/// <summary>
/// make sure a component doesn't "invade" the other components
/// </summary>
/// <param name="modelPoints"></param>
private void EMScaleCovariancesPairwise(List <ModelPoint> modelPoints)
{
foreach (var m1 in modelPoints)
{
ModelPoint maxM = null;
double maxProb = 0;
foreach (var m2 in modelPoints)
{
if (m2 == m1)
{
continue;
}
// pretend m1 is a segment
double prob = m2.Ploidy.Omega * Sigma(m1.MAF, m1.Coverage, m2.Ploidy.Mu, m2.Ploidy.Sigma);
if (prob > maxProb)
{
maxProb = prob;
maxM = m2;
}
}
if (maxProb > 0)
{
double[][] s1 = m1.Ploidy.Sigma;
double det1 = s1[0][0] * s1[1][1] - s1[0][1] * s1[1][0];
double[][] s2 = maxM.Ploidy.Sigma;
double det2 = s2[0][0] * s2[1][1] - s2[0][1] * s2[1][0];
if (det1 <= 1E-7 || det2 <= 1E-7)
{
continue;
}
double ratio = det1 > det2 ? det1 / det2 : det2 / det1;
if (ratio < 4)
{
continue;
}
if (det1 > det2)
{
Scale2DMatrix(s1, 0.8);
Scale2DMatrix(s2, 1.1);
}
else
{
Scale2DMatrix(s2, 0.8);
Scale2DMatrix(s1, 1.1);
}
}
}
}
protected List<ModelPoint> InitializeModelPoints(CoverageModel model)
{
List<ModelPoint> modelPoints = new List<ModelPoint>();
double[] mu = GetProjectedMeanCoverage(model.DiploidCoverage);
double diploidMAF = this.AllPloidies[3].MinorAlleleFrequency; /// %%% Magic number!
// Refine our estimate of diploid MAF:
//double diploidMAF = this.EstimateDiploidMAF(2, model.DiploidCoverage);
/////////////////////////////////////////////
// Update the parameters in each SegmentPloidy object, and construct corresponding SegmentInfo objects
foreach (SegmentPloidy ploidy in this.AllPloidies)
{
ModelPoint point = new ModelPoint();
double pureCoverage = mu[ploidy.CopyNumber];
point.Coverage = pureCoverage;
double pureMAF = ploidy.MinorAlleleFrequency;
point.MAF = pureMAF;
if (double.IsNaN(point.MAF)) point.MAF = 0;
point.Ploidy = ploidy;
modelPoints.Add(point);
point.CN = ploidy.CopyNumber;
ploidy.MixedMinorAlleleFrequency = point.MAF;
ploidy.MixedCoverage = point.Coverage;
}
return modelPoints;
}
/// <summary>
/// Initialize model points given diploid purity modelInitialize model points given somatic purity model
/// </summary>
protected List<ModelPoint> InitializeModelPoints(List<SegmentInfo> segments, double coverage, int percentPurity, int numClusters)
{
List<ModelPoint> modelPoints = new List<ModelPoint>();
CoveragePurityModel model = new CoveragePurityModel();
model.DiploidCoverage = coverage;
model.Purity = percentPurity / 100f;
double[] mu = GetProjectedMeanCoverage(model.DiploidCoverage);
double diploidMAF = this.AllPloidies[3].MinorAlleleFrequency; /// %%% Magic number!
/////////////////////////////////////////////
// Update the parameters in each SegmentPloidy object, and construct corresponding SegmentInfo objects
foreach (SegmentPloidy ploidy in this.AllPloidies)
{
ModelPoint point = new ModelPoint();
double pureCoverage = mu[ploidy.CopyNumber];
point.Coverage = (model.Purity * pureCoverage) + (1 - model.Purity) * model.DiploidCoverage;
double pureMAF = ploidy.MinorAlleleFrequency;
if (ploidy.MajorChromosomeCount * 2 == ploidy.CopyNumber)
{
point.MAF = (model.Purity * ploidy.CopyNumber * pureMAF) + ((1 - model.Purity) * 2 * diploidMAF);
point.MAF /= model.Purity * ploidy.CopyNumber + (1 - model.Purity) * 2;
if (double.IsNaN(point.MAF)) point.MAF = 0;
}
else
{
point.MAF = (model.Purity * ploidy.CopyNumber * pureMAF) + ((1 - model.Purity) * 1);
point.MAF /= model.Purity * ploidy.CopyNumber + (1 - model.Purity) * 2;
}
point.Ploidy = ploidy;
modelPoints.Add(point);
point.CN = ploidy.CopyNumber;
ploidy.MixedMinorAlleleFrequency = point.MAF;
ploidy.MixedCoverage = point.Coverage;
}
// estimate distance between each model point and segments
List<double> modelPointsScore = new List<double>();
foreach (ModelPoint modelPoint in modelPoints)
{
List<double> distanceList = new List<double>();
foreach (SegmentInfo info in segments)
{
if (info.MAF >= 0)
distanceList.Add(GetModelDistance(info.Coverage, modelPoint.Coverage, info.MAF, modelPoint.MAF));
}
distanceList.Sort();
double v15th_percentile = distanceList[Convert.ToInt32(distanceList.Count * 0.15)];
// use model points with good fit to observed values
modelPointsScore.Add(v15th_percentile);
}
// sort list and return indices
var sortedScores = modelPointsScore.Select((x, i) => new KeyValuePair<double, int>(x, i)).OrderBy(x => x.Key).ToList();
List<double> scoresValue = sortedScores.Select(x => x.Key).ToList();
List<int> scoresIndex = sortedScores.Select(x => x.Value).ToList();
List<ModelPoint> selectedModelPoints = new List<ModelPoint>();
for (int i = 0; i < numClusters; i++)
{
modelPoints[scoresIndex[i]].Cluster = i + 1;
selectedModelPoints.Add(modelPoints[scoresIndex[i]]);
}
return selectedModelPoints;
}
// Initialize model points given expected ploidy and purity values
protected List<ModelPoint> InitializeModelPoints(CoveragePurityModel model)
{
List<ModelPoint> modelPoints = new List<ModelPoint>();
double[] mu = GetProjectedMeanCoverage(model.DiploidCoverage);
double diploidMAF = this.AllPloidies[3].MinorAlleleFrequency; /// %%% Magic number!
/////////////////////////////////////////////
// Update the parameters in each SegmentPloidy object, and construct corresponding SegmentInfo objects
foreach (SegmentPloidy ploidy in this.AllPloidies)
{
ModelPoint point = new ModelPoint();
double pureCoverage = mu[ploidy.CopyNumber];
point.Coverage = (model.Purity * pureCoverage) + (1 - model.Purity) * model.DiploidCoverage;
double pureMAF = ploidy.MinorAlleleFrequency;
if (ploidy.MajorChromosomeCount * 2 == ploidy.CopyNumber)
{
point.MAF = (model.Purity * ploidy.CopyNumber * pureMAF) + ((1 - model.Purity) * 2 * diploidMAF);
point.MAF /= model.Purity * ploidy.CopyNumber + (1 - model.Purity) * 2;
if (double.IsNaN(point.MAF)) point.MAF = 0;
}
else
{
point.MAF = (model.Purity * ploidy.CopyNumber * pureMAF) + ((1 - model.Purity) * 1);
point.MAF /= model.Purity * ploidy.CopyNumber + (1 - model.Purity) * 2;
}
point.Ploidy = ploidy;
modelPoints.Add(point);
point.CN = ploidy.CopyNumber;
ploidy.MixedMinorAlleleFrequency = point.MAF;
ploidy.MixedCoverage = point.Coverage;
}
return modelPoints;
}
/// <summary>
/// Initialize model points by subsampling from existing segment Coverage and MAF values.
/// Use distanceThreshold to ensure that both large and small cluster components get subsampled
/// </summary>
protected List<ModelPoint> InitializeModelPoints(List<SegmentInfo> segments, int numClusters, double distanceThreshold)
{
List<ModelPoint> modelPoints = new List<ModelPoint>();
List<SegmentInfo> usableSegments = new List<SegmentInfo>();
List<SegmentInfo> usedSegments = new List<SegmentInfo>();
foreach (SegmentInfo segment in segments) {
if (segment.Cluster != -1 && segment.MAF >= 0)
usableSegments.Add(segment);
}
Random rnd = new Random();
int lastIndex = rnd.Next(1, usableSegments.Count);
usedSegments.Add(usableSegments[lastIndex]);
int counter = 1;
double attempts = 0;
while (counter < numClusters)
{
int newIndex = rnd.Next(1, usableSegments.Count);
attempts += 1.0;
double distance = GetModelDistance(usableSegments[lastIndex].Coverage, usableSegments[newIndex].Coverage, usableSegments[lastIndex].MAF, usableSegments[newIndex].MAF);
if (distance > distanceThreshold || attempts/usableSegments.Count > 0.3) // escape outlier minima
{
usedSegments.Add(usableSegments[newIndex]);
counter++;
lastIndex = newIndex;
attempts = 0;
}
}
// Initialize model points with coverage and MAF values from subsampled segments
for (int i = 0; i < numClusters; i++)
{
ModelPoint point = new ModelPoint();
point.Coverage = usedSegments[i].Coverage;
point.MAF = usedSegments[i].MAF;
SegmentPloidy ploidy = new SegmentPloidy();
ploidy.CopyNumber = 2;
ploidy.MajorChromosomeCount = 1;
point.Ploidy = ploidy;
point.Cluster = i+1;
modelPoints.Add(point);
}
return modelPoints;
}
