private static void pushData(
final ExactACset targetSet,
final ExactACset dependentSet,
final int PLsetIndex,
final List<double[]> genotypeLikelihoods) {
final int totalK = targetSet.getACsum();
for (int j = 1; j < targetSet.getLog10Likelihoods().length; j++) {
if (totalK <= 2 * j) { // skip impossible conformations
final double[] gl = genotypeLikelihoods.get(j);
final double conformationValue =
determineCoefficient(PLsetIndex, j, targetSet.getACcounts().getCounts(), totalK)
+ dependentSet.getLog10Likelihoods()[j - 1]
+ gl[PLsetIndex];
targetSet.getLog10Likelihoods()[j] =
MathUtils.approximateLog10SumLog10(
targetSet.getLog10Likelihoods()[j], conformationValue);
}
}
}
private double calculateAlleleCountConformation(
final ExactACset set,
final List<double[]> genotypeLikelihoods,
final int numChr,
final Deque<ExactACset> ACqueue,
final Map<ExactACcounts, ExactACset> indexesToACset,
final double[] log10AlleleFrequencyPriors,
final StateTracker stateTracker) {
// compute the log10Likelihoods
computeLofK(set, genotypeLikelihoods, log10AlleleFrequencyPriors, stateTracker);
final double log10LofK = set.getLog10Likelihoods()[set.getLog10Likelihoods().length - 1];
// can we abort early because the log10Likelihoods are so small?
if (stateTracker.abort(log10LofK, set.getACcounts(), true, false)) {
return log10LofK;
}
// iterate over higher frequencies if possible
final int ACwiggle = numChr - set.getACsum();
if (ACwiggle
== 0) { // all alternate alleles already sum to 2N so we cannot possibly go to higher
// frequencies
return log10LofK;
}
final int numAltAlleles = set.getACcounts().getCounts().length;
// add conformations for the k+1 case
for (int allele = 0; allele < numAltAlleles; allele++) {
final int[] ACcountsClone = set.getACcounts().getCounts().clone();
ACcountsClone[allele]++;
// to get to this conformation, a sample would need to be AB (remember that ref=0)
final int PLindex = GenotypeLikelihoods.calculatePLindex(0, allele + 1);
updateACset(
ACcountsClone, numChr, set, PLindex, ACqueue, indexesToACset, genotypeLikelihoods);
}
// add conformations for the k+2 case if it makes sense; note that the 2 new alleles may be the
// same or different
if (ACwiggle > 1) {
final List<DependentSet> differentAlleles = new ArrayList<>(numAltAlleles * numAltAlleles);
final List<DependentSet> sameAlleles = new ArrayList<>(numAltAlleles);
for (int allele_i = 0; allele_i < numAltAlleles; allele_i++) {
for (int allele_j = allele_i; allele_j < numAltAlleles; allele_j++) {
final int[] ACcountsClone = set.getACcounts().getCounts().clone();
ACcountsClone[allele_i]++;
ACcountsClone[allele_j]++;
// to get to this conformation, a sample would need to be BB or BC (remember that ref=0,
// so add one to the index)
final int PLindex = GenotypeLikelihoods.calculatePLindex(allele_i + 1, allele_j + 1);
if (allele_i == allele_j) {
sameAlleles.add(new DependentSet(ACcountsClone, PLindex));
} else {
differentAlleles.add(new DependentSet(ACcountsClone, PLindex));
}
}
}
// IMPORTANT: we must first add the cases where the 2 new alleles are different so that the
// queue maintains its ordering
for (final DependentSet dependent : differentAlleles) {
updateACset(
dependent.ACcounts,
numChr,
set,
dependent.PLindex,
ACqueue,
indexesToACset,
genotypeLikelihoods);
}
for (final DependentSet dependent : sameAlleles) {
updateACset(
dependent.ACcounts,
numChr,
set,
dependent.PLindex,
ACqueue,
indexesToACset,
genotypeLikelihoods);
}
}
return log10LofK;
}
private static void computeLofK(
final ExactACset set,
final List<double[]> genotypeLikelihoods,
final double[] log10AlleleFrequencyPriors,
final StateTracker stateTracker) {
set.getLog10Likelihoods()[0] = 0.0; // the zero case
final int totalK = set.getACsum();
// special case for k = 0 over all k
if (totalK == 0) {
for (int j = 1; j < set.getLog10Likelihoods().length; j++) {
set.getLog10Likelihoods()[j] =
set.getLog10Likelihoods()[j - 1] + genotypeLikelihoods.get(j)[HOM_REF_INDEX];
}
final double log10Lof0 = set.getLog10Likelihoods()[set.getLog10Likelihoods().length - 1];
stateTracker.setLog10LikelihoodOfAFzero(log10Lof0);
stateTracker.setLog10PosteriorOfAFzero(log10Lof0 + log10AlleleFrequencyPriors[0]);
return;
}
// if we got here, then k > 0 for at least one k.
// the non-AA possible conformations were already dealt with by pushes from dependent sets;
// now deal with the AA case (which depends on previous cells in this column) and then update
// the L(j,k) value
for (int j = 1; j < set.getLog10Likelihoods().length; j++) {
if (totalK < 2 * j - 1) {
final double[] gl = genotypeLikelihoods.get(j);
final double conformationValue =
MathUtils.log10(2 * j - totalK)
+ MathUtils.log10(2 * j - totalK - 1)
+ set.getLog10Likelihoods()[j - 1]
+ gl[HOM_REF_INDEX];
set.getLog10Likelihoods()[j] =
MathUtils.approximateLog10SumLog10(set.getLog10Likelihoods()[j], conformationValue);
}
final double logDenominator = MathUtils.log10(2 * j) + MathUtils.log10(2 * j - 1);
set.getLog10Likelihoods()[j] = set.getLog10Likelihoods()[j] - logDenominator;
}
double log10LofK = set.getLog10Likelihoods()[set.getLog10Likelihoods().length - 1];
// update the MLE if necessary
stateTracker.updateMLEifNeeded(log10LofK, set.getACcounts().getCounts());
// apply the priors over each alternate allele
for (final int ACcount : set.getACcounts().getCounts()) {
if (ACcount > 0) {
log10LofK += log10AlleleFrequencyPriors[ACcount];
}
}
stateTracker.updateMAPifNeeded(log10LofK, set.getACcounts().getCounts());
}