
--------
>>> eigenvalues, scores, loadings = methods.hwe_normalized_pca(dataset, k=5)
Notes
-----
Variants that are all homozygous reference or all homozygous variant are removed before evaluation.
Parameters
----------
dataset : :class:`.MatrixTable`
Dataset.
k : :obj:`int`
Number of principal components.
compute_loadings : :obj:`bool`
If ``True``, compute row loadings.
as_array : :obj:`bool`
If ``True``, return scores and loadings as an array field. If ``False``, return
one field per element (`PC1`, `PC2`, ... `PCk`).
Returns
-------
(:obj:`list` of :obj:`float`, :class:`.Table`, :class:`.Table`)
List of eigenvalues, table with column scores, table with row loadings.
"""
dataset = dataset.annotate_rows(AC=agg.sum(dataset.GT.num_alt_alleles()),
n_called=agg.count_where(
functions.is_defined(dataset.GT)))
dataset = dataset.filter_rows(
(dataset.AC > 0) & (dataset.AC < 2 * dataset.n_called)).persist()
n_variants = dataset.count_rows()
if n_variants == 0:
raise FatalError(
"Cannot run PCA: found 0 variants after filtering out monomorphic sites."
)
info("Running PCA using {} variants.".format(n_variants))
entry_expr = functions.bind(
dataset.AC / dataset.n_called, lambda mean_gt: functions.cond(
functions.is_defined(dataset.GT), (dataset.GT.num_alt_alleles(
) - mean_gt) / functions.sqrt(mean_gt *
(2 - mean_gt) * n_variants / 2), 0))
result = pca(entry_expr, k, compute_loadings, as_array)
dataset.unpersist()
return result
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)
def test_query(self):
schema = hl.tstruct(a=hl.tint32, b=hl.tint32, c=hl.tint32, d=hl.tint32, e=hl.tstr, f=hl.tarray(hl.tint32))
rows = [{'a': 4, 'b': 1, 'c': 3, 'd': 5, 'e': "hello", 'f': [1, 2, 3]},
{'a': 0, 'b': 5, 'c': 13, 'd': -1, 'e': "cat", 'f': []},
{'a': 4, 'b': 2, 'c': 20, 'd': 3, 'e': "dog", 'f': [5, 6, 7]}]
kt = hl.Table.parallelize(rows, schema)
results = kt.aggregate(hl.Struct(q1=agg.sum(kt.b),
q2=agg.count(),
q3=agg.collect(kt.e),
q4=agg.collect(agg.filter((kt.d >= 5) | (kt.a == 0), kt.e))))
self.assertEqual(results.q1, 8)
self.assertEqual(results.q2, 3)
self.assertEqual(set(results.q3), {"hello", "cat", "dog"})
self.assertEqual(set(results.q4), {"hello", "cat"})
def test_aggregate1(self):
schema = hl.tstruct(a=hl.tint32, b=hl.tint32, c=hl.tint32, d=hl.tint32, e=hl.tstr, f=hl.tarray(hl.tint32))
rows = [{'a': 4, 'b': 1, 'c': 3, 'd': 5, 'e': "hello", 'f': [1, 2, 3]},
{'a': 0, 'b': 5, 'c': 13, 'd': -1, 'e': "cat", 'f': []},
{'a': 4, 'b': 2, 'c': 20, 'd': 3, 'e': "dog", 'f': [5, 6, 7]}]
kt = hl.Table.parallelize(rows, schema)
results = kt.aggregate(hl.Struct(q1=agg.sum(kt.b),
q2=agg.count(),
q3=agg.collect(kt.e),
q4=agg.filter((kt.d >= 5) | (kt.a == 0), agg.collect(kt.e)),
q5=agg.explode(lambda elt: agg.mean(elt), kt.f)))
self.assertEqual(results.q1, 8)
self.assertEqual(results.q2, 3)
self.assertEqual(set(results.q3), {"hello", "cat", "dog"})
self.assertEqual(set(results.q4), {"hello", "cat"})
self.assertAlmostEqual(results.q5, 4)
def _make_tsm_from_call(call_expr,
block_size,
mean_center=False,
hwe_normalize=False):
mt = matrix_table_source('_make_tsm/entry_expr', call_expr)
mt = mt.select_entries(__gt=call_expr.n_alt_alleles())
if mean_center or hwe_normalize:
mt = mt.annotate_rows(__AC=agg.sum(mt.__gt),
__n_called=agg.count_where(hl.is_defined(
mt.__gt)))
mt = mt.filter_rows((mt.__AC > 0) & (mt.__AC < 2 * mt.__n_called))
n_variants = mt.count_rows()
if n_variants == 0:
raise FatalError(
"_make_tsm: found 0 variants after filtering out monomorphic sites."
)
info(
f"_make_tsm: found {n_variants} variants after filtering out monomorphic sites."
)
mt = mt.annotate_rows(__mean_gt=mt.__AC / mt.__n_called)
mt = mt.unfilter_entries()
mt = mt.select_entries(__x=hl.or_else(mt.__gt - mt.__mean_gt, 0.0))
if hwe_normalize:
mt = mt.annotate_rows(
__hwe_scaled_std_dev=hl.sqrt(mt.__mean_gt *
(2 - mt.__mean_gt) * n_variants /
2))
mt = mt.select_entries(__x=mt.__x / mt.__hwe_scaled_std_dev)
else:
mt = mt.select_entries(__x=mt.__gt)
A, ht = mt_to_table_of_ndarray(mt.__x,
block_size,
return_checkpointed_table_also=True)
A = A.persist()
return TallSkinnyMatrix(A, A.ndarray, ht, list(mt.col_key))
def test_grm(self):
tolerance = 0.001
def load_id_file(path):
ids = []
with hl.hadoop_open(path) as f:
for l in f:
r = l.strip().split('\t')
self.assertEqual(len(r), 2)
ids.append(r[1])
return ids
def load_rel(ns, path):
rel = np.zeros((ns, ns))
with hl.hadoop_open(path) as f:
for i, l in enumerate(f):
for j, n in enumerate(map(float, l.strip().split('\t'))):
rel[i, j] = n
self.assertEqual(j, i)
self.assertEqual(i, ns - 1)
return rel
def load_grm(ns, nv, path):
m = np.zeros((ns, ns))
with utils.hadoop_open(path) as f:
i = 0
for l in f:
row = l.strip().split('\t')
self.assertEqual(int(row[2]), nv)
m[int(row[0]) - 1, int(row[1]) - 1] = float(row[3])
i += 1
self.assertEqual(i, ns * (ns + 1) / 2)
return m
def load_bin(ns, path):
m = np.zeros((ns, ns))
with utils.hadoop_open(path, 'rb') as f:
for i in range(ns):
for j in range(i + 1):
b = f.read(4)
self.assertEqual(len(b), 4)
m[i, j] = unpack('<f', bytearray(b))[0]
left = f.read()
self.assertEqual(len(left), 0)
return m
b_file = utils.new_temp_file(prefix="plink")
rel_file = utils.new_temp_file(prefix="test", suffix="rel")
rel_id_file = utils.new_temp_file(prefix="test", suffix="rel.id")
grm_file = utils.new_temp_file(prefix="test", suffix="grm")
grm_bin_file = utils.new_temp_file(prefix="test", suffix="grm.bin")
grm_nbin_file = utils.new_temp_file(prefix="test", suffix="grm.N.bin")
dataset = self.get_dataset()
n_samples = dataset.count_cols()
dataset = dataset.annotate_rows(AC=agg.sum(dataset.GT.n_alt_alleles()),
n_called=agg.count_where(hl.is_defined(dataset.GT)))
dataset = dataset.filter_rows((dataset.AC > 0) & (dataset.AC < 2 * dataset.n_called))
dataset = dataset.filter_rows(dataset.n_called == n_samples).persist()
hl.export_plink(dataset, b_file, id=dataset.s)
sample_ids = [row.s for row in dataset.cols().select('s').collect()]
n_variants = dataset.count_rows()
self.assertGreater(n_variants, 0)
grm = hl.genetic_relatedness_matrix(dataset)
grm.export_id_file(rel_id_file)
############
### rel
p_file = utils.new_temp_file(prefix="plink")
syscall('''plink --bfile {} --make-rel --out {}'''
.format(utils.uri_path(b_file), utils.uri_path(p_file)), shell=True, stdout=DEVNULL, stderr=DEVNULL)
self.assertEqual(load_id_file(p_file + ".rel.id"), sample_ids)
grm.export_rel(rel_file)
self.assertEqual(load_id_file(rel_id_file), sample_ids)
self.assertTrue(np.allclose(load_rel(n_samples, p_file + ".rel"),
load_rel(n_samples, rel_file),
atol=tolerance))
############
### gcta-grm
p_file = utils.new_temp_file(prefix="plink")
syscall('''plink --bfile {} --make-grm-gz --out {}'''
.format(utils.uri_path(b_file), utils.uri_path(p_file)), shell=True, stdout=DEVNULL, stderr=DEVNULL)
self.assertEqual(load_id_file(p_file + ".grm.id"), sample_ids)
grm.export_gcta_grm(grm_file)
self.assertTrue(np.allclose(load_grm(n_samples, n_variants, p_file + ".grm.gz"),
load_grm(n_samples, n_variants, grm_file),
atol=tolerance))
############
### gcta-grm-bin
p_file = utils.new_temp_file(prefix="plink")
syscall('''plink --bfile {} --make-grm-bin --out {}'''
.format(utils.uri_path(b_file), utils.uri_path(p_file)), shell=True, stdout=DEVNULL, stderr=DEVNULL)
self.assertEqual(load_id_file(p_file + ".grm.id"), sample_ids)
grm.export_gcta_grm_bin(grm_bin_file, grm_nbin_file)
self.assertTrue(np.allclose(load_bin(n_samples, p_file + ".grm.bin"),
load_bin(n_samples, grm_bin_file),
atol=tolerance))
self.assertTrue(np.allclose(load_bin(n_samples, p_file + ".grm.N.bin"),
load_bin(n_samples, grm_nbin_file),
atol=tolerance))
def grm(dataset):
"""Compute the Genetic Relatedness Matrix (GRM).
.. include:: ../_templates/req_tvariant.rst
.. include:: ../_templates/req_biallelic.rst
Examples
--------
>>> km = methods.grm(dataset)
Notes
-----
The genetic relationship matrix (GRM) :math:`G` encodes genetic correlation
between each pair of samples. It is defined by :math:`G = MM^T` where
:math:`M` is a standardized version of the genotype matrix, computed as
follows. Let :math:`C` be the :math:`n \\times m` matrix of raw genotypes
in the variant dataset, with rows indexed by :math:`n` samples and columns
indexed by :math:`m` bialellic autosomal variants; :math:`C_{ij}` is the
number of alternate alleles of variant :math:`j` carried by sample
:math:`i`, which can be 0, 1, 2, or missing. For each variant :math:`j`,
the sample alternate allele frequency :math:`p_j` is computed as half the
mean of the non-missing entries of column :math:`j`. Entries of :math:`M`
are then mean-centered and variance-normalized as
.. math::
M_{ij} = \\frac{C_{ij}-2p_j}{\sqrt{2p_j(1-p_j)m}},
with :math:`M_{ij} = 0` for :math:`C_{ij}` missing (i.e. mean genotype
imputation). This scaling normalizes genotype variances to a common value
:math:`1/m` for variants in Hardy-Weinberg equilibrium and is further
motivated in the paper `Patterson, Price and Reich, 2006
<http://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.0020190>`__.
(The resulting amplification of signal from the low end of the allele
frequency spectrum will also introduce noise for rare variants; common
practice is to filter out variants with minor allele frequency below some
cutoff.) The factor :math:`1/m` gives each sample row approximately unit
total variance (assuming linkage equilibrium) so that the diagonal entries
of the GRM are approximately 1. Equivalently,
.. math::
G_{ik} = \\frac{1}{m} \\sum_{j=1}^m \\frac{(C_{ij}-2p_j)(C_{kj}-2p_j)}{2 p_j (1-p_j)}
Warning
-------
Since Hardy-Weinberg normalization cannot be applied to variants that
contain only reference alleles or only alternate alleles, all such variants
are removed prior to calcularing the GRM.
Parameters
----------
dataset : :class:`.MatrixTable`
Dataset to sample from.
Returns
-------
:class:`genetics.KinshipMatrix`
Genetic Relatedness Matrix for all samples.
:rtype:
"""
dataset = dataset.annotate_rows(AC=agg.sum(dataset.GT.num_alt_alleles()),
n_called=agg.count_where(
functions.is_defined(dataset.GT)))
dataset = dataset.filter_rows(
(dataset.AC > 0) & (dataset.AC < 2 * dataset.n_called)).persist()
n_variants = dataset.count_rows()
if n_variants == 0:
raise FatalError(
"Cannot run GRM: found 0 variants after filtering out monomorphic sites."
)
info("Computing GRM using {} variants.".format(n_variants))
normalized_genotype_expr = functions.bind(
dataset.AC / dataset.n_called, lambda mean_gt: functions.cond(
functions.is_defined(dataset.GT), (dataset.GT.num_alt_alleles(
) - mean_gt) / functions.sqrt(mean_gt *
(2 - mean_gt) * n_variants / 2), 0))
bm = BlockMatrix.from_matrix_table(normalized_genotype_expr)
dataset.unpersist()
grm = bm.T.dot(bm)
return KinshipMatrix._from_block_matrix(
dataset.colkey_schema, grm,
[row.s
for row in dataset.cols_table().select('s').collect()], n_variants)