public ContinuousFrequencySNPGenotype(BasePairFrequencies frequencies, int[] originalReadCounts, PileUp pileup = null) :
this(GenotypeCallResult.GenotypeCalled, pileup)
{
Frequencies = frequencies;
OriginalBasePairCounts = originalReadCounts;
var freq = frequencies.Frequencies;
this.indexMax = 0;
var max = freq [0];
for (int i = 1; i < freq.Length; i++)
{
if (freq [i] > max)
{
max = freq [i];
indexMax = i;
}
}
}
/// <summary>
/// Calculates the conditional probability that a base comes from each of A, C, G, T
/// based on the observed read and quality and current parameter settings.
/// </summary>
/// <returns>The gonditional probabilities.</returns>
/// <param name="freqs">Freqs.</param>
/// <param name="data">Data.</param>
/// <param name="bp">Bp.</param>
private static double updateConditionalProbabilities(BasePairFrequencies freqs, double[] data, BaseAndQuality bp)
{
//Calculate conditional probability that the base comes from the observed base.
double probRight = BaseQualityUtils.GetCorrectProbability(bp.PhredScore);
double probWrong = .333333333 * BaseQualityUtils.GetErrorProbability(bp.PhredScore);
double totProb = 0.0;
for (int i = 0; i < data.Length; i++)
{
var freq = freqs.Frequencies[i];
if (freq != 0.0)
{
double prob = freq * (i == bp.Base ? probRight : probWrong);
totProb += prob;
data[i] = prob;
}
}
for (int i = 0; i < data.Length; i++)
{
data[i] /= totProb;
}
return(Math.Log(totProb));
}
private static ContinuousFrequencySNPGenotype callGenotype(PileUp pu)
{
//If it looks like a deletion, skip it
if (pileupHasTooManyIndels(pu))
{
return(new ContinuousFrequencySNPGenotype(GenotypeCallResult.TooManyGaps, pu));
}
// Otherwise, drop gaps, ambiguous bases and low scoring reads.
var filteredBases = pu.Bases.Where(z => z.Base != BaseAndQuality.N_BASE_INDEX &&
z.Base != BaseAndQuality.GAP_BASE_INDEX && z.PhredScore > 17).ToArray();
if (filteredBases.Length == 0)
{
return(new ContinuousFrequencySNPGenotype(GenotypeCallResult.NoData));
}
// initialize the continuous frequency based on counts of bases.
var base_pair_counts = new int[BasePairFrequencies.NUM_BASES];
foreach (var bp in filteredBases)
{
base_pair_counts [bp.Base]++;
}
var freqs = base_pair_counts.Select(x => x / (double)filteredBases.Length).ToArray();
var theta = new BasePairFrequencies(freqs);
//if only one base has data or if
//if we are not doing EM optimization, we are done.
if (base_pair_counts.Count(x => x > 0) > 1 && DO_EM_ESTIMATION)
{
//first make an NumReads * Num_Bases Matrix
double[][] conditionalProbs = new double[filteredBases.Length][];
for (int i = 0; i < filteredBases.Length; i++)
{
conditionalProbs [i] = new double[BasePairFrequencies.NUM_BASES];
}
double likDif = double.MaxValue;
double last_lik = double.MinValue;
while (likDif > 1e-3)
{
double lik = 0;
for (int i = 0; i < conditionalProbs.Length; i++)
{
lik += updateConditionalProbabilities(theta, conditionalProbs [i], filteredBases [i]);
}
likDif = lik - last_lik;
last_lik = lik;
//now update thetas by summing the conditional probability of each value
Array.Clear(theta.Frequencies, 0, BasePairFrequencies.NUM_BASES);
for (int i = 0; i < conditionalProbs.Length; i++)
{
var cur = conditionalProbs [i];
for (int j = 0; j < cur.Length; j++)
{
theta.Frequencies [j] += cur [j];
}
}
for (int j = 0; j < theta.Frequencies.Length; j++)
{
theta.Frequencies [j] /= (double)conditionalProbs.Length;
}
}
}
return(new ContinuousFrequencySNPGenotype(theta, base_pair_counts, pu));
}
