/**
* Takes the interval, finds it in the stash, prints it to the VCF
*
* @param stats The statistics of the interval
* @param refAllele the reference allele
*/
private void outputStatsToVCF(final IntervalStratification stats, final Allele refAllele) {
GenomeLoc interval = stats.getInterval();
final List<Allele> alleles = new ArrayList<>();
final Map<String, Object> attributes = new HashMap<>();
final ArrayList<Genotype> genotypes = new ArrayList<>();
for (String sample : samples) {
final GenotypeBuilder gb = new GenotypeBuilder(sample);
SampleStratification sampleStat = stats.getSampleStatistics(sample);
gb.attribute(
GATKVCFConstants.AVG_INTERVAL_DP_BY_SAMPLE_KEY,
sampleStat.averageCoverage(interval.size()));
gb.attribute(GATKVCFConstants.LOW_COVERAGE_LOCI, sampleStat.getNLowCoveredLoci());
gb.attribute(GATKVCFConstants.ZERO_COVERAGE_LOCI, sampleStat.getNUncoveredLoci());
gb.filters(statusToStrings(stats.getSampleStatistics(sample).callableStatuses(), false));
genotypes.add(gb.make());
}
alleles.add(refAllele);
alleles.add(SYMBOLIC_ALLELE);
VariantContextBuilder vcb =
new VariantContextBuilder(
"DiagnoseTargets",
interval.getContig(),
interval.getStart(),
interval.getStop(),
alleles);
vcb = vcb.log10PError(VariantContext.NO_LOG10_PERROR);
vcb.filters(new LinkedHashSet<>(statusToStrings(stats.callableStatuses(), true)));
attributes.put(VCFConstants.END_KEY, interval.getStop());
attributes.put(GATKVCFConstants.AVG_INTERVAL_DP_KEY, stats.averageCoverage(interval.size()));
attributes.put(GATKVCFConstants.INTERVAL_GC_CONTENT_KEY, stats.gcContent());
vcb = vcb.attributes(attributes);
vcb = vcb.genotypes(genotypes);
vcfWriter.add(vcb.make());
}
/**
* Merges VariantContexts into a single hybrid. Takes genotypes for common samples in priority
* order, if provided. If uniqifySamples is true, the priority order is ignored and names are
* created by concatenating the VC name with the sample name
*
* @param genomeLocParser loc parser
* @param unsortedVCs collection of unsorted VCs
* @param priorityListOfVCs priority list detailing the order in which we should grab the VCs
* @param filteredRecordMergeType merge type for filtered records
* @param genotypeMergeOptions merge option for genotypes
* @param annotateOrigin should we annotate the set it came from?
* @param printMessages should we print messages?
* @param setKey the key name of the set
* @param filteredAreUncalled are filtered records uncalled?
* @param mergeInfoWithMaxAC should we merge in info from the VC with maximum allele count?
* @return new VariantContext representing the merge of unsortedVCs
*/
public static VariantContext simpleMerge(
final GenomeLocParser genomeLocParser,
final Collection<VariantContext> unsortedVCs,
final List<String> priorityListOfVCs,
final FilteredRecordMergeType filteredRecordMergeType,
final GenotypeMergeType genotypeMergeOptions,
final boolean annotateOrigin,
final boolean printMessages,
final String setKey,
final boolean filteredAreUncalled,
final boolean mergeInfoWithMaxAC) {
if (unsortedVCs == null || unsortedVCs.size() == 0) return null;
if (annotateOrigin && priorityListOfVCs == null)
throw new IllegalArgumentException(
"Cannot merge calls and annotate their origins without a complete priority list of VariantContexts");
if (genotypeMergeOptions == GenotypeMergeType.REQUIRE_UNIQUE)
verifyUniqueSampleNames(unsortedVCs);
List<VariantContext> prepaddedVCs =
sortVariantContextsByPriority(unsortedVCs, priorityListOfVCs, genotypeMergeOptions);
// Make sure all variant contexts are padded with reference base in case of indels if necessary
List<VariantContext> VCs = new ArrayList<VariantContext>();
for (VariantContext vc : prepaddedVCs) {
// also a reasonable place to remove filtered calls, if needed
if (!filteredAreUncalled || vc.isNotFiltered())
VCs.add(createVariantContextWithPaddedAlleles(vc, false));
}
if (VCs.size() == 0) // everything is filtered out and we're filteredAreUncalled
return null;
// establish the baseline info from the first VC
final VariantContext first = VCs.get(0);
final String name = first.getSource();
final Allele refAllele = determineReferenceAllele(VCs);
final Set<Allele> alleles = new LinkedHashSet<Allele>();
final Set<String> filters = new TreeSet<String>();
final Map<String, Object> attributes = new TreeMap<String, Object>();
final Set<String> inconsistentAttributes = new HashSet<String>();
final Set<String> variantSources =
new HashSet<
String>(); // contains the set of sources we found in our set of VCs that are variant
final Set<String> rsIDs = new LinkedHashSet<String>(1); // most of the time there's one id
GenomeLoc loc = getLocation(genomeLocParser, first);
int depth = 0;
int maxAC = -1;
final Map<String, Object> attributesWithMaxAC = new TreeMap<String, Object>();
double log10PError = 1;
VariantContext vcWithMaxAC = null;
GenotypesContext genotypes = GenotypesContext.create();
// counting the number of filtered and variant VCs
int nFiltered = 0;
boolean remapped = false;
// cycle through and add info from the other VCs, making sure the loc/reference matches
for (VariantContext vc : VCs) {
if (loc.getStart() != vc.getStart()) // || !first.getReference().equals(vc.getReference()) )
throw new ReviewedStingException(
"BUG: attempting to merge VariantContexts with different start sites: first="
+ first.toString()
+ " second="
+ vc.toString());
if (getLocation(genomeLocParser, vc).size() > loc.size())
loc = getLocation(genomeLocParser, vc); // get the longest location
nFiltered += vc.isFiltered() ? 1 : 0;
if (vc.isVariant()) variantSources.add(vc.getSource());
AlleleMapper alleleMapping = resolveIncompatibleAlleles(refAllele, vc, alleles);
remapped = remapped || alleleMapping.needsRemapping();
alleles.addAll(alleleMapping.values());
mergeGenotypes(
genotypes, vc, alleleMapping, genotypeMergeOptions == GenotypeMergeType.UNIQUIFY);
log10PError = Math.min(log10PError, vc.isVariant() ? vc.getLog10PError() : 1);
filters.addAll(vc.getFilters());
//
// add attributes
//
// special case DP (add it up) and ID (just preserve it)
//
if (vc.hasAttribute(VCFConstants.DEPTH_KEY))
depth += vc.getAttributeAsInt(VCFConstants.DEPTH_KEY, 0);
if (vc.hasID()) rsIDs.add(vc.getID());
if (mergeInfoWithMaxAC && vc.hasAttribute(VCFConstants.ALLELE_COUNT_KEY)) {
String rawAlleleCounts = vc.getAttributeAsString(VCFConstants.ALLELE_COUNT_KEY, null);
// lets see if the string contains a , separator
if (rawAlleleCounts.contains(VCFConstants.INFO_FIELD_ARRAY_SEPARATOR)) {
List<String> alleleCountArray =
Arrays.asList(
rawAlleleCounts
.substring(1, rawAlleleCounts.length() - 1)
.split(VCFConstants.INFO_FIELD_ARRAY_SEPARATOR));
for (String alleleCount : alleleCountArray) {
final int ac = Integer.valueOf(alleleCount.trim());
if (ac > maxAC) {
maxAC = ac;
vcWithMaxAC = vc;
}
}
} else {
final int ac = Integer.valueOf(rawAlleleCounts);
if (ac > maxAC) {
maxAC = ac;
vcWithMaxAC = vc;
}
}
}
for (Map.Entry<String, Object> p : vc.getAttributes().entrySet()) {
String key = p.getKey();
// if we don't like the key already, don't go anywhere
if (!inconsistentAttributes.contains(key)) {
boolean alreadyFound = attributes.containsKey(key);
Object boundValue = attributes.get(key);
boolean boundIsMissingValue =
alreadyFound && boundValue.equals(VCFConstants.MISSING_VALUE_v4);
if (alreadyFound && !boundValue.equals(p.getValue()) && !boundIsMissingValue) {
// we found the value but we're inconsistent, put it in the exclude list
// System.out.printf("Inconsistent INFO values: %s => %s and %s%n", key, boundValue,
// p.getValue());
inconsistentAttributes.add(key);
attributes.remove(key);
} else if (!alreadyFound || boundIsMissingValue) { // no value
// if ( vc != first ) System.out.printf("Adding key %s => %s%n", p.getKey(),
// p.getValue());
attributes.put(key, p.getValue());
}
}
}
}
// if we have more alternate alleles in the merged VC than in one or more of the
// original VCs, we need to strip out the GL/PLs (because they are no longer accurate), as well
// as allele-dependent attributes like AC,AF
for (VariantContext vc : VCs) {
if (vc.alleles.size() == 1) continue;
if (hasPLIncompatibleAlleles(alleles, vc.alleles)) {
if (!genotypes.isEmpty())
logger.warn(
String.format(
"Stripping PLs at %s due incompatible alleles merged=%s vs. single=%s",
genomeLocParser.createGenomeLoc(vc), alleles, vc.alleles));
genotypes = stripPLs(genotypes);
// this will remove stale AC,AF attributed from vc
calculateChromosomeCounts(vc, attributes, true);
break;
}
}
// take the VC with the maxAC and pull the attributes into a modifiable map
if (mergeInfoWithMaxAC && vcWithMaxAC != null) {
attributesWithMaxAC.putAll(vcWithMaxAC.getAttributes());
}
// if at least one record was unfiltered and we want a union, clear all of the filters
if ((filteredRecordMergeType == FilteredRecordMergeType.KEEP_IF_ANY_UNFILTERED
&& nFiltered != VCs.size())
|| filteredRecordMergeType == FilteredRecordMergeType.KEEP_UNCONDITIONAL) filters.clear();
if (annotateOrigin) { // we care about where the call came from
String setValue;
if (nFiltered == 0
&& variantSources.size() == priorityListOfVCs.size()) // nothing was unfiltered
setValue = MERGE_INTERSECTION;
else if (nFiltered == VCs.size()) // everything was filtered out
setValue = MERGE_FILTER_IN_ALL;
else if (variantSources.isEmpty()) // everyone was reference
setValue = MERGE_REF_IN_ALL;
else {
LinkedHashSet<String> s = new LinkedHashSet<String>();
for (VariantContext vc : VCs)
if (vc.isVariant())
s.add(vc.isFiltered() ? MERGE_FILTER_PREFIX + vc.getSource() : vc.getSource());
setValue = Utils.join("-", s);
}
if (setKey != null) {
attributes.put(setKey, setValue);
if (mergeInfoWithMaxAC && vcWithMaxAC != null) {
attributesWithMaxAC.put(setKey, vcWithMaxAC.getSource());
}
}
}
if (depth > 0) attributes.put(VCFConstants.DEPTH_KEY, String.valueOf(depth));
final String ID = rsIDs.isEmpty() ? VCFConstants.EMPTY_ID_FIELD : Utils.join(",", rsIDs);
final VariantContextBuilder builder = new VariantContextBuilder().source(name).id(ID);
builder.loc(loc.getContig(), loc.getStart(), loc.getStop());
builder.alleles(alleles);
builder.genotypes(genotypes);
builder.log10PError(log10PError);
builder.filters(filters).attributes(mergeInfoWithMaxAC ? attributesWithMaxAC : attributes);
// Trim the padded bases of all alleles if necessary
VariantContext merged = createVariantContextWithTrimmedAlleles(builder.make());
if (printMessages && remapped) System.out.printf("Remapped => %s%n", merged);
return merged;
}
/**
* Main entry function to calculate genotypes of a given VC with corresponding GL's
*
* @param tracker Tracker
* @param refContext Reference context
* @param rawContext Raw context
* @param stratifiedContexts Stratified alignment contexts
* @param vc Input VC
* @param model GL calculation model
* @param inheritAttributesFromInputVC Output VC will contain attributes inherited from input vc
* @return VC with assigned genotypes
*/
public VariantCallContext calculateGenotypes(
final RefMetaDataTracker tracker,
final ReferenceContext refContext,
final AlignmentContext rawContext,
Map<String, AlignmentContext> stratifiedContexts,
final VariantContext vc,
final GenotypeLikelihoodsCalculationModel.Model model,
final boolean inheritAttributesFromInputVC,
final Map<String, org.broadinstitute.sting.utils.genotyper.PerReadAlleleLikelihoodMap>
perReadAlleleLikelihoodMap) {
boolean limitedContext =
tracker == null || refContext == null || rawContext == null || stratifiedContexts == null;
// initialize the data for this thread if that hasn't been done yet
if (afcm.get() == null) {
afcm.set(AFCalcFactory.createAFCalc(UAC, N, logger));
}
// estimate our confidence in a reference call and return
if (vc.getNSamples() == 0) {
if (limitedContext) return null;
return (UAC.OutputMode != OUTPUT_MODE.EMIT_ALL_SITES
? estimateReferenceConfidence(vc, stratifiedContexts, getTheta(model), false, 1.0)
: generateEmptyContext(tracker, refContext, stratifiedContexts, rawContext));
}
AFCalcResult AFresult = afcm.get().getLog10PNonRef(vc, getAlleleFrequencyPriors(model));
// is the most likely frequency conformation AC=0 for all alternate alleles?
boolean bestGuessIsRef = true;
// determine which alternate alleles have AF>0
final List<Allele> myAlleles = new ArrayList<Allele>(vc.getAlleles().size());
final List<Integer> alleleCountsofMLE = new ArrayList<Integer>(vc.getAlleles().size());
myAlleles.add(vc.getReference());
for (int i = 0; i < AFresult.getAllelesUsedInGenotyping().size(); i++) {
final Allele alternateAllele = AFresult.getAllelesUsedInGenotyping().get(i);
if (alternateAllele.isReference()) continue;
// we are non-ref if the probability of being non-ref > the emit confidence.
// the emit confidence is phred-scaled, say 30 => 10^-3.
// the posterior AF > 0 is log10: -5 => 10^-5
// we are non-ref if 10^-5 < 10^-3 => -5 < -3
final boolean isNonRef =
AFresult.isPolymorphic(alternateAllele, UAC.STANDARD_CONFIDENCE_FOR_EMITTING / -10.0);
// if the most likely AC is not 0, then this is a good alternate allele to use
if (isNonRef) {
myAlleles.add(alternateAllele);
alleleCountsofMLE.add(AFresult.getAlleleCountAtMLE(alternateAllele));
bestGuessIsRef = false;
}
// if in GENOTYPE_GIVEN_ALLELES mode, we still want to allow the use of a poor allele
else if (UAC.GenotypingMode
== GenotypeLikelihoodsCalculationModel.GENOTYPING_MODE.GENOTYPE_GIVEN_ALLELES) {
myAlleles.add(alternateAllele);
alleleCountsofMLE.add(AFresult.getAlleleCountAtMLE(alternateAllele));
}
}
final double PoFGT0 = Math.pow(10, AFresult.getLog10PosteriorOfAFGT0());
// note the math.abs is necessary because -10 * 0.0 => -0.0 which isn't nice
final double phredScaledConfidence =
Math.abs(
!bestGuessIsRef
|| UAC.GenotypingMode
== GenotypeLikelihoodsCalculationModel.GENOTYPING_MODE
.GENOTYPE_GIVEN_ALLELES
? -10 * AFresult.getLog10PosteriorOfAFEq0()
: -10 * AFresult.getLog10PosteriorOfAFGT0());
// return a null call if we don't pass the confidence cutoff or the most likely allele frequency
// is zero
if (UAC.OutputMode != OUTPUT_MODE.EMIT_ALL_SITES
&& !passesEmitThreshold(phredScaledConfidence, bestGuessIsRef)) {
// technically, at this point our confidence in a reference call isn't accurately estimated
// because it didn't take into account samples with no data, so let's get a better estimate
return limitedContext
? null
: estimateReferenceConfidence(vc, stratifiedContexts, getTheta(model), true, PoFGT0);
}
// start constructing the resulting VC
final GenomeLoc loc = genomeLocParser.createGenomeLoc(vc);
final VariantContextBuilder builder =
new VariantContextBuilder(
"UG_call", loc.getContig(), loc.getStart(), loc.getStop(), myAlleles);
builder.log10PError(phredScaledConfidence / -10.0);
if (!passesCallThreshold(phredScaledConfidence)) builder.filters(filter);
// create the genotypes
final GenotypesContext genotypes = afcm.get().subsetAlleles(vc, myAlleles, true, ploidy);
builder.genotypes(genotypes);
// print out stats if we have a writer
if (verboseWriter != null && !limitedContext)
printVerboseData(refContext.getLocus().toString(), vc, PoFGT0, phredScaledConfidence, model);
// *** note that calculating strand bias involves overwriting data structures, so we do that
// last
final HashMap<String, Object> attributes = new HashMap<String, Object>();
// inherit attributed from input vc if requested
if (inheritAttributesFromInputVC) attributes.putAll(vc.getAttributes());
// if the site was downsampled, record that fact
if (!limitedContext && rawContext.hasPileupBeenDownsampled())
attributes.put(VCFConstants.DOWNSAMPLED_KEY, true);
if (UAC.ANNOTATE_NUMBER_OF_ALLELES_DISCOVERED)
attributes.put(NUMBER_OF_DISCOVERED_ALLELES_KEY, vc.getAlternateAlleles().size());
// add the MLE AC and AF annotations
if (alleleCountsofMLE.size() > 0) {
attributes.put(VCFConstants.MLE_ALLELE_COUNT_KEY, alleleCountsofMLE);
final int AN = builder.make().getCalledChrCount();
final ArrayList<Double> MLEfrequencies = new ArrayList<Double>(alleleCountsofMLE.size());
// the MLEAC is allowed to be larger than the AN (e.g. in the case of all PLs being 0, the GT
// is ./. but the exact model may arbitrarily choose an AC>1)
for (int AC : alleleCountsofMLE) MLEfrequencies.add(Math.min(1.0, (double) AC / (double) AN));
attributes.put(VCFConstants.MLE_ALLELE_FREQUENCY_KEY, MLEfrequencies);
}
if (UAC.COMPUTE_SLOD && !limitedContext && !bestGuessIsRef) {
// final boolean DEBUG_SLOD = false;
// the overall lod
// double overallLog10PofNull = AFresult.log10AlleleFrequencyPosteriors[0];
double overallLog10PofF = AFresult.getLog10LikelihoodOfAFGT0();
// if ( DEBUG_SLOD ) System.out.println("overallLog10PofF=" + overallLog10PofF);
List<Allele> allAllelesToUse = builder.make().getAlleles();
// the forward lod
VariantContext vcForward =
calculateLikelihoods(
tracker,
refContext,
stratifiedContexts,
AlignmentContextUtils.ReadOrientation.FORWARD,
allAllelesToUse,
false,
model,
perReadAlleleLikelihoodMap);
AFresult = afcm.get().getLog10PNonRef(vcForward, getAlleleFrequencyPriors(model));
// double[] normalizedLog10Posteriors =
// MathUtils.normalizeFromLog10(AFresult.log10AlleleFrequencyPosteriors, true);
double forwardLog10PofNull = AFresult.getLog10LikelihoodOfAFEq0();
double forwardLog10PofF = AFresult.getLog10LikelihoodOfAFGT0();
// if ( DEBUG_SLOD ) System.out.println("forwardLog10PofNull=" + forwardLog10PofNull + ",
// forwardLog10PofF=" + forwardLog10PofF);
// the reverse lod
VariantContext vcReverse =
calculateLikelihoods(
tracker,
refContext,
stratifiedContexts,
AlignmentContextUtils.ReadOrientation.REVERSE,
allAllelesToUse,
false,
model,
perReadAlleleLikelihoodMap);
AFresult = afcm.get().getLog10PNonRef(vcReverse, getAlleleFrequencyPriors(model));
// normalizedLog10Posteriors =
// MathUtils.normalizeFromLog10(AFresult.log10AlleleFrequencyPosteriors, true);
double reverseLog10PofNull = AFresult.getLog10LikelihoodOfAFEq0();
double reverseLog10PofF = AFresult.getLog10LikelihoodOfAFGT0();
// if ( DEBUG_SLOD ) System.out.println("reverseLog10PofNull=" + reverseLog10PofNull + ",
// reverseLog10PofF=" + reverseLog10PofF);
double forwardLod = forwardLog10PofF + reverseLog10PofNull - overallLog10PofF;
double reverseLod = reverseLog10PofF + forwardLog10PofNull - overallLog10PofF;
// if ( DEBUG_SLOD ) System.out.println("forward lod=" + forwardLod + ", reverse lod=" +
// reverseLod);
// strand score is max bias between forward and reverse strands
double strandScore = Math.max(forwardLod, reverseLod);
// rescale by a factor of 10
strandScore *= 10.0;
// logger.debug(String.format("SLOD=%f", strandScore));
if (!Double.isNaN(strandScore)) attributes.put("SB", strandScore);
}
// finish constructing the resulting VC
builder.attributes(attributes);
VariantContext vcCall = builder.make();
// if we are subsetting alleles (either because there were too many or because some were not
// polymorphic)
// then we may need to trim the alleles (because the original VariantContext may have had to pad
// at the end).
if (myAlleles.size() != vc.getAlleles().size()
&& !limitedContext) // limitedContext callers need to handle allele trimming on their own to
// keep their perReadAlleleLikelihoodMap alleles in sync
vcCall = VariantContextUtils.reverseTrimAlleles(vcCall);
if (annotationEngine != null
&& !limitedContext) { // limitedContext callers need to handle annotations on their own by
// calling their own annotationEngine
// Note: we want to use the *unfiltered* and *unBAQed* context for the annotations
final ReadBackedPileup pileup = rawContext.getBasePileup();
stratifiedContexts = AlignmentContextUtils.splitContextBySampleName(pileup);
vcCall =
annotationEngine.annotateContext(
tracker, refContext, stratifiedContexts, vcCall, perReadAlleleLikelihoodMap);
}
return new VariantCallContext(vcCall, confidentlyCalled(phredScaledConfidence, PoFGT0));
}