static void writeEstimate(ostream& out,
const ContigNode& id0, const ContigNode& id1,
unsigned len0, unsigned len1,
const Pairs& pairs, const PMF& pmf)
{
if (pairs.size() < opt::npairs)
return;
DistanceEst est;
est.distance = estimateDistance(len0, len1,
pairs, pmf, est.numPairs);
est.stdDev = pmf.getSampleStdDev(est.numPairs);
std::pair<ContigNode, ContigNode> e(id0, id1 ^ id0.sense());
if (est.numPairs >= opt::npairs) {
if (opt::format == DOT) {
#pragma omp critical(out)
out << get(g_contigNames, e) << " [" << est << "]\n";
} else
out << ' ' << get(g_contigNames, id1) << ',' << est;
} else if (opt::verbose > 1) {
#pragma omp critical(cerr)
cerr << "warning: " << get(g_contigNames, e)
<< " [d=" << est.distance << "] "
<< est.numPairs << " of " << pairs.size()
<< " pairs fit the expected distribution\n";
}
}
// term of non-HW HMatrix
inline double term_ij(
double Bp,
double Bq,
const PMF<Allele> &p,
const PMF<Allele> &q,
const BackFreq &hback,
Allele const &i,
Allele const &j)
{
// NB in this term (i, j) is ordered
double ret = p.val(i) * q.val(j);
if (Bq>0) ret += Bq * p.val(i) * hback(j, i);
if (Bp>0) ret += Bp * q.val(j) * hback(i, j);
if (Bp*Bq>0) ret += Bp * Bq * hback.pOrdered(make_pair(i,j));
return ret;
}
/** Estimate the distance between two contigs using the difference of
* the population mean and the sample mean.
* @param numPairs [out] the number of pairs that agree with the
* expected distribution
* @return the estimated distance
*/
static int estimateDistanceUsingMean(
const std::vector<int>& samples, const PMF& pmf,
unsigned& numPairs)
{
Histogram h(samples.begin(), samples.end());
int d = (int)round(pmf.mean() - h.mean());
// Count the number of samples that agree with the distribution.
unsigned n = 0;
for (Histogram::const_iterator it = h.begin();
it != h.end(); ++it)
if (pmf[it->first + d] > pmf.minProbability())
n += it->second;
numPairs = n;
return d;
}
// This gives non-HW treatment of the 'F' term
HMatrix
makeHMatrixNHW(
const PMF<Allele> &p,
const PMF<Allele> &q,
const BackFreq &hback,
float delta,
bool sparse)
{
// How much background to add in?
// If the input HMatrix is not normalized then we make up the difference
// with background. Otherwise we add in delta.
double Bp = std::max((double)delta, 1 - p.sum()); // background for p
double Bq = std::max((double)delta, 1 - q.sum()); // background for q
// apply this formula:
// H(ij) = p(i)q(j) + Bq p(i)b(j|i) + Bp (q(j)b(i|j) + Bp Bq Bij
HMatrix ret;
// loop over all elements in the background. This gives us the upper
// triangular terms only. We need to sum over all terms.
HMatrix &href = (HMatrix&)hback; // to use base class member
PMF< std::pair<Allele, Allele> >::iterator it;
for (it = href.m_pmf.begin(); it != href.m_pmf.end(); ++it)
{
Allele i = it->first.first;
Allele j = it->first.second;
double h_ij = term_ij(Bp, Bq, p, q, hback, i, j); // upper-triangular term
if (i != j)
{
h_ij += term_ij(Bp, Bq, p, q, hback, j, i); // lower-triangular term
}
if (!sparse || h_ij > 0) ret.set(i, j, h_ij);
}
ret.normalize(); // just in case
return ret;
}
// check for alleles not in population database
bool
AlleleSet::checkBackground(PMF<Allele> const &background) const
{
bool ret = true;
std::vector< PMF<Allele> >::const_iterator ip;
for(ip = m_pmfs.begin(); ip != m_pmfs.end(); ++ip)
{
PMF<Allele>::const_iterator ia;
for(ia = ip->begin(); ia != ip->end(); ++ia)
{
if (background.find(ia->first) == background.end())
{
// Allele not in database.
warn << startl << "allele not in population database: " << ia->first.string() << std::endl;
ret = false;
}
}
}
return ret;
}
/** Compute the log likelihood that these samples came from the
* specified distribution shifted by the parameter theta.
* @param theta the parameter of the PMF, f_theta(x)
* @param samples the samples
* @param pmf the probability mass function
* @return the log likelihood
*/
static pair<double, unsigned>
computeLikelihood(int theta, const Histogram& samples, const PMF& pmf)
{
double likelihood = 0;
unsigned nsamples = 0;
for (Histogram::const_iterator it = samples.begin();
it != samples.end(); ++it) {
double p = pmf[it->first + theta];
unsigned n = it->second;
likelihood += n * log(p);
if (p > pmf.minProbability())
nsamples += n;
}
return make_pair(likelihood, nsamples);
}
