private void EMComputePosteriorProbs(List<ModelPoint> modelPoints, SegmentInfo segment)
{
double tempsum1 = 0;
Dictionary<ModelPoint, double> temp = new Dictionary<ModelPoint, double>();
foreach (var modelPoint in modelPoints)
{
temp[modelPoint] = modelPoint.Ploidy.Omega * Sigma(segment.MAF, segment.Coverage, modelPoint.Ploidy.Mu, modelPoint.Ploidy.Sigma);
tempsum1 += temp[modelPoint];
}
if (segment.PosteriorProbs == null) { segment.PosteriorProbs = new Dictionary<ModelPoint, double>(); }
int bestCluster = 0;
double bestProb = 0;
foreach (var modelPoint in modelPoints)
{
segment.PosteriorProbs[modelPoint] = temp[modelPoint] / tempsum1;
if (segment.PosteriorProbs[modelPoint] > bestProb) {
bestCluster = modelPoint.Cluster.Value;
bestProb = temp[modelPoint] / tempsum1;
}
if (Double.IsNaN(segment.PosteriorProbs[modelPoint])) { segment.PosteriorProbs[modelPoint] = 0; }
}
segment.Cluster = bestCluster;
}
public int CallVariants(string variantFrequencyFile, string inFile, string outFile, string ploidyBedPath, string referenceFolder, string sampleName,
string truthDataPath)
{
if (!string.IsNullOrEmpty(truthDataPath))
{
this.CNOracle = new CopyNumberOracle();
this.CNOracle.LoadKnownCN(truthDataPath);
}
this.Segments = CanvasSegment.ReadSegments(inFile);
this.TempFolder = Path.GetDirectoryName(inFile);
if (this.Segments.Count == 0)
{
Console.WriteLine("CanvasDiploidCaller: No segments loaded; no CNV calls will be made.");
CanvasSegment.WriteSegments(outFile, this.Segments, referenceFolder, sampleName, null, null);
return 0;
}
PloidyInfo ploidy = null;
if (!string.IsNullOrEmpty(ploidyBedPath)) ploidy = PloidyInfo.LoadPloidyFromBedFile(ploidyBedPath);
// load MAF
this.MeanCoverage = CanvasIO.LoadVariantFrequencies(variantFrequencyFile, this.Segments);
int medianVariantCoverage = AggregateVariantCoverage(ref this.Segments);
// Create new models for different copy number states
this.InitializePloidies();
// Compute statistics on the copy number two regions
float[] diploidCounts = AggregateCounts(ref this.Segments);
DiploidCoverage = CanvasCommon.Utilities.Mean(diploidCounts);
CoverageWeightingFactor = CoverageWeighting / DiploidCoverage;
// new coverage model
this.Model = new CoverageModel();
Model.DiploidCoverage = DiploidCoverage;
List<SegmentInfo> segments = new List<SegmentInfo>();
foreach (CanvasSegment segment in this.Segments)
{
SegmentInfo info = new SegmentInfo();
info.Segment = segment;
List<double> MAF = new List<double>();
foreach (float value in segment.VariantFrequencies) MAF.Add(value > 0.5 ? 1 - value : value);
if (MAF.Count > 0)
{
info.MAF = CanvasCommon.Utilities.Median(MAF);
}
else
{
info.MAF = -1;
}
info.Coverage = CanvasCommon.Utilities.Median(segment.Counts);
if (this.Segments.Count > 100)
{
info.Weight = segment.End - segment.Begin;
}
else
{
info.Weight = segment.BinCount;
}
segments.Add(info);
}
// Assign copy number and major chromosome count for each segment
bool useGaussianMixtureModel = false; // For now, this is set false, since we saw weird performance on chrY (CANV-115):
if (useGaussianMixtureModel)
{
// optimize model covariance
double likelihood = FitGaussians(Model, segments);
AssignPloidyCallsGaussianMixture();
}
else
{
AssignPloidyCallsDistance(Model, segments, medianVariantCoverage);
}
// Merge neighboring segments that got the same copy number call.
CanvasSegment.MergeSegments(ref this.Segments);
CanvasSegment.AssignQualityScores(this.Segments, CanvasSegment.QScoreMethod.LogisticGermline);
List<string> extraHeaders = new List<string>();
string coverageOutputPath = CanvasCommon.Utilities.GetCoverageAndVariantFrequencyOutputPath(outFile);
CanvasSegment.WriteCoveragePlotData(this.Segments, Model.DiploidCoverage, ploidy, coverageOutputPath, referenceFolder);
if (this.CNOracle != null)
{
this.GenerateReportVersusKnownCN();
}
if (ploidy != null && !string.IsNullOrEmpty(ploidy.HeaderLine)) extraHeaders.Add(ploidy.HeaderLine);
CanvasSegment.WriteSegments(outFile, this.Segments, referenceFolder, sampleName, extraHeaders, ploidy);
return 0;
}
static public List<SegmentInfo> GetUsableSegmentsForModeling(List<CanvasSegment> segments)
{
// Get the average count everwhere. Exclude segments whose coverage is >2x this average.
float overallMean = 0;
int overallCount = 0;
foreach (CanvasSegment segment in segments)
{
foreach (float value in segment.Counts)
{
overallCount++;
overallMean += value;
}
}
overallMean /= Math.Max(1, overallCount);
List<SegmentInfo> usableSegments = new List<SegmentInfo>();
foreach (CanvasSegment segment in segments)
{
if (segment.End - segment.Begin < 5000) continue;
SegmentInfo info = new SegmentInfo();
info.Segment = segment;
// If the segment has few or no variants, then don't use the MAF for this segment - set to -1 (no frequency)
// Typically a segment will have no variants if it's on chrX or chrY and starling knows not to call a
// heterozygous variant there (other than in the PAR regions).
if (segment.VariantFrequencies.Count < MinimumVariantFrequenciesForInformativeSegment)
{
info.MAF = -1;
}
else
{
List<double> MAF = new List<double>();
foreach (float value in segment.VariantFrequencies) MAF.Add(value > 0.5 ? 1 - value : value);
MAF.Sort();
info.MAF = MAF[MAF.Count / 2];
}
info.Coverage = CanvasCommon.Utilities.Median(segment.Counts);
if (info.Coverage > overallMean * 2) continue;
if (segments.Count > 100)
{
info.Weight = segment.End - segment.Begin;
}
else
{
info.Weight = segment.BinCount;
}
if (segment.VariantFrequencies.Count < 10)
{
info.Weight *= (double)segment.VariantFrequencies.Count / 10;
}
usableSegments.Add(info);
}
return usableSegments;
}
