/// <summary>
/// Setup: Model various copy ploidies.
/// </summary>
public void InitializePloidies()
{
Console.WriteLine("{0} Initialize ploidy models...", DateTime.Now);
_allPloidies = new List <SegmentPloidy>();
Utilities.EstimateDiploidMAF(2, MeanCoverage);
for (int copyNumber = 0; copyNumber <= MaximumCopyNumber; copyNumber++)
{
for (int majorCount = copyNumber; majorCount * 2 >= copyNumber; majorCount--)
{
SegmentPloidy ploidy = new SegmentPloidy
{
CopyNumber = copyNumber,
MajorChromosomeCount = majorCount,
Index = _allPloidies.Count
};
_allPloidies.Add(ploidy);
if (copyNumber == 0)
{
ploidy.MinorAlleleFrequency = 0.01; // should reflect sequencing error rate
continue;
}
float variantFrequency = majorCount / (float)copyNumber;
ploidy.MinorAlleleFrequency = variantFrequency < 0.5 ? variantFrequency : 1 - variantFrequency;
if (majorCount * 2 == copyNumber)
{
ploidy.MinorAlleleFrequency = Utilities.EstimateDiploidMAF(copyNumber, MeanCoverage);
}
}
}
Console.WriteLine("{0} Ploidy models prepared.", DateTime.Now);
}
/// <summary>
/// Setup: Model various copy ploidies.
/// </summary>
public void InitializePloidies()
{
Console.WriteLine("{0} Initialize ploidy models...", DateTime.Now);
this.AllPloidies = new List <SegmentPloidy>();
double diploidPredictedMAF = CanvasCommon.Utilities.EstimateDiploidMAF(2, this.MeanCoverage);
for (int copyNumber = 0; copyNumber <= MaximumCopyNumber; copyNumber++)
{
for (int majorCount = copyNumber; majorCount * 2 >= copyNumber; majorCount--)
{
SegmentPloidy ploidy = new SegmentPloidy();
ploidy.CopyNumber = copyNumber;
ploidy.MajorChromosomeCount = majorCount;
ploidy.ID = AllPloidies.Count;
AllPloidies.Add(ploidy);
if (copyNumber == 0)
{
ploidy.MinorAlleleFrequency = 0.01; // should reflect sequencing error rate
continue;
}
float VF = majorCount / (float)copyNumber;
ploidy.MinorAlleleFrequency = (VF < 0.5 ? VF : 1 - VF);
if (majorCount * 2 == copyNumber)
{
ploidy.MinorAlleleFrequency = CanvasCommon.Utilities.EstimateDiploidMAF(copyNumber, this.MeanCoverage);
}
}
}
Console.WriteLine("{0} Ploidy models prepared.", DateTime.Now);
}
/// <summary>
/// Assign a SegmentPloidy to each CanvasSegment, based on which model matches this segment best:
/// </summary>
void AssignPloidyCallsGaussianMixture()
{
// For segments with (almost) no variants alleles at all, we'll assign them a dummy MAF, and
// we simply won't consider MAF when determining the closest ploidy:
double dummyMAF = -1;
foreach (CanvasSegment segment in this.Segments)
{
// Compute (MAF, Coverage) for this segment:
List <double> MAF = new List <double>();
foreach (float VF in segment.VariantFrequencies)
{
MAF.Add(VF > 0.5 ? 1 - VF : VF);
}
double medianCoverage = CanvasCommon.Utilities.Median(segment.Counts);
double medianMAF = dummyMAF;
SegmentPloidy bestPloidy = null;
double bestProbability = 0;
if (MAF.Count >= 10)
{
medianMAF = Utilities.Median(MAF);
}
Dictionary <SegmentPloidy, double> posteriorProbabilities = GaussianMixtureModel.EMComputePosteriorProbs(AllPloidies, medianMAF, medianCoverage);
// Find the closest ploidy.
foreach (SegmentPloidy ploidy in AllPloidies)
{
if (bestPloidy == null || posteriorProbabilities[ploidy] > bestProbability)
{
bestProbability = posteriorProbabilities[ploidy];
bestPloidy = ploidy;
}
}
if (bestProbability == 0)
{
// Sanity-check: If we didn't find anything with probability > 0, then fall back to the simplest possible
// thing: Call purely on coverage.
segment.CopyNumber = (int)Math.Round(2 * medianCoverage / this.DiploidCoverage);
segment.MajorChromosomeCount = null;
}
else
{
segment.CopyNumber = bestPloidy.CopyNumber;
segment.MajorChromosomeCount = bestPloidy.MajorChromosomeCount;
if (MAF.Count < 10)
{
segment.MajorChromosomeCount = null; // Don't assign MCC if we don't have variant allele frequencies
}
}
}
}
private void AssignPloidyCallsDistance(CoverageModel model)
{
InitializeModelPoints(model);
foreach (CanvasSegment segment in _allSegments)
{
// Compute (MAF, Coverage) for this segment:
List <double> mafs = new List <double>();
foreach (float variantFrequency in segment.Balleles.Frequencies)
{
mafs.Add(variantFrequency > 0.5 ? 1 - variantFrequency : variantFrequency);
}
int expectedSnpDensityCutoff = (segment.Length) / MedianHetSnpsDistance / 2;
double medianCoverage = Utilities.Median(segment.Counts);
double medianMaf = -1;
SegmentPloidy bestPloidy = null;
if (mafs.Count >= Math.Max(10, expectedSnpDensityCutoff))
{
medianMaf = Utilities.Median(mafs);
}
double bestDistance = double.MaxValue;
double secondBestDistance = double.MaxValue;
foreach (SegmentPloidy ploidy in _allPloidies)
{
double diff = (ploidy.MixedCoverage - medianCoverage) * _coverageWeightingFactor;
double distance = diff * diff;
if (mafs.Count >= Math.Max(10, expectedSnpDensityCutoff))
{
diff = ploidy.MixedMinorAlleleFrequency - medianMaf;
distance += diff * diff;
}
if (distance < bestDistance)
{
secondBestDistance = bestDistance;
bestDistance = distance;
bestPloidy = ploidy;
}
else if (distance < secondBestDistance)
{
secondBestDistance = distance;
}
}
if (bestPloidy != null)
{
segment.CopyNumber = bestPloidy.CopyNumber;
segment.MajorChromosomeCount = bestPloidy.MajorChromosomeCount;
}
segment.ModelDistance = bestDistance;
segment.RunnerUpModelDistance = secondBestDistance;
if (mafs.Count < 10)
{
segment.MajorChromosomeCount = null; // Don't assign MCC if we don't have variant allele frequencies
}
}
}
private void AssignPloidyCallsDistance(CoverageModel model, List <SegmentInfo> segments, int medianVariantCoverage)
{
List <ModelPoint> modelPoints = InitializeModelPoints(model);
foreach (CanvasSegment segment in this.Segments)
{
// Compute (MAF, Coverage) for this segment:
List <double> MAF = new List <double>();
foreach (float VF in segment.VariantFrequencies)
{
MAF.Add(VF > 0.5 ? 1 - VF : VF);
}
int expectedSnpDensityCutoff = (segment.End - segment.Begin) / MedianHetSnpsDistance / 2;
List <Tuple <float, float> > weightedVariantFrequencies = new List <Tuple <float, float> >();
double medianCoverage = CanvasCommon.Utilities.Median(segment.Counts);
double medianMAF = -1;
SegmentPloidy bestPloidy = null;
if (MAF.Count >= Math.Max(10, expectedSnpDensityCutoff))
{
medianMAF = Utilities.Median(MAF);
}
double bestDistance = double.MaxValue;
double secondBestDistance = double.MaxValue;
foreach (SegmentPloidy ploidy in AllPloidies)
{
double diff = (ploidy.MixedCoverage - medianCoverage) * CoverageWeightingFactor;
double distance = diff * diff;
if (MAF.Count >= Math.Max(10, expectedSnpDensityCutoff))
{
diff = ploidy.MixedMinorAlleleFrequency - medianMAF;
distance += diff * diff;
}
if (distance < bestDistance)
{
secondBestDistance = bestDistance;
bestDistance = distance;
bestPloidy = ploidy;
}
else if (distance < secondBestDistance)
{
secondBestDistance = distance;
}
}
segment.CopyNumber = bestPloidy.CopyNumber;
segment.ModelDistance = bestDistance;
segment.RunnerUpModelDistance = secondBestDistance;
segment.MajorChromosomeCount = bestPloidy.MajorChromosomeCount;
if (MAF.Count < 10)
{
segment.MajorChromosomeCount = null; // Don't assign MCC if we don't have variant allele frequencies
}
}
}
