def __handle_sib_founder_family(self, request, family):
'''A) Align the POO phases of all phased quasi-founder siblings.
B) If alignment was successful, phase unphased sibs in the same family using IBD segments
between them and the phased siblings.
Output long IBD segments among the siblings. We only output long segments since we know
sibs share long segments with the parents.'''
problem, params = request.problem, request.params
g, h = problem.g, problem.haplotype
genotyped_children = im.gt.genotyped_children(problem, family)
# Find phased children, identify IBD segments, build paternal haplotypes (by coloring
# child haplotypes), align children POO phases
phased_children = np.array([x for x in genotyped_children if h.fill_fraction(sample=x) >= params.surrogate_parent_fill_threshold])
if params.debug:
print 'Phased children', phased_children
if not phased_children.size:
# No phased children
return
elif phased_children.size == 1:
# If there's just one kid, no need to align
poo, separation = np.array([1.0]), 1.0
else:
# At least two children ==> look for segments and try to align their POO phase.
segments_phased = im.ibd_distant_hap.among_samples_segments(problem, phased_children, request.params)
if params.debug:
print 'Segments among phased children:'
print segments_phased
if segments_phased.length:
pa = im.color.hap_color.hap_colors(list(it.product(sorted(phased_children), im.constants.ALLELES)), segments_phased, max_colors=4)
poo, separation, _, _ = im.color.hap_color.best_hap_alignment_to_colors(pa)
if params.debug:
print pa
print 'POO phases', poo
print 'Separation measure', separation
print 'Parental haplotype coverage', pa.color_sequence_coverage(np.arange(4))
else: separation = 0.0 # no segments, can't align and can't trust kids phased the unphased samples
if np.abs(separation) > params.poo_coloring_measure_threshold:
# Alignment of all kids succeeded
if params.debug:
print 'Alignment successful'
# Align POO phases - flip haplotypes of children with reversed haplotype
# This is a -local- alignment within this family. Still need to globally align all POO phases.
# This is done by the poo module, after phasing and IBD segment calculation+index are done.
flipped_children = phased_children[poo < -params.poo_coloring_measure_threshold]
h_children = problem.h[:, flipped_children, :]
problem.h[:, flipped_children, PATERNAL] = h_children[:, :, MATERNAL]
problem.h[:, flipped_children, MATERNAL] = h_children[:, :, PATERNAL]
# Phase each unphased sib 'sample' using IBD segments between sample and phase_children
# Note: phased sibs these are all full sibs of the proband, not half sibs. So they are
# m=2 meioses away from the proband.
for sample in [x for x in genotyped_children if h.fill_fraction(sample=x, snps=im.gt.where_heterozygous(g, x)) < params.het_fill_threshold]:
relatives = RelativeCollection.from_sibs(sample, phased_children)
if relatives.length > 0:
# segments = sum((im.ibd_distant_hap.hap_segments(IbdProblem(problem, (sample, allele), (sib, allele), None, params))
# for sib, allele in it.product(phased_children, ALLELES)), im.segment.SegmentSet([]))
segments = im.idist.ibd_segments_with_relatives(problem, sample,
relatives.info['index'], params,
prob_ibd_calculator=prob_ibd_hmm)
# Since j is POO-aligned and is a sib of i, segments can only be between (i,a),(j,a).
# Replace the dummy i allele in segments by j's allele.
segments = im.segment.SegmentSet(im.segment.Segment(x.snp, [(x.samples[0][0], x.samples[1][1]), x.samples[1]], x.bp,
error_snps=x.error_snps, confidence=x.confidence, cm=x.cm)
for x in segments)
if params.debug:
print 'Segments after assigning sib alleles to proband haplotypes:'
print segments
phase_by_segment_priority(problem, segments)
else:
if params.debug:
print 'Alignment unsuccessful'
def __handle_sib_founder_family(self, request, family):
'''A) Align the POO phases of all phased quasi-founder siblings.
B) If alignment was successful, phase unphased sibs in the same family using IBD segments
between them and the phased siblings.
Output long IBD segments among the siblings. We only output long segments since we know
sibs share long segments with the parents.'''
problem, params = request.problem, request.params
g, h = problem.g, problem.haplotype
genotyped_children = im.gt.genotyped_children(problem, family)
# Find phased children, identify IBD segments, build paternal haplotypes (by coloring
# child haplotypes), align children POO phases
phased_children = np.array([
x for x in genotyped_children
if h.fill_fraction(sample=x) >= params.surrogate_parent_fill_threshold
])
if params.debug:
print 'Phased children', phased_children
if not phased_children.size:
# No phased children
return
elif phased_children.size == 1:
# If there's just one kid, no need to align
poo, separation = np.array([1.0]), 1.0
else:
# At least two children ==> look for segments and try to align their POO phase.
segments_phased = im.ibd_distant_hap.among_samples_segments(
problem, phased_children, request.params)
if params.debug:
print 'Segments among phased children:'
print segments_phased
if segments_phased.length:
pa = im.color.hap_color.hap_colors(list(
it.product(sorted(phased_children), im.constants.ALLELES)),
segments_phased,
max_colors=4)
poo, separation, _, _ = im.color.hap_color.best_hap_alignment_to_colors(
pa)
if params.debug:
print pa
print 'POO phases', poo
print 'Separation measure', separation
print 'Parental haplotype coverage', pa.color_sequence_coverage(
np.arange(4))
else:
separation = 0.0 # no segments, can't align and can't trust kids phased the unphased samples
if np.abs(separation) > params.poo_coloring_measure_threshold:
# Alignment of all kids succeeded
if params.debug:
print 'Alignment successful'
# Align POO phases - flip haplotypes of children with reversed haplotype
# This is a -local- alignment within this family. Still need to globally align all POO phases.
# This is done by the poo module, after phasing and IBD segment calculation+index are done.
flipped_children = phased_children[
poo < -params.poo_coloring_measure_threshold]
h_children = problem.h[:, flipped_children, :]
problem.h[:, flipped_children, PATERNAL] = h_children[:, :, MATERNAL]
problem.h[:, flipped_children, MATERNAL] = h_children[:, :, PATERNAL]
# Phase each unphased sib 'sample' using IBD segments between sample and phase_children
# Note: phased sibs these are all full sibs of the proband, not half sibs. So they are
# m=2 meioses away from the proband.
for sample in [
x for x in genotyped_children if
h.fill_fraction(sample=x, snps=im.gt.where_heterozygous(
g, x)) < params.het_fill_threshold
]:
relatives = RelativeCollection.from_sibs(sample, phased_children)
if relatives.length > 0:
# segments = sum((im.ibd_distant_hap.hap_segments(IbdProblem(problem, (sample, allele), (sib, allele), None, params))
# for sib, allele in it.product(phased_children, ALLELES)), im.segment.SegmentSet([]))
segments = im.idist.ibd_segments_with_relatives(
problem,
sample,
relatives.info['index'],
params,
prob_ibd_calculator=prob_ibd_hmm)
# Since j is POO-aligned and is a sib of i, segments can only be between (i,a),(j,a).
# Replace the dummy i allele in segments by j's allele.
segments = im.segment.SegmentSet(
im.segment.Segment(x.snp, [(
x.samples[0][0], x.samples[1][1]), x.samples[1]],
x.bp,
error_snps=x.error_snps,
confidence=x.confidence,
cm=x.cm) for x in segments)
if params.debug:
print 'Segments after assigning sib alleles to proband haplotypes:'
print segments
phase_by_segment_priority(problem, segments)
else:
if params.debug:
print 'Alignment unsuccessful'
def __pass(self, request, phased_fill, max_path_length, target_fill):
'''A single pass of distant relative phasing : loop over unphased individuals and
comSpare each to a set of its distant relatives on the undirected pedigree graph.
Use HMM to estimate IBD segments.'''
problem, params = request.problem, request.params
# debug = params.debug
debug = True
chunk_size = 2 # Chunk size of relatives to process at a time
chunk_growth_size = 1.5 # 1.2 # Chunk size growth factor
if debug:
print '/' * 80
print 'Distant phasing pass: phased_fill %.2f, max_path_length %d, target_fill %.2f' % \
(phased_fill, max_path_length, target_fill)
print '/' * 80
if self.single_sample is not None:
samples = [self.single_sample]
filled = problem.fill_fraction()
else:
filled = problem.fill_fraction()
samples = filled[np.where(filled[:, 1] < phased_fill)[0], 0].astype(int)
filled = filled[samples, :]
filled = filled[np.argsort(filled[:, 1])]
samples = filled[:, 0].astype(int)
if params.selected_mode:
# Restrict samples to selected samples only. Use a boolean array
# 'selected' to indicate which entries of 'samples' are in the selected set.
selected = np.in1d(params.selected_samples, samples)
samples, filled = samples[selected], filled[selected]
if debug:
print 'Target phasing fraction: %.2f, max_path_length %d' % (target_fill, max_path_length)
print 'samples to be phased (< %.2f phased):' % (phased_fill,)
print filled
for i, sample in enumerate([int(x) for x in samples]):
if debug:
print '#' * 60
print 'Distant phasing sample %d' % (sample,)
print '#' * 60
# Find all relatives and sort them by sc depth, then by desc fill%
relatives = RelativeCollection.in_neighborhood(sample, problem, 1, max_path_length, phased_fill)
if relatives.length:
fill = np.concatenate((problem.fill_fraction(sample=relatives.info['index']), relatives.info['distance'][:, 0][np.newaxis].transpose()), axis=1)
index = np.lexsort((-fill[:, 1], fill[:, 2]))
sorted_relatives = relatives.info['index'][index]
# Define a sequence of increasingly-larger chunks. TODO: wrap in iterator
n = relatives.length
ind = np.concatenate(([0], np.round(chunk_size * chunk_growth_size **
np.arange(np.floor(np.log(n) / np.log(chunk_growth_size)
- np.log(chunk_size)))).astype(int)))
if ind[-1] < n: ind = np.concatenate((ind, [n]))
if debug: print 'Chunk size ind', ind
else:
sorted_relatives = []
ind = []
# Look for increasingly-distant relatives until we get IBD coverage sufficient to phase
# the entire sample's chromosome
fill = problem.fill_fraction_of_sample(sample)
k = 0
for k, chunk in enumerate(map(lambda k: (ind[k], ind[k + 1]), range(len(ind) - 1))):
relatives_chunk = sorted_relatives[chunk[START]:chunk[STOP]]
if debug:
print '-' * 70
print 'Processing sample %d, %d/%d fill %f, chunk #%d' % \
(sample, i + 1, len(samples), fill, k + 1)
print 'Relatives fill'
print problem.fill_fraction(sample=relatives_chunk)
if self.single_sample is not None: print 'h[0,%d] = %s' % (self.single_sample, repr(problem.h[0, self.single_sample, :]))
if relatives.length:
segments = im.idist.ibd_segments_with_relatives(problem, sample, relatives_chunk, params,
prob_ibd_calculator=prob_ibd_hmm)
# Not grouping to disjoint at this point, maybe in the future
if debug:
print 'IBD segments, sample-surrogate parent'
print segments
print 'Phasing sample by segment priority'
phase_by_segment_priority(problem, segments)
# Increment relative set depth
fill = problem.fill_fraction_of_sample(sample)
if fill >= target_fill:
if debug: print 'Reached target at sample %d, %d/%d fill %f' % (sample, k + 1, len(samples), fill)
break
if debug: print 'Done with sample %d, %d/%d fill %f' % (sample, k + 1, len(samples), fill)
# Debugging, process only the first max_samples samples
if i == self.max_samples - 1: break
return False
def __pass(self, request, phased_fill, max_path_length, target_fill):
'''A single pass of distant relative phasing : loop over unphased individuals and
comSpare each to a set of its distant relatives on the undirected pedigree graph.
Use HMM to estimate IBD segments.'''
problem, params = request.problem, request.params
# debug = params.debug
debug = True
chunk_size = 2 # Chunk size of relatives to process at a time
chunk_growth_size = 1.5 # 1.2 # Chunk size growth factor
if debug:
print '/' * 80
print 'Distant phasing pass: phased_fill %.2f, max_path_length %d, target_fill %.2f' % \
(phased_fill, max_path_length, target_fill)
print '/' * 80
if self.single_sample is not None:
samples = [self.single_sample]
filled = problem.fill_fraction()
else:
filled = problem.fill_fraction()
samples = filled[np.where(filled[:, 1] < phased_fill)[0],
0].astype(int)
filled = filled[samples, :]
filled = filled[np.argsort(filled[:, 1])]
samples = filled[:, 0].astype(int)
if params.selected_mode:
# Restrict samples to selected samples only. Use a boolean array
# 'selected' to indicate which entries of 'samples' are in the selected set.
selected = np.in1d(params.selected_samples, samples)
samples, filled = samples[selected], filled[selected]
if debug:
print 'Target phasing fraction: %.2f, max_path_length %d' % (
target_fill, max_path_length)
print 'samples to be phased (< %.2f phased):' % (phased_fill, )
print filled
for i, sample in enumerate([int(x) for x in samples]):
if debug:
print '#' * 60
print 'Distant phasing sample %d' % (sample, )
print '#' * 60
# Find all relatives and sort them by sc depth, then by desc fill%
relatives = RelativeCollection.in_neighborhood(
sample, problem, 1, max_path_length, phased_fill)
if relatives.length:
fill = np.concatenate(
(problem.fill_fraction(sample=relatives.info['index']),
relatives.info['distance'][:, 0][np.newaxis].transpose()),
axis=1)
index = np.lexsort((-fill[:, 1], fill[:, 2]))
sorted_relatives = relatives.info['index'][index]
# Define a sequence of increasingly-larger chunks. TODO: wrap in iterator
n = relatives.length
ind = np.concatenate(
([0],
np.round(chunk_size * chunk_growth_size**np.arange(
np.floor(
np.log(n) / np.log(chunk_growth_size) -
np.log(chunk_size)))).astype(int)))
if ind[-1] < n: ind = np.concatenate((ind, [n]))
if debug: print 'Chunk size ind', ind
else:
sorted_relatives = []
ind = []
# Look for increasingly-distant relatives until we get IBD coverage sufficient to phase
# the entire sample's chromosome
fill = problem.fill_fraction_of_sample(sample)
k = 0
for k, chunk in enumerate(
map(lambda k: (ind[k], ind[k + 1]), range(len(ind) - 1))):
relatives_chunk = sorted_relatives[chunk[START]:chunk[STOP]]
if debug:
print '-' * 70
print 'Processing sample %d, %d/%d fill %f, chunk #%d' % \
(sample, i + 1, len(samples), fill, k + 1)
print 'Relatives fill'
print problem.fill_fraction(sample=relatives_chunk)
if self.single_sample is not None:
print 'h[0,%d] = %s' % (
self.single_sample,
repr(problem.h[0, self.single_sample, :]))
if relatives.length:
segments = im.idist.ibd_segments_with_relatives(
problem,
sample,
relatives_chunk,
params,
prob_ibd_calculator=prob_ibd_hmm)
# Not grouping to disjoint at this point, maybe in the future
if debug:
print 'IBD segments, sample-surrogate parent'
print segments
print 'Phasing sample by segment priority'
phase_by_segment_priority(problem, segments)
# Increment relative set depth
fill = problem.fill_fraction_of_sample(sample)
if fill >= target_fill:
if debug:
print 'Reached target at sample %d, %d/%d fill %f' % (
sample, k + 1, len(samples), fill)
break
if debug:
print 'Done with sample %d, %d/%d fill %f' % (
sample, k + 1, len(samples), fill)
# Debugging, process only the first max_samples samples
if i == self.max_samples - 1: break
return False