private bool IsForcedAllele(CalledAllele baseCalledAllele)
{
if (ForcedGtAlleles == null)
{
return(false);
}
var allele = new Tuple <string, int, string, string>(baseCalledAllele.Chromosome, baseCalledAllele.ReferencePosition, baseCalledAllele.ReferenceAllele, baseCalledAllele.AlternateAllele);
return(ForcedGtAlleles.Contains(allele));
}
//this is not as obvious as it seems. What is the depth of a het-alt1-alt2, when you have two insertions of different length?
//Least controversial is to take the maximum.
public int GetDepthCountInt(IEnumerable <CalledAllele> variants)
{
CalledAllele firstVariant = variants.First();
int totalDepth = 0;
foreach (var variant in variants)
{
totalDepth = Math.Max(variant.TotalCoverage, totalDepth);
}
return(totalDepth);
}
private void CallCandidate(CalledAllele mnv, bool isReference)
{
mnv.NoiseLevelApplied = _bamParams.MinimumBaseCallQuality;
//tjd
//since any input variant passed filters, we assume it is not SB'ed.
//Before the big Scylla refactor, we used to copy the SB number from the original variant at this index. but thats not really consistent with the new call.
//gb
//yes, we can assume that the component variants were not strand biased according to the thresholds used in the original Pisces run, but that doesn't mean that they're necessarily -100
//tjd:
//Correct. Alternatly, here are some other options:
// A) We could leave the SB tag off the final var call, or set it to Nan,
// but that could be a downstream parsing problem.
// B) We could make some number up based the SB of the input variants
// (some max value to represent the worst case of the component variants in the nbdh...?)
// C) Implement a fully stranded phasing model.
// IMO, if we put work into this, we should do C. This will probably be a feature request before too long anyway.
//For the moment, nothing affects actually pass/Fail varcall decisions.
mnv.StrandBiasResults.GATKBiasScore = -100;
if (mnv.VariantQscore < _callerParams.MinimumVariantQScore)
{
if (isReference)
{
mnv.Genotype = Pisces.Domain.Types.Genotype.RefLikeNoCall;
}
else
{
mnv.Genotype = Pisces.Domain.Types.Genotype.AltLikeNoCall;
}
mnv.GenotypeQscore = 0;
}
if (mnv.Frequency < _callerParams.MinimumFrequency)
{
if (isReference)
{
mnv.Genotype = Pisces.Domain.Types.Genotype.RefLikeNoCall;
}
else
{
mnv.Genotype = Pisces.Domain.Types.Genotype.AltLikeNoCall;
}
mnv.VariantQscore = 0;
mnv.GenotypeQscore = 0;
}
//as an alternative to setting them to nocalls, we could just omit them from gvcf.
}
/// <summary>
/// Step forward with the reader, assembling a list of variants at your CurrentVariant position.
/// </summary>
/// <param name="Reader"></param>
/// <param name="CurrentVariant"></param>
/// <param name="BackLogExists"></param>
/// <param name="TheBackLog"></param>
/// <returns></returns>
private static List <CalledAllele> AssembleColocatedList(
AlleleReader Reader, CalledAllele CurrentVariant, AlleleCompareByLoci alleleOrdering,
ref bool BackLogExists, ref List <CalledAllele> TheBackLog)
{
List <CalledAllele> CoLocatedVariants = new List <CalledAllele>();
bool ContinueReadA = true;
var NextVariantList = new List <CalledAllele>();
while (ContinueReadA)
{
if (BackLogExists)
{
NextVariantList = TheBackLog;
BackLogExists = false;
}
else
{
ContinueReadA = Reader.GetNextVariants(out NextVariantList);
if (!ContinueReadA)
{
break;
}
}
// VarOrder = -1 if Current comes first, 0 if co-located.
int VarOrder = (alleleOrdering.OrderAlleles(CurrentVariant, NextVariantList.First()));
switch (VarOrder)
{
case 0: //the variant we just got is at out current position
CoLocatedVariants.AddRange(NextVariantList);
break;
case -1: //the variant we just got is after our current position, and needs to go to the backlog.
TheBackLog = NextVariantList; //NextVariant;
ContinueReadA = false;
BackLogExists = true;
break;
default: //
{
throw new InvalidDataException("Vcf needs to be ordered.");
}
}
}
if (!BackLogExists)
{
TheBackLog = null;
}
return(CoLocatedVariants);
}
private static bool IsPotentialOverlap(CalledAllele callableAllele, CalledAllele failedMnv)
{
return(callableAllele.Coordinate >= failedMnv.Coordinate &&
callableAllele.Chromosome == failedMnv.Chromosome &&
callableAllele.Coordinate <= (failedMnv.Coordinate + failedMnv.Alternate.Length) &&
callableAllele.Alternate.Length <= failedMnv.Alternate.Length &&
callableAllele.Coordinate + callableAllele.Alternate.Length <= (failedMnv.Coordinate + failedMnv.Alternate.Length) &&
(callableAllele.Type == AlleleCategory.Mnv ||
callableAllele.Type == AlleleCategory.Snv ||
callableAllele.Type == AlleleCategory.Reference));
}
public static int OrderVariants(CalledAllele a, CalledAllele b, bool mFirst)
{
var vcfVariantA = new VcfVariant {
ReferencePosition = a.Coordinate, ReferenceName = a.Chromosome
};
var vcfVariantB = new VcfVariant {
ReferencePosition = b.Coordinate, ReferenceName = b.Chromosome
};
return(Extensions.OrderVariants(vcfVariantA, vcfVariantB, mFirst));
}
public static double GetAlternateAlleleFrequency(CalledAllele variant)
{
if (variant.HasAnAltAllele)
{
return(variant.Frequency);
}
else
{
return((double)GetAlternateAlleleSupport(variant) / variant.TotalCoverage);
}
}
public static int GetAlternateAlleleSupport(CalledAllele variant)
{
if (variant.HasAnAltAllele)
{
return(variant.AlleleSupport);
}
else
{
return(variant.TotalCoverage - variant.AlleleSupport);
}
}
public static MutationCategory GetMutationCategory(
CalledAllele variant)
{
if (variant.Type == Pisces.Domain.Types.AlleleCategory.Reference)
{
return(MutationCategory.Reference);
}
return(GetMutationCategory(variant.ReferenceAllele, variant.AlternateAllele));
}
private void ComputeCoverageTestInternalWeighted(CalledAllele variant, List <AlleleCount> stagedCounts,
int expectedCoverageLowerBound, int expectedCoverageUpperBound, float expectedWeightLowerBound, float expectedWeightUpperBound)
{
var mockStateManager = CreateMockStateManager(stagedCounts, 0);
new CoverageCalculator(true).Compute(variant, mockStateManager);
Assert.True(expectedCoverageLowerBound <= variant.TotalCoverage);
Assert.True(expectedCoverageUpperBound >= variant.TotalCoverage);
Assert.True(expectedWeightLowerBound <= variant.UnanchoredCoverageWeight);
Assert.True(expectedWeightUpperBound >= variant.UnanchoredCoverageWeight);
}
private static bool CheckIfUsed(List <CalledAllele> usedAlleles, CalledAllele originalCall)
{
foreach (var allele in usedAlleles)
{
if (originalCall.IsSameAllele(allele))
{
return(true);
}
}
return(false);
}
private static CalledAllele SetUpCalledAllele()
{
var v = new CalledAllele();
v.TotalCoverage = 100;
v.ReferenceSupport = 90;
v.AlleleSupport = 10;
v.GenotypeQscore = 72;
v.NoiseLevelApplied = 20;
v.VariantQscore = 666;
return(v);
}
public static void CheckVariantsMatch(VcfVariant baseline, CalledAllele test)
{
Assert.Equal(baseline.ReferenceAllele, test.ReferenceAllele);
Assert.Equal(baseline.VariantAlleles[0], test.AlternateAllele);
Assert.Equal(baseline.VariantAlleles.Length, 1);
Assert.Equal(baseline.ReferenceName, test.Chromosome);
Assert.Equal(baseline.ReferencePosition, test.ReferencePosition);
int numAlts = (baseline.VariantAlleles[0] == ".") ? 0 : baseline.VariantAlleles.Length;
Assert.Equal(VcfVariantUtilities.MapGTString(baseline.Genotypes[0]["GT"], numAlts), test.Genotype);
}
public void CandidateAllele_CheckType()
{
var allele = new CalledAllele()
{
ReferenceAllele = "A", AlternateAllele = "."
};
allele.SetType();
Assert.Equal(AlleleCategory.Reference, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "A", AlternateAllele = "A"
};
allele.SetType();
Assert.Equal(AlleleCategory.Reference, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "A", AlternateAllele = "C"
};
allele.SetType();
Assert.Equal(AlleleCategory.Snv, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "AC", AlternateAllele = "CG"
};
allele.SetType();
Assert.Equal(AlleleCategory.Mnv, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "AAA", AlternateAllele = "A"
};
allele.SetType();
Assert.Equal(AlleleCategory.Deletion, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "A", AlternateAllele = "ACG"
};
allele.SetType();
Assert.Equal(AlleleCategory.Insertion, allele.Type);
allele = new CalledAllele()
{
ReferenceAllele = "AFGGGG", AlternateAllele = "ACG"
};
allele.SetType();
Assert.Equal(AlleleCategory.Unsupported, allele.Type);
}
private static void ApplyFilters(CalledAllele allele, int?minCoverageFilter, int?variantQscoreThreshold, bool filterSingleStrandVariants, float?variantFreqFilter, float?lowGenotypeqFilter, int?indelRepeatFilter,
RMxNFilterSettings rMxNFilterSettings, bool hasStitchedSource, ChrReference chrReference)
{
//Reset filters
allele.Filters.Clear();
if (minCoverageFilter.HasValue && allele.TotalCoverage < minCoverageFilter)
{
allele.AddFilter(FilterType.LowDepth);
}
if (variantQscoreThreshold.HasValue && allele.VariantQscore < variantQscoreThreshold && (allele.TotalCoverage != 0))
{
//note we wont flag it for Qscore, if its got zero depth, because in that case, the Q score calc was not made anyway.
allele.AddFilter(FilterType.LowVariantQscore);
}
if (allele.Type != AlleleCategory.Reference)
{
if (!allele.StrandBiasResults.BiasAcceptable ||
(filterSingleStrandVariants && !allele.StrandBiasResults.VarPresentOnBothStrands))
{
allele.AddFilter(FilterType.StrandBias);
}
if (indelRepeatFilter.HasValue && indelRepeatFilter > 0)
{
var indelRepeatLength = ComputeIndelRepeatLength(allele, chrReference.Sequence);
if (indelRepeatFilter <= indelRepeatLength)
{
allele.AddFilter(FilterType.IndelRepeatLength);
}
}
if (RMxNCalculator.ShouldFilter(allele, rMxNFilterSettings, chrReference.Sequence))
{
allele.AddFilter(FilterType.RMxN);
}
if (variantFreqFilter.HasValue && allele.Frequency < variantFreqFilter)
{
allele.AddFilter(FilterType.LowVariantFrequency);
}
if (hasStitchedSource) //can only happen for insertions and MNVs
{
if (allele.Alternate.Contains("N"))
{
allele.AddFilter(FilterType.StrandBias);
}
}
}
}
// jg todo - set numnocalls - appears to only be applicable to SNVs
private static void SetFractionNoCall(CalledAllele allele)
{
var allReads = (float)(allele.TotalCoverage + allele.NumNoCalls);
if (allReads == 0)
{
allele.FractionNoCalls = 0;
}
else
{
allele.FractionNoCalls = allele.NumNoCalls / allReads;
}
}
public void GetMultiAllelicQScores()
{
CalledAllele variant1 = TestHelper.CreateDummyAllele("chr1", 1000, "A", "C", 30, 12);
CalledAllele variant2 = TestHelper.CreateDummyAllele("chr1", 1000, "A", "T", 30, 11);
MixtureModelResult result = AdaptiveGenotyperCalculator.GetMultiAllelicQScores(variant1, variant2,
new List <double[]> {
Means, Means
});
// The 4th GP should always be the minimum because that reflects the 1/2 call
Assert.Equal(4, result.GenotypePosteriors.ToList().IndexOf(result.GenotypePosteriors.Min()));
}
protected virtual void CalculateSinglePoint(CalledAllele allele, IAlleleSource alleleCountSource)
{
//TODO: Is there a reason why we don't reallocate the stitched coverage here for point mutations? (as we do with spanning ones)
// sum up all observations at that point
var variant = allele as CalledAllele;
for (var direction = 0; direction < Constants.NumDirectionTypes; direction++)
{
foreach (var alleleType in Constants.CoverageContributingAlleles)
{
allele.EstimatedCoverageByDirection[direction] += alleleCountSource.GetAlleleCount(allele.ReferencePosition, alleleType, (DirectionType)direction);
allele.SumOfBaseQuality += alleleCountSource.GetSumOfAlleleBaseQualities(allele.ReferencePosition, alleleType, (DirectionType)direction);
if (alleleType != AlleleHelper.GetAlleleType(allele.ReferenceAllele))
{
continue;
}
if (variant != null)
{
variant.ReferenceSupport += alleleCountSource.GetAlleleCount(variant.ReferencePosition, alleleType,
(DirectionType)direction);
}
}
allele.TotalCoverage += allele.EstimatedCoverageByDirection[direction];
// For single point variants, for now, we're calling everything confident coverage
allele.ConfidentCoverageStart += allele.EstimatedCoverageByDirection[direction];
allele.ConfidentCoverageEnd += allele.EstimatedCoverageByDirection[direction];
allele.NumNoCalls += alleleCountSource.GetAlleleCount(allele.ReferencePosition, AlleleType.N, (DirectionType)direction);
}
// adjust for reference counts already taken up by gapped mnvs
// note: it's possible that the ref count taken up by a gapped mnv is greater than depth at that ref position.
// this is possible when collapsing is true, and some gapped ref positions have low quality (or are N).
// in these cases, they get collapsed to the mnv and count towards support, but those specific alleles were never added to region's allele counts because they are low quality.
// collapsing is the correct thing to do, so this is ok. we should just make sure to cap at 0.
var gappedRefCounts = alleleCountSource.GetGappedMnvRefCount(allele.ReferencePosition);
if (allele.Type == AlleleCategory.Snv && variant != null)
{
variant.ReferenceSupport = Math.Max(0, variant.ReferenceSupport - gappedRefCounts);
}
else if (allele.Type == AlleleCategory.Reference)
{
allele.AlleleSupport = Math.Max(0, allele.AlleleSupport - gappedRefCounts);
}
}
public CalledAllele ReCallAsRef(CalledAllele usedVariant, int numRefCallsSuckedUpByOtherVariants)
{
var newRef = PhasedVariantExtractor.Create(
usedVariant.Chromosome, usedVariant.Coordinate,
usedVariant.Reference.Substring(0, 1), ".",
Math.Max(0, usedVariant.ReferenceSupport - numRefCallsSuckedUpByOtherVariants), usedVariant.TotalCoverage, Pisces.Domain.Types.AlleleCategory.Reference,
_bamParams.MinimumBaseCallQuality, _callerParams.MaximumVariantQScore);
CallCandidate(newRef, true);
AddFilters(newRef, true);
return(newRef);
}
/// <summary>
/// Assign a q-score to a SNP, given (CallCount / Coverage) frequency.
/// </summary>
public static void Compute(CalledAllele allele, int maxQScore, int estimatedBaseCallQuality)
{
allele.NoiseLevelApplied = estimatedBaseCallQuality;
if (allele.TotalCoverage == 0)
{
allele.VariantQscore = 0;
}
else
{
allele.VariantQscore = AssignPoissonQScore(allele.AlleleSupport, allele.TotalCoverage, estimatedBaseCallQuality,
maxQScore);
}
}
private static int CombineQualitiesByTakingMinValue(CalledAllele VariantA, CalledAllele VariantB)
{
if (VariantB == null)
{
return(VariantA.VariantQscore);
}
if (VariantA == null)
{
return(VariantB.VariantQscore);
}
return(Math.Min(VariantA.VariantQscore, VariantB.VariantQscore));
}
public static void Compute(CalledAllele variant, int maxQScore, float?filterThreshold)
{
if (filterThreshold.HasValue)
{
var acceptanceCriteria = (float)filterThreshold;
//restrict use to SNPs for now. We have not done any testing on this for indels, and indel coverage calc for amplicons would need special handling.
if (variant.Type == AlleleCategory.Snv)
{
variant.AmpliconBiasResults = CalculateAmpliconBias(variant.SupportByAmplicon, variant.CoverageByAmplicon,
acceptanceCriteria, maxQScore);
}
}
}
/// <summary>
/// Calculates the Q score and genotype posteriors of a 1/2 locus using a multinomial distribution.
/// This method is called from Pisces variant caller.
/// </summary>
/// <param name="allele1"></param>
/// <param name="allele2"></param>
/// <param name="models">IList of models for allele1 and allele 2. Each model is a 3 element double array.</param>
/// <returns>RecalibratedVariant that contains the Q score and genotype posteriors.</returns>
public static MixtureModelResult GetMultiAllelicQScores(CalledAllele allele1, CalledAllele allele2,
IList <double[]> models)
{
int totalCoverage = allele1.TotalCoverage;
int[] alleleDepths = new int[3];
alleleDepths[2] = allele2.AlleleSupport;
alleleDepths[1] = allele1.AlleleSupport;
// "Reference" here is not really reference--it is just everything else that was not the top two alleles
alleleDepths[0] = Math.Max(totalCoverage - alleleDepths[1] - alleleDepths[2], 0);
return(MixtureModel.GetMultinomialQScores(alleleDepths, totalCoverage, models));
}
/// <summary>
/// Combines two variants made by PhasedVariantExtractor.
/// Simplistic in that it only worries about the relevant fields, and not every possible field.
/// LIMITATION: this assumes that allele1 and allele2 are the same variant in terms of chr, pos, ref / alt alleles.
/// see CalledAllele.IsSameAlle() . This should be true for the variants in this method
/// </summary>
/// <param name="allele1"></param>
/// <param name="allele2"></param>
/// <param name="maxQscore"></param>
/// <returns></returns>
public static CalledAllele CombinePhasedVariants(CalledAllele allele1, CalledAllele allele2, int maxQscore)
{
//NOTE this assumes that allele1 and allele2 are the same variant in terms of chr, pos, ref / alt alleles.
//see CalledAllele.IsSameAlle() . This should be true for the variants in this method
CalledAllele result = Create(allele1.Chromosome, allele1.ReferencePosition, allele1.ReferenceAllele, allele1.AlternateAllele,
allele1.AlleleSupport + allele2.AlleleSupport,
(allele1.NumNoCalls + allele2.NumNoCalls) / 2,
(allele1.TotalCoverage + allele2.TotalCoverage) / 2,
(allele1.ReferenceSupport + allele2.ReferenceSupport) / 2,
allele1.Type, allele1.NoiseLevelApplied,
maxQscore);
return(result);
}
private void TestCalculation(CalledAllele variant, double frequency, double depth, int expectedValue)
{
variant.TotalCoverage = (int)depth;
variant.AlleleSupport = (int)(depth * frequency);
if (variant.Genotype == Genotype.HomozygousRef)
{
variant.AlleleSupport = (int)(depth * (1.0 - frequency));
}
MixtureModelResult result = AdaptiveGenotyperCalculator.GetGenotypeAndQScore(variant, Means, Priors);
Assert.Equal(expectedValue, result.QScore);
}
private static void TestCalculation(CalledAllele variant, double frequency, double depth, int expectedValue)
{
variant.TotalCoverage = (int)depth;
variant.AlleleSupport = (int)(depth * frequency);
if (variant.Genotype == Genotype.HomozygousRef)
{
variant.AlleleSupport = (int)(depth * (1.0 - frequency));
}
int result = DiploidGenotypeQualityCalculator.Compute(variant, 0, int.MaxValue);
Assert.Equal(expectedValue, result);
}
public static void PrintBiasStats(StreamWriter writer, CalledAllele variant)
{
if (variant.ReferenceAllele == variant.AlternateAllele)
{
return; //skip ref calls.
}
var strandBiasResults = StatsToString(variant.StrandBiasResults);
StringBuilder sb = new StringBuilder(string.Format("{0}\t{1}\t{2}\t{3}\t",
variant.Chromosome, variant.ReferencePosition, variant.ReferenceAllele,
variant.AlternateAllele));
sb.Append(strandBiasResults);
writer.WriteLine(sb.ToString());
}
/// <summary>
/// Calculates repeats for insertions and deletions by scanning up to 50 base pairs of the chromosome reference on either side of the allele coordinate.
/// Duplicates of the allele alternate found in the reference are summed to compute the overall repeat length.. Useful for filtering some
/// indels - e.g. we mistrust a call of AAAAAAAA versus reference AAAAAAAAA, since it may be
/// polymerase slippage during PCR in sample prep, rather than an actual mutation.
/// </summary>
private static int ComputeIndelRepeatLength(CalledAllele allele, string referenceBases)
{
if (String.IsNullOrEmpty(referenceBases))
{
return(0);
}
if (allele.Type != AlleleCategory.Insertion && allele.Type != AlleleCategory.Deletion && allele.Type != AlleleCategory.Snv)
{
return(0);
}
// Logic from GetFlankingBases:
var stringPos = allele.ReferencePosition - 1;
var upstreamBegin = stringPos - FlankingBaseCount;
var upstreamEnd = stringPos - 1;
var downstreamBegin = stringPos;
var downstreamEnd = stringPos + FlankingBaseCount - 1;
if (upstreamBegin < 0)
{
upstreamBegin = 0;
}
if (downstreamBegin < 0)
{
downstreamBegin = 0;
}
if (downstreamEnd >= referenceBases.Length)
{
downstreamEnd = referenceBases.Length - 1;
}
if (upstreamEnd >= referenceBases.Length)
{
upstreamEnd = referenceBases.Length - 1;
}
var upstreamBases = String.Empty;
if (upstreamEnd >= 0)
{
upstreamBases =
referenceBases.Substring(upstreamBegin, upstreamEnd - upstreamBegin + 1).ToUpper();
}
var downstreamBases =
referenceBases.Substring(downstreamBegin, downstreamEnd - downstreamBegin + 1)
.ToUpper();
var longestRepeatLength = CheckVariantRepeatCount(allele, upstreamBases, downstreamBases);
return(longestRepeatLength);
}
public void AddRejectedPhasedVariant(CalledAllele variant)
{
var match = _rejectedPhasedVariants.Find(v => v.IsSameAllele(variant));
if (match == null)
{
_rejectedPhasedVariants.Add(variant);
}
else
{
var combinedVar = PhasedVariantExtractor.CombinePhasedVariants(match, variant, MaxQScore);
_rejectedPhasedVariants.Remove(match);
_rejectedPhasedVariants.Add(combinedVar);
}
}
private bool MatchVariants(CalledAllele BaseCalledAllele, CandidateAllele candidateVariant, int?expectedSupport = null, float?expectedFreq = null)
{
if (BaseCalledAllele.Chromosome == candidateVariant.Chromosome &&
BaseCalledAllele.ReferencePosition == candidateVariant.ReferencePosition &&
BaseCalledAllele.ReferenceAllele == candidateVariant.ReferenceAllele &&
BaseCalledAllele.AlternateAllele == candidateVariant.AlternateAllele &&
BaseCalledAllele.Type == candidateVariant.Type &&
(expectedFreq == null || BaseCalledAllele.Frequency == expectedFreq) &&
(expectedSupport == null || BaseCalledAllele.AlleleSupport == expectedSupport)
)
{
return(true);
}
return(false);
}