def _extract(self, chromvariant, locset, num_type_dict):
clone = ChromVariants(chromvariant.chrom, chromvariant._max_indel_len)
for loc in locset:
var = chromvariant.all_variants[loc]
clone._add_variant(var)
num_type_dict[var.var_type] += 1
clone._ensure_sorted()
return clone
def _extract(self,chromvariant,locset,num_type_dict):
clone = ChromVariants(chromvariant.chrom, chromvariant._max_indel_len)
for loc in locset:
var = chromvariant.all_variants[loc]
clone._add_variant(var)
num_type_dict[var.var_type] += 1
clone._ensure_sorted()
return clone
class ChromVariantStats:
"""Stats for a certain contig's worth of variants."""
def __init__(self, true_var, pred_var, true_positives, false_positives,
false_negatives, concordance):
self.chrom = true_var.chrom
self.true_var = true_var
self.pred_var = pred_var
self.num_true = dict(
map(lambda t: [t, true_var.var_num(t)], VARIANT_TYPE))
self.num_pred = dict(
map(lambda t: [t, pred_var.var_num(t)], VARIANT_TYPE))
self.num_tp = _type_dict() # true positives as int
self.num_fp = _type_dict()
self.num_fn = _type_dict()
self.false_positives = self._extract(pred_var, false_positives,
self.num_fp) # chromvariants
self.false_negatives = self._extract(true_var, false_negatives,
self.num_fn) # chromvariants
self.true_positives = self._extract(true_var, true_positives,
self.num_tp) # chromvariants
self.rescued_vars = ChromVariants(
self.chrom,
self.false_positives._max_indel_len) # populate with rescued vars
self.intersect_bad = None # set externally
self.known_fp = None # set externally
self.calls_at_known_fp = None # set externally
self.known_fp_variants = None # set externally
self.genotype_concordance = concordance
# for loc in true_positives:
# var = true_var.all_variants[loc]
# self.num_tp[var.var_type] += 1
def _extract(self, chromvariant, locset, num_type_dict):
clone = ChromVariants(chromvariant.chrom, chromvariant._max_indel_len)
for loc in locset:
var = chromvariant.all_variants[loc]
clone._add_variant(var)
num_type_dict[var.var_type] += 1
clone._ensure_sorted()
return clone
def rectify(self, ref, window):
"""Rescue variants from VCF ambiguity.
Given reference genome and window of sequence comparison,
fix each error estimated to be an artifact of VCF ambiguity.
Calling this function strictly improves realism of evaluation,
but doesn't perfectly detect ambiguity.
TODO: Don't do redundant calculation for overlapping windows
TODO: This should return two bools: events match, and genotypes match
Note: Here the window is a single integer for the size. Here it gets
converted specifically to an interval, and that's its subsequent usage in the stack.
"""
locs_to_rescue = list(
map(lambda loc: loc, self.false_negatives.all_locations))
# note we needed to force a copy here, since rescue_mission is modifying the false-negative sets
for loc in locs_to_rescue:
if (loc in self.false_negatives.all_variants
): # if the element is still in the set of false negatives
new_tp, rm_fp, rescued_vars = rescue_mission(
self.false_negatives, self.false_positives,
self.true_positives, loc, ref, window)
for t in VARIANT_TYPE:
# seemingly odd accounting. The number of predicted variants *changes* as a result of rescuing.
# e.g. 2 predicted FPs are in fact 1 FN. So
# -- remove 2 predicted variants
# -- remove 2 false positives
# -- remove 1 false negative
# -- add 1 true positive
self.num_pred[t] -= rm_fp[t]
self.num_fp[t] -= rm_fp[t]
self.num_pred[t] += new_tp[t]
self.num_fn[t] -= new_tp[t]
self.num_tp[t] += new_tp[t]
for v in rescued_vars:
self.rescued_vars._add_variant(v)
def _nrd_counts(self, var_type):
genoGenoCounts = self.genotype_concordance[var_type]
nWrong = genoGenoCounts[GENOTYPE_TYPE.HOM_REF][GENOTYPE_TYPE.HET]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_REF][GENOTYPE_TYPE.HOM_VAR]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HET][GENOTYPE_TYPE.HOM_REF]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HET][GENOTYPE_TYPE.HOM_VAR]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][GENOTYPE_TYPE.HOM_REF]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][GENOTYPE_TYPE.HET]
nTotal = nWrong + genoGenoCounts[GENOTYPE_TYPE.HET][
GENOTYPE_TYPE.HET] + genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][
GENOTYPE_TYPE.HOM_VAR]
return (nWrong, nTotal)
def to_dict(self, var_type):
stats = {}
stats['num_true'] = self.num_true[var_type]
stats['num_pred'] = self.num_pred[var_type]
stats['false_positives'] = self.num_fp[var_type]
stats['false_negatives'] = self.num_fn[var_type]
stats['good_predictions'] = self.num_tp[var_type]
stats['intersect_bad'] = len(self.intersect_bad[var_type])
nrd_wrong, nrd_total = self._nrd_counts(var_type)
stats['nrd_wrong'] = nrd_wrong
stats['nrd_total'] = nrd_total
stats['known_fp_calls'] = self.calls_at_known_fp[
var_type] if self.calls_at_known_fp else 0
stats['known_fp'] = self.known_fp[
var_type] if self.calls_at_known_fp else 0
return stats
class ChromVariantStats:
"""Stats for a certain contig's worth of variants."""
def __init__(self, true_var, pred_var, true_positives,
false_positives, false_negatives,concordance):
self.chrom = true_var.chrom
self.true_var = true_var
self.pred_var = pred_var
self.num_true = dict(map(lambda t: [t,true_var.var_num(t)],VARIANT_TYPE))
self.num_pred = dict(map(lambda t: [t,pred_var.var_num(t)],VARIANT_TYPE))
self.num_tp = _type_dict() # true positives as int
self.num_fp = _type_dict()
self.num_fn = _type_dict()
self.false_positives = self._extract(pred_var,false_positives,self.num_fp) # chromvariants
self.false_negatives = self._extract(true_var,false_negatives,self.num_fn) # chromvariants
self.true_positives = self._extract(true_var,true_positives,self.num_tp) # chromvariants
self.rescued_vars = ChromVariants(self.chrom,self.false_positives._max_indel_len) # populate with rescued vars
self.intersect_bad = None # set externally
self.known_fp = None # set externally
self.calls_at_known_fp = None # set externally
self.known_fp_variants = None # set externally
self.genotype_concordance = concordance
# for loc in true_positives:
# var = true_var.all_variants[loc]
# self.num_tp[var.var_type] += 1
def _extract(self,chromvariant,locset,num_type_dict):
clone = ChromVariants(chromvariant.chrom, chromvariant._max_indel_len)
for loc in locset:
var = chromvariant.all_variants[loc]
clone._add_variant(var)
num_type_dict[var.var_type] += 1
clone._ensure_sorted()
return clone
def rectify(self, ref, window):
"""Rescue variants from VCF ambiguity.
Given reference genome and window of sequence comparison,
fix each error estimated to be an artifact of VCF ambiguity.
Calling this function strictly improves realism of evaluation,
but doesn't perfectly detect ambiguity.
TODO: Don't do redundant calculation for overlapping windows
TODO: This should return two bools: events match, and genotypes match
Note: Here the window is a single integer for the size. Here it gets
converted specifically to an interval, and that's its subsequent usage in the stack.
"""
locs_to_rescue = list(map(lambda loc: loc, self.false_negatives.all_locations))
# note we needed to force a copy here, since rescue_mission is modifying the false-negative sets
for loc in locs_to_rescue:
if ( loc in self.false_negatives.all_variants ): # if the element is still in the set of false negatives
new_tp,rm_fp,rescued_vars = rescue_mission(self.false_negatives,self.false_positives,self.true_positives,loc,ref,window)
for t in VARIANT_TYPE:
# seemingly odd accounting. The number of predicted variants *changes* as a result of rescuing.
# e.g. 2 predicted FPs are in fact 1 FN. So
# -- remove 2 predicted variants
# -- remove 2 false positives
# -- remove 1 false negative
# -- add 1 true positive
self.num_pred[t] -= rm_fp[t]
self.num_fp[t] -= rm_fp[t]
self.num_pred[t] += new_tp[t]
self.num_fn[t] -= new_tp[t]
self.num_tp[t] += new_tp[t]
for v in rescued_vars:
self.rescued_vars._add_variant(v)
def _nrd_counts(self,var_type):
genoGenoCounts = self.genotype_concordance[var_type]
nWrong = genoGenoCounts[GENOTYPE_TYPE.HOM_REF][GENOTYPE_TYPE.HET]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_REF][GENOTYPE_TYPE.HOM_VAR]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HET][GENOTYPE_TYPE.HOM_REF]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HET][GENOTYPE_TYPE.HOM_VAR]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][GENOTYPE_TYPE.HOM_REF]
nWrong += genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][GENOTYPE_TYPE.HET]
nTotal = nWrong + genoGenoCounts[GENOTYPE_TYPE.HET][GENOTYPE_TYPE.HET] + genoGenoCounts[GENOTYPE_TYPE.HOM_VAR][GENOTYPE_TYPE.HOM_VAR]
return (nWrong,nTotal)
def to_dict(self,var_type):
stats = {}
stats['num_true'] = self.num_true[var_type]
stats['num_pred'] = self.num_pred[var_type]
stats['false_positives'] = self.num_fp[var_type]
stats['false_negatives'] = self.num_fn[var_type]
stats['good_predictions'] = self.num_tp[var_type]
stats['intersect_bad'] = len(self.intersect_bad[var_type])
nrd_wrong,nrd_total = self._nrd_counts(var_type)
stats['nrd_wrong'] = nrd_wrong
stats['nrd_total'] = nrd_total
stats['known_fp_calls'] = self.calls_at_known_fp[var_type] if self.calls_at_known_fp else 0
stats['known_fp'] = self.known_fp[var_type] if self.calls_at_known_fp else 0
return stats