private GVstatus getGVstatus(final VariantContext vc) {
return (!vc.hasAttribute("GV"))
? GVstatus.NONE
: (vc.getAttribute("GV").equals("T") ? GVstatus.T : GVstatus.F);
}
@Override
public void accumulate(final VariantContext ctx) {
logger.record(ctx.getContig(), ctx.getStart());
final String variantChrom = ctx.getContig();
final int variantPos = ctx.getStart();
// Skip anything a little too funky
if (ctx.isFiltered()) return;
if (!ctx.isVariant()) return;
if (SKIP_CHROMS.contains(variantChrom)) return;
for (final MendelianViolationMetrics trio : trios) {
final Genotype momGt = ctx.getGenotype(trio.MOTHER);
final Genotype dadGt = ctx.getGenotype(trio.FATHER);
final Genotype kidGt = ctx.getGenotype(trio.OFFSPRING);
// if any genotype:
// - has a non-snp allele; or
// - lacks a reference allele
//
// then ignore this trio
if (CollectionUtil.makeList(momGt, dadGt, kidGt)
.stream()
.anyMatch(
gt ->
gt.isHetNonRef()
|| Stream.concat(Stream.of(ctx.getReference()), gt.getAlleles().stream())
.anyMatch(a -> a.length() != 1 || a.isSymbolic()))) {
continue;
}
// if between the trio there are more than 2 alleles including the reference, continue
if (Stream.concat(
Collections.singleton(ctx.getReference()).stream(),
CollectionUtil.makeList(momGt, dadGt, kidGt)
.stream()
.flatMap(gt -> gt.getAlleles().stream()))
.collect(Collectors.toSet())
.size()
> 2) continue;
// Test to make sure:
// 1) That the site is in fact variant in the trio
// 2) that the offspring doesn't have a really wacky het allele balance
if (!isVariant(momGt, dadGt, kidGt)) continue;
if (kidGt.isHet()) {
final int[] ad = kidGt.getAD();
if (ad == null) continue;
final List<Integer> adOfAlleles =
kidGt
.getAlleles()
.stream()
.map(a -> ad[ctx.getAlleleIndex(a)])
.collect(Collectors.toList());
final double minAlleleFraction =
Math.min(adOfAlleles.get(0), adOfAlleles.get(1))
/ (double) (adOfAlleles.get(0) + adOfAlleles.get(1));
if (minAlleleFraction < MIN_HET_FRACTION) continue;
}
///////////////////////////////////////////////////////////////
// Determine whether the offspring should be haploid at this
// locus and which is the parental donor of the haploid genotype
///////////////////////////////////////////////////////////////
boolean haploid = false;
Genotype haploidParentalGenotype = null;
if (FEMALE_CHROMS.contains(variantChrom) && trio.OFFSPRING_SEX != Sex.Unknown) {
if (trio.OFFSPRING_SEX == Sex.Female) {
// famale
haploid = false;
} else if (isInPseudoAutosomalRegion(variantChrom, variantPos)) {
// male but in PAR on X, so diploid
haploid = false;
} else {
// male, out of PAR on X, haploid
haploid = true;
haploidParentalGenotype = momGt;
}
}
// the PAR on the male chromosome should be masked so that reads
// align to the female chromosomes instead, so there's no point
// of worrying about that here.
if (MALE_CHROMS.contains(variantChrom)) {
if (trio.OFFSPRING_SEX == Sex.Male) {
haploid = true;
haploidParentalGenotype = dadGt;
} else {
continue;
}
}
// We only want to look at sites where we have high enough confidence that the genotypes we
// are looking at are
// interesting. We want to ensure that parents are always GQ>=MIN_GQ, and that the kid is
// either GQ>=MIN_GQ or in the
// case where kid is het that the phred-scaled-likelihood of being reference is >=MIN_GQ.
if (haploid
&& (haploidParentalGenotype.isNoCall() || haploidParentalGenotype.getGQ() < MIN_GQ))
continue;
if (!haploid
&& (momGt.isNoCall()
|| momGt.getGQ() < MIN_GQ
|| dadGt.isNoCall()
|| dadGt.getGQ() < MIN_GQ)) continue;
if (kidGt.isNoCall()) continue;
if (momGt.isHomRef() && dadGt.isHomRef() && !kidGt.isHomRef()) {
if (kidGt.getPL()[0] < MIN_GQ) continue;
} else if (kidGt.getGQ() < MIN_GQ) continue;
// Also filter on the DP for each of the samples - it's possible to miss hets when DP is too
// low
if (haploid && (kidGt.getDP() < MIN_DP || haploidParentalGenotype.getDP() < MIN_DP)) continue;
if (!haploid && (kidGt.getDP() < MIN_DP || momGt.getDP() < MIN_DP || dadGt.getDP() < MIN_DP))
continue;
trio.NUM_VARIANT_SITES++;
///////////////////////////////////////////////////////////////
// First test for haploid violations
///////////////////////////////////////////////////////////////
MendelianViolation type = null;
if (haploid) {
if (kidGt.isHet()) continue; // Should not see heterozygous calls at haploid regions
if (!haploidParentalGenotype.getAlleles().contains(kidGt.getAllele(0))) {
if (kidGt.isHomRef()) {
type = MendelianViolation.Haploid_Other;
trio.NUM_HAPLOID_OTHER++;
} else {
type = MendelianViolation.Haploid_Denovo;
trio.NUM_HAPLOID_DENOVO++;
}
}
}
///////////////////////////////////////////////////////////////
// Then test for diploid mendelian violations
///////////////////////////////////////////////////////////////
else if (isMendelianViolation(momGt, dadGt, kidGt)) {
if (momGt.isHomRef() && dadGt.isHomRef() && !kidGt.isHomRef()) {
trio.NUM_DIPLOID_DENOVO++;
type = MendelianViolation.Diploid_Denovo;
} else if (momGt.isHomVar() && dadGt.isHomVar() && kidGt.isHet()) {
trio.NUM_HOMVAR_HOMVAR_HET++;
type = MendelianViolation.HomVar_HomVar_Het;
} else if (kidGt.isHom()
&& ((momGt.isHomRef() && dadGt.isHomVar()) || (momGt.isHomVar() && dadGt.isHomRef()))) {
trio.NUM_HOMREF_HOMVAR_HOM++;
type = MendelianViolation.HomRef_HomVar_Hom;
} else if (kidGt.isHom()
&& ((momGt.isHom() && dadGt.isHet()) || (momGt.isHet() && dadGt.isHom()))) {
trio.NUM_HOM_HET_HOM++;
type = MendelianViolation.Hom_Het_Hom;
} else {
trio.NUM_OTHER++;
type = MendelianViolation.Other;
}
}
// Output a record into the family's violation VCF
if (type != null) {
// Create a new Context subsetted to the three samples
final VariantContextBuilder builder = new VariantContextBuilder(ctx);
builder.genotypes(
ctx.getGenotypes()
.subsetToSamples(CollectionUtil.makeSet(trio.MOTHER, trio.FATHER, trio.OFFSPRING)));
builder.attribute(MENDELIAN_VIOLATION_KEY, type.name());
// Copy over some useful attributes from the full context
if (ctx.hasAttribute(VCFConstants.ALLELE_COUNT_KEY))
builder.attribute(ORIGINAL_AC, ctx.getAttribute(VCFConstants.ALLELE_COUNT_KEY));
if (ctx.hasAttribute(VCFConstants.ALLELE_FREQUENCY_KEY))
builder.attribute(ORIGINAL_AF, ctx.getAttribute(VCFConstants.ALLELE_FREQUENCY_KEY));
if (ctx.hasAttribute(VCFConstants.ALLELE_NUMBER_KEY))
builder.attribute(ORIGINAL_AN, ctx.getAttribute(VCFConstants.ALLELE_NUMBER_KEY));
// Write out the variant record
familyToViolations.get(trio.FAMILY_ID).add(builder.make());
}
}
}
public CountedData map(
RefMetaDataTracker tracker, ReferenceContext ref, AlignmentContext context) {
final CountedData counter = new CountedData();
// For some reason RodWalkers get map calls with null trackers
if (tracker == null) return counter;
VariantContext vcComp = tracker.getFirstValue(alleles);
if (vcComp == null) return counter;
// todo - not sure I want this, may be misleading to filter extended indel events.
if (isInsideExtendedIndel(vcComp, ref)) return counter;
// Do not operate on variants that are not covered to the optional minimum depth
if (!context.hasReads()
|| (minDepth > 0 && context.getBasePileup().getBases().length < minDepth)) {
counter.nUncovered = 1L;
final GVstatus status = getGVstatus(vcComp);
if (status == GVstatus.T) counter.nAltNotCalled = 1L;
else if (status == GVstatus.F) counter.nRefNotCalled = 1L;
else counter.nNoStatusNotCalled = 1L;
return counter;
}
VariantCallContext call;
if (vcComp.isSNP()) {
call = snpEngine.calculateLikelihoodsAndGenotypes(tracker, ref, context).get(0);
} else if (vcComp.isIndel()) {
call = indelEngine.calculateLikelihoodsAndGenotypes(tracker, ref, context).get(0);
} else if (bamIsTruth) {
// assume it's a SNP if no variation is present; this is necessary so that we can test
// supposed monomorphic sites against the truth bam
call = snpEngine.calculateLikelihoodsAndGenotypes(tracker, ref, context).get(0);
} else {
logger.info(
"Not SNP or INDEL "
+ vcComp.getChr()
+ ":"
+ vcComp.getStart()
+ " "
+ vcComp.getAlleles());
return counter;
}
boolean writeVariant = true;
if (bamIsTruth) {
if (call.confidentlyCalled) {
// If truth is a confident REF call
if (call.isVariant()) {
if (vcComp.isVariant()) counter.nAltCalledAlt = 1L;
else {
counter.nAltCalledRef = 1L;
if (printInterestingSites)
System.out.println("Truth=ALT Call=REF at " + call.getChr() + ":" + call.getStart());
}
}
// If truth is a confident ALT call
else {
if (vcComp.isVariant()) {
counter.nRefCalledAlt = 1L;
if (printInterestingSites)
System.out.println("Truth=REF Call=ALT at " + call.getChr() + ":" + call.getStart());
} else counter.nRefCalledRef = 1L;
}
} else {
counter.nNotConfidentCalls = 1L;
if (printInterestingSites)
System.out.println("Truth is not confident at " + call.getChr() + ":" + call.getStart());
writeVariant = false;
}
} else {
// if (!vcComp.hasExtendedAttribute("GV"))
// throw new UserException.BadInput("Variant has no GV annotation in the INFO
// field. " + vcComp.getChr() + ":" + vcComp.getStart());
final GVstatus status = getGVstatus(vcComp);
if (call.isCalledAlt(callConf)) {
if (status == GVstatus.T) counter.nAltCalledAlt = 1L;
else if (status == GVstatus.F) {
counter.nRefCalledAlt = 1L;
if (printInterestingSites)
System.out.println("Truth=REF Call=ALT at " + call.getChr() + ":" + call.getStart());
} else counter.nNoStatusCalledAlt = 1L;
} else if (call.isCalledRef(callConf)) {
if (status == GVstatus.T) {
counter.nAltCalledRef = 1L;
if (printInterestingSites)
System.out.println("Truth=ALT Call=REF at " + call.getChr() + ":" + call.getStart());
} else if (status == GVstatus.F) counter.nRefCalledRef = 1L;
else counter.nNoStatusCalledRef = 1L;
} else {
counter.nNotConfidentCalls = 1L;
if (status == GVstatus.T) counter.nAltNotCalled = 1L;
else if (status == GVstatus.F) counter.nRefNotCalled = 1L;
else counter.nNoStatusNotCalled = 1L;
if (printInterestingSites)
System.out.println("Truth is not confident at " + call.getChr() + ":" + call.getStart());
writeVariant = false;
}
}
if (vcfWriter != null && writeVariant) {
if (!vcComp.hasAttribute(GATKVCFConstants.GENOTYPE_AND_VALIDATE_STATUS_KEY)) {
vcfWriter.add(
new VariantContextBuilder(vcComp)
.attribute(
GATKVCFConstants.GENOTYPE_AND_VALIDATE_STATUS_KEY,
call.isCalledAlt(callConf) ? "ALT" : "REF")
.make());
} else vcfWriter.add(vcComp);
}
return counter;
}