/// Incomlete beta functions.
double
ibeta(double a, double b, double x)
{
gsl_sf_result result;
int stat = gsl_sf_beta_inc_e(a, b, x, &result);
if (stat != GSL_SUCCESS)
{
std::ostringstream msg("Error in ibeta:");
msg << " a=" << a << " b=" << b << " x=" << x;
throw std::runtime_error(msg.str());
}
else
return result.val;
}
/* To adjust for local mappable K-mer density, just reduce the densityWindowSize
(from 50,000 etc) to however many sites in that window are uniquely mappable
by K-mers. But that throws off the probZ calc in a bad way.
I use adjustedNumSitesInCluster as an estimate of how many sites I would observe
in the density window if all the sites were actually mappable.
*/
double calculateZScore( int basesSpannedByCluster, int numSitesInCluster,
int densityWindowSize, int numSitesInDensityWindow, int numMappableSites )
{
int fewEnoughSites = 2; //densityWinSmall; //densityWindowSize / 2;
// Now using exact counts from the interval, so we shouldn't need this anymore:
if (numMappableSites < fewEnoughSites)
{
// let's put a limit on the adjustment
std::cerr << "Warning: only " << numMappableSites
<< " of " << densityWindowSize
<< " are mappable, increasing to "
<< fewEnoughSites << "..." << std::endl;
numMappableSites = fewEnoughSites;
}
// Adjust probZ using mappable sites.
double probZ = ((double)basesSpannedByCluster) / (double)numMappableSites;
double meanZ = numSitesInDensityWindow * probZ;
double sdZ = std::sqrt(numSitesInDensityWindow * probZ * (1-probZ));
double zScore = (numSitesInCluster - meanZ) / sdZ;
// p-value = P(X >= k), where k = number tags in the cluster
// = 1 - P(X < k) = 1 - P(X <= k - 1). For the binomial distribution, the cdf is given by
// the normalized (or regularized) incomplete beta function, I_x(a,b) (see wikipedia), and
// P(X <= k-1) = I_(1-p)(n - (k-1), k), where p = probability, n = number of tags in the
// density window.
// Number mappable bases genome-wide, from ~rthurman/proj/dhs-peaks/results/fdr/fdr.R
if ( !useGenomeDensWin )
return zScore;
else
{
double probZgw = ((double)basesSpannedByCluster) / mpblGenomeSize;
double meanZgw = backgroundTotalTagCount * probZgw;
double sdZgw = std::sqrt(backgroundTotalTagCount * probZgw * (1-probZgw));
double zScoregw = (numSitesInCluster - meanZgw) / sdZgw;
//std::printf("zScoregw = %f, zScore = %f\n", zScoregw, zScore);
gsl_sf_result beta_result;
double pVal, pValgw;
pVal = 0;
pValgw = 0;
int status = gsl_sf_beta_inc_e(numSitesInCluster, numSitesInDensityWindow - numSitesInCluster + 1, probZ, &beta_result);
if(status == 15){
//std::printf("Warning: underflow in pval computation, setting pVal = 0.\n");
pVal = 0.0;
}
else if(status){
//std::printf("gsl_sf_beta_inc_e error: status = %d, error = %s, setting pVal = 1.\n", status, gsl_strerror(status));
pVal = 1.0;
}
else
{
pVal = beta_result.val;
}
status = gsl_sf_beta_inc_e(numSitesInCluster, backgroundTotalTagCount - numSitesInCluster + 1, probZgw, &beta_result);
if(status == 15){
//std::printf("Warning: underflow in pval computation, setting pValgw = 0.\n");
pValgw = 0.0;
}
else if(status){
//std::printf("gsl_sf_beta_inc_e error: status = %d, error = %s, setting pValgw = 1.\n", status, gsl_strerror(status));
pValgw = 1.0;
}
else
{
pValgw = beta_result.val;
}
//std::printf("zScoregw = %f, zScore = %f, pVal = %g, pValgw = %g\n", zScoregw, zScore, pVal, pValgw);
if (zScoregw < zScore){
//std::printf("Genome-wide density used.\n");
genomeDensZ += zScoregw;
numGenomeDens++;
return zScoregw;
}else{
numLocalDens++;
localDensZ += zScore;
return zScore;
}
}
}
double gsl_sf_beta_inc(const double a, const double b, const double x)
{
EVAL_RESULT(gsl_sf_beta_inc_e(a, b, x, &result));
}