private static void GetWeightedAverageBinCount(IEnumerable <string> binnedPaths, string mergedBinnedPath,
NexteraManifest manifest = null)
{
int sampleCount = binnedPaths.Count();
if (sampleCount == 1) // copy file
{
if (File.Exists(binnedPaths.First()))
{
if (File.Exists(mergedBinnedPath))
{
File.Delete(mergedBinnedPath);
}
File.Copy(binnedPaths.First(), mergedBinnedPath);
}
}
else // merge normal samples
{
double[] weights = new double[sampleCount];
List <double>[] binCountsBySample = new List <double> [sampleCount];
for (int sampleIndex = 0; sampleIndex < sampleCount; sampleIndex++)
{
string binnedPath = binnedPaths.ElementAt(sampleIndex);
var binCounts = new BinCounts(binnedPath, manifest: manifest);
List <double> counts = binCounts.AllCounts;
// If a manifest is available, get the median of bins overlapping the targeted regions only.
// For small panels, there could be a lot of bins with zero count and the median would be 0 if taken over all the bins, resulting in division by zero.
double median = binCounts.OnTargetMedianBinCount;
weights[sampleIndex] = median > 0 ? 1.0 / median : 0;
binCountsBySample[sampleIndex] = counts;
}
double weightSum = weights.Sum();
for (int i = 0; i < sampleCount; i++)
{
weights[i] /= weightSum;
} // so weights sum to 1
// Computed weighted average of bin counts across samples
using (GzipReader reader = new GzipReader(binnedPaths.First()))
using (GzipWriter writer = new GzipWriter(mergedBinnedPath))
{
string line;
string[] toks;
int lineIdx = 0;
while ((line = reader.ReadLine()) != null)
{
toks = line.Split('\t');
double weightedBinCount = 0;
for (int i = 0; i < sampleCount; i++)
{
weightedBinCount += weights[i] * binCountsBySample[i][lineIdx];
}
toks[3] = String.Format("{0}", weightedBinCount);
writer.WriteLine(String.Join("\t", toks));
lineIdx++;
}
}
}
}
public void Run(IFileLocation outputFile)
{
int bestNormalSampleIndex = 0;
int normalSampleCount = _controlBinnedFiles.Count();
if (normalSampleCount > 1) // find the best normal
{
List <double[]> binCountsByNormalSample = new List <double[]>();
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
var controlBinnedFile = _controlBinnedFiles.ElementAt(normalSampleIndex);
var binCounts = new BinCounts(controlBinnedFile.FullName, manifest: _manifest);
List <double> counts = binCounts.OnTargetCounts;
double median = binCounts.OnTargetMedianBinCount;
// If a manifest is available, get the median of bins overlapping the targeted regions only.
// For small panels, there could be a lot of bins with zero count and the median would be 0 if taken over all the bins, resulting in division by zero.
double weight = median > 0 ? 1.0 / median : 0;
binCountsByNormalSample.Add(counts.Select(cnt => cnt * weight).ToArray());
}
double[] tumorBinCounts;
{
var binCounts = new BinCounts(_sampleBinnedFile.FullName, manifest: _manifest);
List <double> counts = binCounts.OnTargetCounts;
double tumorMedian = binCounts.OnTargetMedianBinCount;
double tumorWeight = tumorMedian > 0 ? 1.0 / tumorMedian : 0;
tumorBinCounts = counts.Select(cnt => cnt * tumorWeight).ToArray();
}
// Find the best normal sample
bestNormalSampleIndex = -1;
double minMeanSquaredLogRatios = double.PositiveInfinity;
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
// Get the sum of squared log ratios
var result = GetMeanSquaredLogRatios(tumorBinCounts, binCountsByNormalSample[normalSampleIndex]);
double meanSquaredLogRatios = result.Item1;
int ignoredBinCount = result.Item2;
// TODO: Skip a (bad) normal sample if too many bins were ignored.
// Donavan's script skips a normal sample if more than 100 log ratios is NA.
// The cut-off is likely panel-dependent.
if (meanSquaredLogRatios < minMeanSquaredLogRatios)
{
minMeanSquaredLogRatios = meanSquaredLogRatios;
bestNormalSampleIndex = normalSampleIndex;
}
}
}
// copy file
var srcBinnedFile = _controlBinnedFiles.ElementAt(bestNormalSampleIndex);
if (outputFile.Exists)
{
outputFile.Delete();
}
srcBinnedFile.CopyTo(outputFile);
}
/// <summary>
/// Pick the best normal control that has the smallest mean squared log-ratios (LR2s).
/// </summary>
/// <param name="tumorBinnedPath"></param>
/// <param name="normalBinnedPaths"></param>
/// <param name="bestBinnedPath"></param>
/// <param name="manifest"></param>
private static void GetBestLR2BinCount(string tumorBinnedPath, IEnumerable<string> normalBinnedPaths, string bestBinnedPath,
NexteraManifest manifest = null)
{
int bestNormalSampleIndex = 0;
int normalSampleCount = normalBinnedPaths.Count();
if (normalSampleCount > 1) // find the best normal
{
List<double[]> binCountsByNormalSample = new List<double[]>();
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
string normalBinnedPath = normalBinnedPaths.ElementAt(normalSampleIndex);
var binCounts = new BinCounts(normalBinnedPath, manifest: manifest);
List<double> counts = binCounts.OnTargetCounts;
double median = binCounts.OnTargetMedianBinCount;
// If a manifest is available, get the median of bins overlapping the targeted regions only.
// For small panels, there could be a lot of bins with zero count and the median would be 0 if taken over all the bins, resulting in division by zero.
double weight = median > 0 ? 1.0 / median : 0;
binCountsByNormalSample.Add(counts.Select(cnt => cnt * weight).ToArray());
}
double[] tumorBinCounts;
{
var binCounts = new BinCounts(tumorBinnedPath, manifest: manifest);
List<double> counts = binCounts.OnTargetCounts;
double tumorMedian = binCounts.OnTargetMedianBinCount;
double tumorWeight = tumorMedian > 0 ? 1.0 / tumorMedian : 0;
tumorBinCounts = counts.Select(cnt => cnt * tumorWeight).ToArray();
}
// Find the best normal sample
bestNormalSampleIndex = -1;
double minMeanSquaredLogRatios = double.PositiveInfinity;
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
// Get the sum of squared log ratios
var result = GetMeanSquaredLogRatios(tumorBinCounts, binCountsByNormalSample[normalSampleIndex]);
double meanSquaredLogRatios = result.Item1;
int ignoredBinCount = result.Item2;
// TODO: Skip a (bad) normal sample if too many bins were ignored.
// Donavan's script skips a normal sample if more than 100 log ratios is NA.
// The cut-off is likely panel-dependent.
if (meanSquaredLogRatios < minMeanSquaredLogRatios)
{
minMeanSquaredLogRatios = meanSquaredLogRatios;
bestNormalSampleIndex = normalSampleIndex;
}
}
}
// copy file
string srcBinnedPath = normalBinnedPaths.ElementAt(bestNormalSampleIndex);
if (File.Exists(srcBinnedPath))
{
if (File.Exists(bestBinnedPath)) { File.Delete(bestBinnedPath); }
File.Copy(srcBinnedPath, bestBinnedPath);
}
}
private static void GetWeightedAverageBinCount(IEnumerable<string> binnedPaths, string mergedBinnedPath,
NexteraManifest manifest = null)
{
int sampleCount = binnedPaths.Count();
if (sampleCount == 1) // copy file
{
if (File.Exists(binnedPaths.First()))
{
if (File.Exists(mergedBinnedPath)) { File.Delete(mergedBinnedPath); }
File.Copy(binnedPaths.First(), mergedBinnedPath);
}
}
else // merge normal samples
{
double[] weights = new double[sampleCount];
List<double>[] binCountsBySample = new List<double>[sampleCount];
for (int sampleIndex = 0; sampleIndex < sampleCount; sampleIndex++)
{
string binnedPath = binnedPaths.ElementAt(sampleIndex);
var binCounts = new BinCounts(binnedPath, manifest: manifest);
List<double> counts = binCounts.AllCounts;
// If a manifest is available, get the median of bins overlapping the targeted regions only.
// For small panels, there could be a lot of bins with zero count and the median would be 0 if taken over all the bins, resulting in division by zero.
double median = binCounts.OnTargetMedianBinCount;
weights[sampleIndex] = median > 0 ? 1.0 / median : 0;
binCountsBySample[sampleIndex] = counts;
}
double weightSum = weights.Sum();
for (int i = 0; i < sampleCount; i++) { weights[i] /= weightSum; } // so weights sum to 1
// Computed weighted average of bin counts across samples
using (GzipReader reader = new GzipReader(binnedPaths.First()))
using (GzipWriter writer = new GzipWriter(mergedBinnedPath))
{
string line;
string[] toks;
int lineIdx = 0;
while ((line = reader.ReadLine()) != null)
{
toks = line.Split('\t');
double weightedBinCount = 0;
for (int i = 0; i < sampleCount; i++) { weightedBinCount += weights[i] * binCountsBySample[i][lineIdx]; }
toks[3] = String.Format("{0}", weightedBinCount);
writer.WriteLine(String.Join("\t", toks));
lineIdx++;
}
}
}
}
public void Run(IFileLocation outputFile)
{
List <SampleGenomicBin> sampleBins = CanvasIO.ReadFromTextFile(_sampleBinnedFile.FullName);
VerifyBinOrder(sampleBins);
// set bin count to 1 if less than 1
foreach (var bin in sampleBins)
{
bin.Count = Math.Max(1, bin.Count);
}
// center the sample
var centeredSampleVector = Enumerable.Zip(sampleBins, _model.Mu, (bin, mu) => (double)bin.Count - mu.Count).ToArray();
// project onto the axes
var projectedSampleVector = CanvasCommon.Utilities.Project(centeredSampleVector, _model.Axes);
// undo centering and set bin count to 1 if less than 1
var referenceVector = Enumerable.Zip(_model.Mu, projectedSampleVector, (bin, count) => Math.Max(1, bin.Count + count));
// write temporary reference count file
var tempReferenceFile = new FileLocation(Path.GetTempFileName());
var tempReferenceBins = Enumerable.Zip(sampleBins, referenceVector,
(bin, count) => new SampleGenomicBin(bin.GenomicBin.Chromosome, bin.Start, bin.Stop, bin.GenomicBin.GC, (float)count));
CanvasIO.WriteToTextFile(tempReferenceFile.FullName, tempReferenceBins);
// calcualte median ratio
var ratios = new BinCounts(_ratioCalculator.Run(_sampleBinnedFile, tempReferenceFile), _manifest);
double medianRatio = ratios.OnTargetMedianBinCount;
// delete temporary reference count file
tempReferenceFile.Delete();
// multiply reference counts by the median ratio
var referenceBins = Enumerable.Zip(sampleBins, referenceVector,
(bin, count) => new SampleGenomicBin(bin.GenomicBin.Chromosome, bin.Start, bin.Stop, bin.GenomicBin.GC, (float)(count * medianRatio)));
// write reference count file
CanvasIO.WriteToTextFile(outputFile.FullName, referenceBins);
}
/// <summary>
/// Pick the best normal control that has the smallest mean squared log-ratios (LR2s).
/// </summary>
/// <param name="tumorBinnedPath"></param>
/// <param name="normalBinnedPaths"></param>
/// <param name="bestBinnedPath"></param>
/// <param name="manifest"></param>
private static void GetBestLR2BinCount(string tumorBinnedPath, IEnumerable <string> normalBinnedPaths, string bestBinnedPath,
NexteraManifest manifest = null)
{
int bestNormalSampleIndex = 0;
int normalSampleCount = normalBinnedPaths.Count();
if (normalSampleCount > 1) // find the best normal
{
List <double[]> binCountsByNormalSample = new List <double[]>();
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
string normalBinnedPath = normalBinnedPaths.ElementAt(normalSampleIndex);
var binCounts = new BinCounts(normalBinnedPath, manifest: manifest);
List <double> counts = binCounts.OnTargetCounts;
double median = binCounts.OnTargetMedianBinCount;
// If a manifest is available, get the median of bins overlapping the targeted regions only.
// For small panels, there could be a lot of bins with zero count and the median would be 0 if taken over all the bins, resulting in division by zero.
double weight = median > 0 ? 1.0 / median : 0;
binCountsByNormalSample.Add(counts.Select(cnt => cnt * weight).ToArray());
}
double[] tumorBinCounts;
{
var binCounts = new BinCounts(tumorBinnedPath, manifest: manifest);
List <double> counts = binCounts.OnTargetCounts;
double tumorMedian = binCounts.OnTargetMedianBinCount;
double tumorWeight = tumorMedian > 0 ? 1.0 / tumorMedian : 0;
tumorBinCounts = counts.Select(cnt => cnt * tumorWeight).ToArray();
}
// Find the best normal sample
bestNormalSampleIndex = -1;
double minMeanSquaredLogRatios = double.PositiveInfinity;
for (int normalSampleIndex = 0; normalSampleIndex < normalSampleCount; normalSampleIndex++)
{
// Get the sum of squared log ratios
var result = GetMeanSquaredLogRatios(tumorBinCounts, binCountsByNormalSample[normalSampleIndex]);
double meanSquaredLogRatios = result.Item1;
int ignoredBinCount = result.Item2;
// TODO: Skip a (bad) normal sample if too many bins were ignored.
// Donavan's script skips a normal sample if more than 100 log ratios is NA.
// The cut-off is likely panel-dependent.
if (meanSquaredLogRatios < minMeanSquaredLogRatios)
{
minMeanSquaredLogRatios = meanSquaredLogRatios;
bestNormalSampleIndex = normalSampleIndex;
}
}
}
// copy file
string srcBinnedPath = normalBinnedPaths.ElementAt(bestNormalSampleIndex);
if (File.Exists(srcBinnedPath))
{
if (File.Exists(bestBinnedPath))
{
File.Delete(bestBinnedPath);
}
File.Copy(srcBinnedPath, bestBinnedPath);
}
}