public Dictionary <string, Segmentation.Segment[]> Run(List <Segmentation> segmentation)
{
Dictionary <string, List <SampleGenomicBin> > commonCNVintervals = null;
if (_commonCnVs != null)
{
commonCNVintervals = CanvasCommon.Utilities.LoadBedFile(_commonCnVs);
CanvasCommon.Utilities.SortAndOverlapCheck(commonCNVintervals, _commonCnVs);
}
var segmentByChr = new Dictionary <string, Segmentation.Segment[]>();
var cts = new CancellationTokenSource();
Parallel.ForEach(
segmentation.First().ScoreByChr.Keys,
new ParallelOptions
{
CancellationToken = cts.Token,
MaxDegreeOfParallelism = Environment.ProcessorCount,
TaskScheduler = TaskScheduler.Default
},
chr =>
{
var breakpoints = new List <int>();
int length = segmentation.First().ScoreByChr[chr].Length;
var startByChr = segmentation.First().StartByChr[chr];
var endByChr = segmentation.First().EndByChr[chr];
var multiSampleCoverage = new List <List <double> >(length);
for (int i = 0; i < length; i++)
{
multiSampleCoverage.Add(segmentation.Select(x => x.ScoreByChr[chr][i]).ToList());
}
if (length > _minSize)
{
var haploidMeans = new List <double>(_nHiddenStates);
var negativeBinomialDistributions = InitializeNegativeBinomialEmission(multiSampleCoverage, _nHiddenStates, haploidMeans);
var hmm = new HiddenMarkovModel(multiSampleCoverage, negativeBinomialDistributions, haploidMeans);
Console.WriteLine($"{DateTime.Now} Launching HMM task for chromosome {chr}");
if (_nSamples <= 3)
{
hmm.FindMaximalLikelihood(multiSampleCoverage);
}
var bestPathViterbi = hmm.BestPathViterbi(multiSampleCoverage, startByChr, haploidMeans);
Console.WriteLine($"{DateTime.Now} Completed HMM task for chromosome {chr}");
breakpoints.Add(0);
for (int i = 1; i < length; i++)
{
if (bestPathViterbi[i] - bestPathViterbi[i - 1] != 0)
{
breakpoints.Add(i);
}
}
if (_commonCnVs != null)
{
if (commonCNVintervals.ContainsKey(chr))
{
var remappedCommonCNVintervals = Segmentation.RemapCommonRegions(commonCNVintervals[chr], startByChr, endByChr);
var oldbreakpoints = breakpoints;
breakpoints = Segmentation.OverlapCommonRegions(oldbreakpoints, remappedCommonCNVintervals);
}
}
var segments = Segmentation.DeriveSegments(breakpoints, length, startByChr, endByChr);
lock (segmentByChr)
{
segmentByChr[chr] = segments;
}
}
});
Console.WriteLine("{0} Completed HMM tasks", DateTime.Now);
Console.WriteLine("{0} Segmentation results complete", DateTime.Now);
return(segmentByChr);
}
public Dictionary <string, SegmentationInput.Segment[]> Run(List <SegmentationInput> segmentation, bool isPerSample)
{
var segmentByChr = new Dictionary <string, SegmentationInput.Segment[]>();
var cts = new CancellationTokenSource();
// Compute whole-genome median and inter-quartile-range-based pseudo-variance for each sample;
// it would be better to exclude regions that are not diploid, and we should really be
// using a different variance for each copy number, but using these values is better than
// using the per-chromosome mean and variance, which have the following problems:
// - chromosomes with a lot of outliers can get a very high variance
// - chromosomes that have a whole-chromosome CNV or a CNV that affects a lot of the chromosome
// can have problematic estimates
var medians = new List <double>();
var pseudoVariances = new List <double>();
foreach (var singleSampleSegmentation in segmentation)
{
var cvgVals = new List <float>();
foreach (var chr in singleSampleSegmentation.CoverageInfo.CoverageByChr.Keys)
{
cvgVals.AddRange(singleSampleSegmentation.CoverageInfo.CoverageByChr[chr].Select(x => (float)x));
}
var quartiles = CanvasCommon.Utilities.Quartiles(cvgVals);
medians.Add(quartiles.Item2);
var iqr = quartiles.Item3 - quartiles.Item1;
pseudoVariances.Add(iqr * iqr);
//Console.WriteLine($"Global estimation of median and pseudovariance: {quartiles.Item2} {iqr * iqr}");
}
Parallel.ForEach(
segmentation.First().CoverageInfo.CoverageByChr.Keys,
new ParallelOptions
{
CancellationToken = cts.Token,
MaxDegreeOfParallelism = Environment.ProcessorCount,
TaskScheduler = TaskScheduler.Default
},
chr =>
{
var breakpoints = new List <int>();
int length = segmentation.First().CoverageInfo.CoverageByChr[chr].Length;
var startByChr = segmentation.First().CoverageInfo.StartByChr[chr];
var endByChr = segmentation.First().CoverageInfo.EndByChr[chr];
var multiSampleCoverage = new List <List <double> >(length);
for (int i = 0; i < length; i++)
{
multiSampleCoverage.Add(segmentation.Select(x => x.CoverageInfo.CoverageByChr[chr][i]).ToList());
}
if (length > _minSize)
{
var haploidMeans = new List <double>(_nHiddenStates);
var negativeBinomialDistributions = isPerSample ?
InitializeNegativeBinomialEmission(multiSampleCoverage, _nHiddenStates, haploidMeans, medians, pseudoVariances)
: InitializeNegativeBinomialEmission(multiSampleCoverage, _nHiddenStates, haploidMeans, null, null);
//for (int j = 0; j < 1; j++)
// for (int i = 0; i < 190; i++)
// {
// Console.WriteLine($"NegBin smp {j} count {i}: {negativeBinomialDistributions[0].Probability(j, i)} {negativeBinomialDistributions[1].Probability(j, i)} {negativeBinomialDistributions[2].Probability(j, i)} {negativeBinomialDistributions[3].Probability(j, i)} {negativeBinomialDistributions[4].Probability(j, i)}");
// }
var hmm = new HiddenMarkovModel(multiSampleCoverage, negativeBinomialDistributions, haploidMeans, isPerSample);
Console.WriteLine($"{DateTime.Now} Launching HMM task for chromosome {chr}");
//if (_nSamples == 1)
// hmm.FindMaximalLikelihood(multiSampleCoverage);
var bestPathViterbi = hmm.BestPathViterbi(multiSampleCoverage, startByChr, haploidMeans);
Console.WriteLine($"{DateTime.Now} Completed HMM task for chromosome {chr}");
breakpoints.Add(0);
for (int i = 1; i < length; i++)
{
if (bestPathViterbi[i] - bestPathViterbi[i - 1] != 0)
{
breakpoints.Add(i);
}
}
var segments = SegmentationInput.DeriveSegments(breakpoints, length, startByChr, endByChr);
lock (segmentByChr)
{
segmentByChr[chr] = segments;
}
}
});
Console.WriteLine("{0} Completed HMM tasks", DateTime.Now);
Console.WriteLine("{0} Segmentation results complete", DateTime.Now);
return(segmentByChr);
}