bool IDEvaluationBase::getPoints(std::vector<PeptideIdentification>& peptides /* cannot be const, to avoid copy */,
const std::vector<double>& q_value_thresholds, MSSpectrum<>& points)
{
points.clear(true);
FalseDiscoveryRate fdr;
fdr.setParameters(param_.copy("fdr:", true));
try
{
fdr.apply(peptides); // computes a q-value (if its params are correct)
}
catch (Exception::MissingInformation)
{
LOG_FATAL_ERROR << "Tool failed due to missing information (see above)." << std::endl;
return false;
}
// get list of q-values and sort them
std::vector<double> q_values;
q_values.reserve(peptides.size());
for (vector<PeptideIdentification>::iterator it = peptides.begin(); it != peptides.end(); ++it)
{
it->assignRanks();
if (it->getHits().size() > 0)
q_values.push_back(it->getHits()[0].getScore());
}
std::sort(q_values.begin(), q_values.end());
for (Size i = 0; i < q_value_thresholds.size(); ++i)
{
// get position in sorted q-values where cutoff is reached
std::vector<double>::iterator pos = std::upper_bound(q_values.begin(), q_values.end(), q_value_thresholds[i]);
Peak1D p;
p.setMZ(q_value_thresholds[i] * 100);
p.setIntensity(std::distance(q_values.begin(), pos));
points.push_back(p);
}
return true;
}
void PeakPickerHiRes::pick(const MSSpectrum& input, MSSpectrum& output, std::vector<PeakBoundary>& boundaries, bool check_spacings) const
{
// copy meta data of the input spectrum
output.clear(true);
output.SpectrumSettings::operator=(input);
output.MetaInfoInterface::operator=(input);
output.setRT(input.getRT());
output.setMSLevel(input.getMSLevel());
output.setName(input.getName());
output.setType(SpectrumSettings::CENTROID);
if (report_FWHM_)
{
output.getFloatDataArrays().resize(1);
output.getFloatDataArrays()[0].setName( report_FWHM_as_ppm_ ? "FWHM_ppm" : "FWHM");
}
// don't pick a spectrum with less than 5 data points
if (input.size() < 5) return;
// if both spacing constraints are disabled, don't check spacings at all:
if ((spacing_difference_ == std::numeric_limits<double>::infinity()) &&
(spacing_difference_gap_ == std::numeric_limits<double>::infinity()))
{
check_spacings = false;
}
// signal-to-noise estimation
SignalToNoiseEstimatorMedian<MSSpectrum > snt;
snt.setParameters(param_.copy("SignalToNoise:", true));
if (signal_to_noise_ > 0.0)
{
snt.init(input);
}
// find local maxima in profile data
for (Size i = 2; i < input.size() - 2; ++i)
{
double central_peak_mz = input[i].getMZ(), central_peak_int = input[i].getIntensity();
double left_neighbor_mz = input[i - 1].getMZ(), left_neighbor_int = input[i - 1].getIntensity();
double right_neighbor_mz = input[i + 1].getMZ(), right_neighbor_int = input[i + 1].getIntensity();
// do not interpolate when the left or right support is a zero-data-point
if (std::fabs(left_neighbor_int) < std::numeric_limits<double>::epsilon()) continue;
if (std::fabs(right_neighbor_int) < std::numeric_limits<double>::epsilon()) continue;
// MZ spacing sanity checks
double left_to_central = 0.0, central_to_right = 0.0, min_spacing = 0.0;
if (check_spacings)
{
left_to_central = central_peak_mz - left_neighbor_mz;
central_to_right = right_neighbor_mz - central_peak_mz;
min_spacing = (left_to_central < central_to_right) ? left_to_central : central_to_right;
}
double act_snt = 0.0, act_snt_l1 = 0.0, act_snt_r1 = 0.0;
if (signal_to_noise_ > 0.0)
{
act_snt = snt.getSignalToNoise(input[i]);
act_snt_l1 = snt.getSignalToNoise(input[i - 1]);
act_snt_r1 = snt.getSignalToNoise(input[i + 1]);
}
// look for peak cores meeting MZ and intensity/SNT criteria
if ((central_peak_int > left_neighbor_int) &&
(central_peak_int > right_neighbor_int) &&
(act_snt >= signal_to_noise_) &&
(act_snt_l1 >= signal_to_noise_) &&
(act_snt_r1 >= signal_to_noise_) &&
(!check_spacings ||
((left_to_central < spacing_difference_ * min_spacing) &&
(central_to_right < spacing_difference_ * min_spacing))))
{
// special case: if a peak core is surrounded by more intense
// satellite peaks (indicates oscillation rather than
// real peaks) -> remove
double act_snt_l2 = 0.0, act_snt_r2 = 0.0;
if (signal_to_noise_ > 0.0)
{
act_snt_l2 = snt.getSignalToNoise(input[i - 2]);
act_snt_r2 = snt.getSignalToNoise(input[i + 2]);
}
// checking signal-to-noise?
if ((i > 1) &&
(i + 2 < input.size()) &&
(left_neighbor_int < input[i - 2].getIntensity()) &&
(right_neighbor_int < input[i + 2].getIntensity()) &&
(act_snt_l2 >= signal_to_noise_) &&
(act_snt_r2 >= signal_to_noise_) &&
(!check_spacings ||
((left_neighbor_mz - input[i - 2].getMZ() < spacing_difference_ * min_spacing) &&
(input[i + 2].getMZ() - right_neighbor_mz < spacing_difference_ * min_spacing))))
{
++i;
continue;
}
std::map<double, double> peak_raw_data;
peak_raw_data[central_peak_mz] = central_peak_int;
peak_raw_data[left_neighbor_mz] = left_neighbor_int;
peak_raw_data[right_neighbor_mz] = right_neighbor_int;
// peak core found, now extend it
// to the left
Size k = 2;
bool previous_zero_left(false); // no need to extend peak if previous intensity was zero
Size missing_left(0);
Size left_boundary(i - 1); // index of the left boundary for the spline interpolation
while ((k <= i) && // prevent underflow
(i - k + 1 > 0) &&
!previous_zero_left &&
(missing_left <= missing_) &&
(input[i - k].getIntensity() <= peak_raw_data.begin()->second) &&
(!check_spacings ||
(peak_raw_data.begin()->first - input[i - k].getMZ() < spacing_difference_gap_ * min_spacing)))
{
double act_snt_lk = 0.0;
if (signal_to_noise_ > 0.0)
{
act_snt_lk = snt.getSignalToNoise(input[i - k]);
}
if ((act_snt_lk >= signal_to_noise_) &&
(!check_spacings ||
(peak_raw_data.begin()->first - input[i - k].getMZ() < spacing_difference_ * min_spacing)))
{
peak_raw_data[input[i - k].getMZ()] = input[i - k].getIntensity();
}
else
{
++missing_left;
if (missing_left <= missing_)
{
peak_raw_data[input[i - k].getMZ()] = input[i - k].getIntensity();
}
}
previous_zero_left = (input[i - k].getIntensity() == 0);
left_boundary = i - k;
++k;
}
// to the right
k = 2;
bool previous_zero_right(false); // no need to extend peak if previous intensity was zero
Size missing_right(0);
Size right_boundary(i+1); // index of the right boundary for the spline interpolation
while ((i + k < input.size()) &&
!previous_zero_right &&
(missing_right <= missing_) &&
(input[i + k].getIntensity() <= peak_raw_data.rbegin()->second) &&
(!check_spacings ||
(input[i + k].getMZ() - peak_raw_data.rbegin()->first < spacing_difference_gap_ * min_spacing)))
{
double act_snt_rk = 0.0;
if (signal_to_noise_ > 0.0)
{
act_snt_rk = snt.getSignalToNoise(input[i + k]);
}
if ((act_snt_rk >= signal_to_noise_) &&
(!check_spacings ||
(input[i + k].getMZ() - peak_raw_data.rbegin()->first < spacing_difference_ * min_spacing)))
{
peak_raw_data[input[i + k].getMZ()] = input[i + k].getIntensity();
}
else
{
++missing_right;
if (missing_right <= missing_)
{
peak_raw_data[input[i + k].getMZ()] = input[i + k].getIntensity();
}
}
previous_zero_right = (input[i + k].getIntensity() == 0);
right_boundary = i + k;
++k;
}
// skip if the minimal number of 3 points for fitting is not reached
if (peak_raw_data.size() < 3) continue;
CubicSpline2d peak_spline (peak_raw_data);
// calculate maximum by evaluating the spline's 1st derivative
// (bisection method)
double max_peak_mz = central_peak_mz;
double max_peak_int = central_peak_int;
double threshold = 1e-6;
OpenMS::Math::spline_bisection(peak_spline, left_neighbor_mz, right_neighbor_mz, max_peak_mz, max_peak_int, threshold);
//
// compute FWHM
//
if (report_FWHM_)
{
double fwhm_int = max_peak_int / 2.0;
threshold = 0.01 * fwhm_int;
double mz_mid, int_mid;
// left:
double mz_left = peak_raw_data.begin()->first;
double mz_center = max_peak_mz;
if (peak_spline.eval(mz_left) > fwhm_int)
{ // the spline ends before half max is reached -- take the leftmost point (probably an underestimation)
mz_mid = mz_left;
}
else
{
do
{
mz_mid = mz_left / 2 + mz_center / 2;
int_mid = peak_spline.eval(mz_mid);
if (int_mid < fwhm_int)
{
mz_left = mz_mid;
}
else
{
mz_center = mz_mid;
}
} while (fabs(int_mid - fwhm_int) > threshold);
}
const double fwhm_left_mz = mz_mid;
// right ...
double mz_right = peak_raw_data.rbegin()->first;
mz_center = max_peak_mz;
if (peak_spline.eval(mz_right) > fwhm_int)
{ // the spline ends before half max is reached -- take the rightmost point (probably an underestimation)
mz_mid = mz_right;
}
else
{
do
{
mz_mid = mz_right / 2 + mz_center / 2;
int_mid = peak_spline.eval(mz_mid);
if (int_mid < fwhm_int)
{
mz_right = mz_mid;
}
else
{
mz_center = mz_mid;
}
} while (fabs(int_mid - fwhm_int) > threshold);
}
const double fwhm_right_mz = mz_mid;
const double fwhm_absolute = fwhm_right_mz - fwhm_left_mz;
output.getFloatDataArrays()[0].push_back( report_FWHM_as_ppm_ ? fwhm_absolute / max_peak_mz * 1e6 : fwhm_absolute);
} // FWHM
// save picked peak into output spectrum
Peak1D peak;
PeakBoundary peak_boundary;
peak.setMZ(max_peak_mz);
peak.setIntensity(max_peak_int);
peak_boundary.mz_min = input[left_boundary].getMZ();
peak_boundary.mz_max = input[right_boundary].getMZ();
output.push_back(peak);
boundaries.push_back(peak_boundary);
// jump over profile data points that have been considered already
i = i + k - 1;
}
}
return;
}