EmgScoring.h

Go to the documentation of this file.

// Copyright (c) 2002-present, OpenMS Inc. -- EKU Tuebingen, ETH Zurich, and FU Berlin
// SPDX-License-Identifier: BSD-3-Clause
//
// --------------------------------------------------------------------------
// $Maintainer: Hannes Roest $
// $Authors: Hannes Roest $
// --------------------------------------------------------------------------
 
#pragma once
 
#include <OpenMS/FEATUREFINDER/EmgFitter1D.h>
#include <OpenMS/FEATUREFINDER/EmgModel.h>
#include <OpenMS/PROCESSING/SMOOTHING/GaussFilter.h>
 
#include <OpenMS/KERNEL/MRMFeature.h>
#include <OpenMS/KERNEL/MRMTransitionGroup.h>
#include <OpenMS/KERNEL/MSSpectrum.h>
 
#include <OpenMS/KERNEL/StandardTypes.h>
 
#include <vector>
#include <cmath> // for isnan
 
namespace OpenMS
{
 
  class EmgScoring
  {
 
  public :
 
    EmgScoring() = default;
 
    ~EmgScoring() = default;
 
    void setFitterParam(const Param& param)
    {
      fitter_emg1D_params_ = param;
    }
 
    Param getDefaults()
    {
      return EmgFitter1D().getDefaults();
    }
 
    template<typename SpectrumType, class TransitionT>
    double calcElutionFitScore(MRMFeature & mrmfeature, MRMTransitionGroup<SpectrumType, TransitionT> & transition_group) const
    {
      double avg_score = 0;
      bool smooth_data = false;
 
      for (Size k = 0; k < transition_group.size(); k++)
      {
        // get the id, then find the corresponding transition and features within this peakgroup
        String native_id = transition_group.getChromatograms()[k].getNativeID();
        Feature f = mrmfeature.getFeature(native_id);
        OPENMS_PRECONDITION(f.getConvexHulls().size() == 1, "Convex hulls need to have exactly one hull point structure");
 
        //TODO think about penalizing aborted fits even more. Currently -1 is just the "lowest" pearson correlation to
        // a fit that you can have.
        double fscore = elutionModelFit(f.getConvexHulls()[0].getHullPoints(), smooth_data);
        avg_score += fscore;
      }
 
      avg_score /= transition_group.size();
      return avg_score;
    }
 
    // Fxn from FeatureFinderAlgorithmMRM
    // TODO: check whether we can leave out some of the steps here, e.g. gaussian smoothing
    double elutionModelFit(const ConvexHull2D::PointArrayType& current_section, bool smooth_data) const
    {
      // We need at least 2 datapoints in order to create a fit
      if (current_section.size() < 2)
      {
        return -1;
      }
 
      // local PeakType is a small hack since here we *need* data of type
      // Peak1D, otherwise our fitter will not accept it.
      typedef Peak1D LocalPeakType;
 
      // -- cut line 301 of FeatureFinderAlgorithmMRM
      std::vector<LocalPeakType> data_to_fit;
      prepareFit_(current_section, data_to_fit, smooth_data);
      std::unique_ptr<InterpolationModel> model_rt;
      double quality = fitRT_(data_to_fit, model_rt);
      // cut line 354 of FeatureFinderAlgorithmMRM
 
      return quality;
    }
 
  protected:
    template<class LocalPeakType>
    double fitRT_(std::vector<LocalPeakType>& rt_input_data, std::unique_ptr<InterpolationModel>& model) const
    {
      EmgFitter1D fitter_emg1D;
      fitter_emg1D.setParameters(fitter_emg1D_params_);
      // Construct model for rt
      // NaN is checked in fit1d: if (std::isnan(quality)) quality = -1.0;
      return fitter_emg1D.fit1d(rt_input_data, model);
    }
 
    // Fxn from FeatureFinderAlgorithmMRM
    // TODO: check whether we can leave out some of the steps here, e.g. gaussian smoothing
    template<class LocalPeakType>
    void prepareFit_(const ConvexHull2D::PointArrayType & current_section, std::vector<LocalPeakType> & data_to_fit, bool smooth_data) const
    {
      // typedef Peak1D LocalPeakType;
      PeakSpectrum filter_spec;
      // first smooth the data to prevent outliers from destroying the fit
      for (const auto& pa : current_section)
      {
        LocalPeakType p;
        using IntensityType = typename LocalPeakType::IntensityType;
        p.setMZ(pa.getX());
        p.setIntensity(IntensityType(pa.getY()));
        filter_spec.push_back(p);
      }
 
      // add two peaks at the beginning and at the end for better fit
      // therefore calculate average distance first
      std::vector<double> distances;
      for (Size j = 1; j < filter_spec.size(); ++j)
      {
        distances.push_back(filter_spec[j].getMZ() - filter_spec[j - 1].getMZ());
      }
      double dist_average = std::accumulate(distances.begin(), distances.end(), 0.0) / (double) distances.size();
 
      // append peaks
      Peak1D new_peak;
      new_peak.setIntensity(0);
      new_peak.setMZ(filter_spec.back().getMZ() + dist_average);
      filter_spec.push_back(new_peak);
      new_peak.setMZ(filter_spec.back().getMZ() + dist_average);
      filter_spec.push_back(new_peak);
      new_peak.setMZ(filter_spec.back().getMZ() + dist_average);
      filter_spec.push_back(new_peak);
 
      // prepend peaks
      new_peak.setMZ(filter_spec.front().getMZ() - dist_average);
      filter_spec.insert(filter_spec.begin(), new_peak);
      new_peak.setMZ(filter_spec.front().getMZ() - dist_average);
      filter_spec.insert(filter_spec.begin(), new_peak);
      new_peak.setMZ(filter_spec.front().getMZ() - dist_average);
      filter_spec.insert(filter_spec.begin(), new_peak);
 
      // To get an estimate of the peak quality, we probably should not smooth
      // and/or transform the data.
      if (smooth_data)
      {
        GaussFilter filter;
        Param filter_param(filter.getParameters());
        filter.setParameters(filter_param);
        filter_param.setValue("gaussian_width", 4 * dist_average);
        filter.setParameters(filter_param);
        filter.filter(filter_spec);
      }
 
      // transform the data for fitting and fit RT profile
      for (Size j = 0; j != filter_spec.size(); ++j)
      {
        LocalPeakType p;
        p.setPosition(filter_spec[j].getMZ());
        p.setIntensity(filter_spec[j].getIntensity());
        data_to_fit.push_back(p);
      }
    }
 
    Param fitter_emg1D_params_;
  };
 
}