Ejemplo n.º 1
1
  public void Simulation(
      Parameter1 p,
      ArrayList qtl,
      double[] env,
      double[][] w,
      double[][] m,
      ArrayList sm,
      boolean isSelectMarker) {
    Param1 = p;
    QTL = qtl;
    map = new double[m.length][];
    for (int i = 0; i < map.length; i++) {
      map[i] = new double[m[i].length];
      System.arraycopy(m[i], 0, map[i], 0, map[i].length);
    }
    weight = new double[w.length][];
    for (int i = 0; i < w.length; i++) {
      weight[i] = new double[w[i].length];
      System.arraycopy(w[i], 0, weight[i], 0, weight[i].length);
    }
    for (int i = 0; i < map.length; i++) {
      Num_interval += map[i].length == 1 ? map[i].length : map[i].length - 1;
    }
    if (sm != null) {
      selectedMarker = sm;
    } else {
      if (isSelectMarker) {
        selectMarker();
      }
    }
    correctLogBon();
    makeFullMap(QTL, map);
    setup(QTL, map, Param1);
    SimulationResults = new ArrayList();
    for (int i_rep = 0; i_rep < Param1.rep; i_rep++) {
      if (Param1.pt.compareTo("F2") == 0) {
        ap =
            new F2Population(
                Param1.populationSize,
                Param1.pheNum.length,
                Param1.pt,
                distance,
                Param1.seed + i_rep,
                Param1.mu,
                env,
                Param1.sd,
                QTL,
                Param1.mf);
      } else if (Param1.pt.compareTo("DH") == 0) {
        ap =
            new DHPopulation(
                Param1.populationSize,
                Param1.pheNum.length,
                Param1.pt,
                distance,
                Param1.seed + i_rep,
                Param1.mu,
                env,
                Param1.sd,
                QTL,
                Param1.mf);
      } else {
        ap =
            new BackCrossPopulation(
                Param1.populationSize,
                Param1.pheNum.length,
                Param1.pt,
                distance,
                Param1.seed + i_rep,
                Param1.mu,
                env,
                Param1.sd,
                QTL,
                Param1.mf);
      }
      ap.ProducePopulation();

      for (int i = 0; i < Param1.pheNum.length; i++) {
        ap.ProducePhenotype(i, Param1.MU, Param1.T);
      }
      PrintStream cim = null;
      try {
        cim = new PrintStream(new BufferedOutputStream(new FileOutputStream("CIM.mcd")));
      } catch (Exception e) {

      }
      ap.CIMFormat(cim);
      cim.close();
      gs = new GenomeScan(ap, Param1.step);
      gs.CalculateIPP();

      if (Param1.permutation > 0 && i_rep == 0) {
        PrintStream PoutPW = null;
        PrintStream PoutPL = null;
        PrintStream PoutW = null;
        PrintStream PoutL = null;
        try {
          PoutW = new PrintStream(new BufferedOutputStream(new FileOutputStream("PermuWald.txt")));
          PoutL = new PrintStream(new BufferedOutputStream(new FileOutputStream("PermuLN.txt")));
          PoutPW =
              new PrintStream(new BufferedOutputStream(new FileOutputStream("PermuWaldP.txt")));
          PoutPL = new PrintStream(new BufferedOutputStream(new FileOutputStream("PermuLNP.txt")));
        } catch (Exception E) {
          E.printStackTrace(System.err);
        }

        thresholdLOD = new ArrayList();
        thresholdAP = new ArrayList();
        thresholdDP = new ArrayList();
        thresholdPwald = new ArrayList();
        thresholdPF = new ArrayList();
        for (int i_permu = 0; i_permu < Param1.permutation; i_permu++) {
          ArrayList IMStatistic = MappingProcedure(i_rep, i_permu + 1);
          ArrayList IMLOD = new ArrayList();
          ArrayList IMAT = new ArrayList();
          ArrayList IMDT = new ArrayList();
          ArrayList IMAP = new ArrayList();
          ArrayList IMDP = new ArrayList();
          ArrayList IMPWald = new ArrayList();
          ArrayList IMPF = new ArrayList();
          int c = 0;
          int ct = 0;
          int bt = 0;

          for (int j = 0; j < IMStatistic.size(); j++) {
            PointMappingStatistic pms = (PointMappingStatistic) IMStatistic.get(j);
            PoutPW.print(pms.get_P_wald() + " ");
            PoutPL.print(pms.get_P_LN() + " ");
            PoutW.print(pms.get_wald() + " ");
            PoutL.print(pms.get_LN() + " ");
            IMLOD.add(new Double(pms.get_LOD()));
            IMAT.add(new Double(pms.get_tStatistic_additive()));
            IMDT.add(new Double(pms.get_tStatistic_dominance()));
            IMAP.add(new Double(pms.get_logP_additive()));
            IMPWald.add(new Double(pms.get_logP_wald()));
            IMPF.add(new Double(pms.get_logP_F()));
            if (pms.get_logP_additive() > LogBon) {
              c++;
            }
            if (pms.get_P_additive() < BonT) {
              bt++;
            }
            if (weight.length > 1) {
              IMDP.add(new Double(pms.get_logP_dominance()));
            }
          }
          PoutPW.println();
          PoutPL.println();
          PoutW.println();
          PoutL.println();
          if (c > 0) {
            typeI_LogBon += 1;
          }
          thresholdLOD.add(Collections.max(IMLOD));
          thresholdAP.add(Collections.max(IMAP));
          if (weight.length > 1) {
            thresholdDP.add(Collections.max(IMDP));
          }
          thresholdPF.add(Collections.max(IMPF));
          thresholdPwald.add(Collections.max(IMPWald));
        }
        PoutPW.close();
        PoutPL.close();
        PoutW.close();
        PoutL.close();
      }
      ArrayList IMLOD = MappingProcedure(i_rep, 0);
      SimulationResults.add(IMLOD);
    }
  }
Ejemplo n.º 2
0
  public void SummuarySimulation() {
    calculatePower();
    PrintStream Pout1 = null;
    PrintStream Pout2 = null;
    PrintStream Pout3 = null;
    PrintStream Pout4 = null;
    PrintStream Pout5 = null;
    PrintStream Pout6 = null;
    ChiSquaredDistribution chi = new ChiSquaredDistributionImpl(weight.length);
    try {
      Pout1 = new PrintStream(new BufferedOutputStream(new FileOutputStream("LogAP.txt")));
      Pout2 = new PrintStream(new BufferedOutputStream(new FileOutputStream("LNP.txt")));
      Pout3 = new PrintStream(new BufferedOutputStream(new FileOutputStream("LN.txt")));
      Pout4 = new PrintStream(new BufferedOutputStream(new FileOutputStream("Wald.txt")));
      Pout5 = new PrintStream(new BufferedOutputStream(new FileOutputStream("F.txt")));
      Pout6 = new PrintStream(new BufferedOutputStream(new FileOutputStream("LogDP.txt")));
    } catch (Exception E) {
      E.printStackTrace(System.err);
    }
    for (int i = 0; i < SimulationResults.size(); i++) {
      ArrayList PointStatistics = (ArrayList) SimulationResults.get(i);
      double[] LRT = new double[PointStatistics.size()];
      double[] WALD = new double[PointStatistics.size()];
      DescriptiveStatistics dsLRT = new DescriptiveStatisticsImpl();
      DescriptiveStatistics dsWALD = new DescriptiveStatisticsImpl();
      SimpleRegression sr = new SimpleRegression();
      for (int j = 0; j < PointStatistics.size(); j++) {
        PointMappingStatistic pms = (PointMappingStatistic) PointStatistics.get(j);
        double lod = pms.get_LOD() > 0 ? pms.get_LOD() : 0;
        double ln = pms.get_LN();
        double p = 0;
        try {
          p = chi.cumulativeProbability(ln);
        } catch (Exception E) {
          E.printStackTrace(System.err);
        }
        double logLOD = -1 * Math.log10(1 - p);
        Pout1.print(pms.get_logP_additive() + " ");
        Pout2.print(logLOD + " ");
        Pout3.print(pms.get_LN() + " ");
        Pout4.print(pms.get_wald() + " ");
        Pout5.print(pms.get_logP_F() + " ");
        Pout6.print(pms.get_logP_dominance() + " ");

        dsLRT.addValue(pms.get_LN());
        dsWALD.addValue(pms.get_wald());
        LRT[j] = pms.get_LN();
        WALD[j] = pms.get_wald();
        sr.addData(pms.get_LN(), pms.get_wald());
      }
      System.out.println(
          dsLRT.getMean()
              + " +- "
              + dsLRT.getStandardDeviation()
              + " "
              + dsWALD.getMean()
              + " +- "
              + dsWALD.getStandardDeviation()
              + " cor "
              + sr.getR());
      dsLRT.clear();
      dsWALD.clear();
      sr.clear();
      Pout1.println();
      Pout2.println();
      Pout3.println();
      Pout4.println();
      Pout5.println();
      Pout6.println();
    }
    Pout1.close();
    Pout2.close();
    Pout3.close();
    Pout4.close();
    Pout5.close();
    Pout6.close();
  }
Ejemplo n.º 3
0
  private void calculatePower() {
    ChiSquaredDistribution chi = new ChiSquaredDistributionImpl(weight.length);
    Collections.sort(thresholdAP);
    Collections.sort(thresholdDP);
    Collections.sort(thresholdLOD);
    Collections.sort(thresholdPwald);
    Collections.sort(thresholdPF);
    EmpiricalLogBonAP =
        ((Double) thresholdAP.get((new Double((thresholdAP.size() - 1) * 0.95)).intValue()))
            .doubleValue();
    if (weight.length > 1) {
      EmpiricalLogBonDP =
          ((Double) thresholdDP.get((new Double((thresholdDP.size() - 1) * 0.95)).intValue()))
              .doubleValue();
    }

    threshold_LOD =
        ((Double) thresholdLOD.get((new Double((thresholdLOD.size() - 1) * 0.95)).intValue()))
            .doubleValue();
    threshold_lodp = 0;
    double ln = threshold_LOD / 0.217;
    threshold_Pwald =
        ((Double) thresholdPwald.get((new Double((thresholdPwald.size() - 1) * 0.95)).intValue()))
            .doubleValue();
    threshold_PF =
        ((Double) thresholdPF.get((new Double((thresholdPF.size() - 1) * 0.95)).intValue()))
            .doubleValue();
    try {
      threshold_lodp = -1 * Math.log10(1 - chi.cumulativeProbability(ln));
    } catch (Exception E) {
      E.printStackTrace(System.err);
    }
    System.out.println("BonT: " + BonT);
    System.out.println("LogBon: " + LogBon);
    System.out.println("EmpLogBonAP: " + EmpiricalLogBonAP);
    System.out.println("EmpLogBonDP: " + EmpiricalLogBonDP);
    System.out.println(
        "threshold_LOD: " + "log10:" + threshold_LOD + " LR: " + ln + " P(ln) " + threshold_lodp);
    System.out.println("threshold_PWald: " + threshold_Pwald);
    System.out.println("threshold_PF: " + threshold_PF);
    for (Iterator e = SimulationResults.iterator(); e.hasNext(); ) {
      ArrayList result = (ArrayList) e.next();
      for (Iterator e1 = result.iterator(); e1.hasNext(); ) {
        PointMappingStatistic pms = (PointMappingStatistic) e1.next();
        String key = pms.getKey();
        if (Param1.permutation > 0
            && pointPowerF.containsKey(key)
            && pms.get_logP_F() > threshold_PF) {
          int p = ((Integer) pointPowerF.get(key)).intValue();
          p++;
          pointPowerF.put(key, new Integer(p));
        }
        if (Param1.permutation > 0
            && pointPowerWald.containsKey(key)
            && pms.get_logP_wald() > threshold_Pwald) {
          int p = ((Integer) pointPowerWald.get(key)).intValue();
          p++;
          pointPowerWald.put(key, new Integer(p));
        }
        if (Param1.permutation > 0
            && pointPowerLOD.containsKey(key)
            && pms.get_LOD() > threshold_LOD) {
          int p = ((Integer) pointPowerLOD.get(key)).intValue();
          p++;
          pointPowerLOD.put(key, new Integer(p));
        }
        if (pointPowerAPvalue.containsKey(key) && pms.get_logP_additive() > LogBon) {
          int p = ((Integer) pointPowerAPvalue.get(key)).intValue();
          p++;
          pointPowerAPvalue.put(key, new Integer(p));
        }
        if (pointPowerEmpiricalAPvalue.containsKey(key)
            && pms.get_logP_additive() > EmpiricalLogBonAP) {
          int p = ((Integer) pointPowerEmpiricalAPvalue.get(key)).intValue();
          p++;
          pointPowerEmpiricalAPvalue.put(key, new Integer(p));
        }
        if (weight.length > 1) {
          if (pointPowerDPvalue.containsKey(key) && pms.get_logP_dominance() > LogBon) {
            int p = ((Integer) pointPowerDPvalue.get(key)).intValue();
            p++;
            pointPowerDPvalue.put(key, new Integer(p));
          }
          if (pointPowerEmpiricalDPvalue.containsKey(key)
              && pms.get_logP_dominance() > EmpiricalLogBonDP) {
            int p = ((Integer) pointPowerEmpiricalDPvalue.get(key)).intValue();
            p++;
            pointPowerEmpiricalDPvalue.put(key, new Integer(p));
          }
        }
      }
    }
    Set keys = pointPowerLOD.keySet();
    System.out.println("type I LogBon: " + typeI_LogBon / Param1.permutation);
    System.out.println("QTL\tFP\tWald\tLOD\tBonAP\tEmpAP\tBonDP\tEmpDP");
    for (Iterator e = keys.iterator(); e.hasNext(); ) {
      String key = (String) e.next();
      System.out.print(key + "\t");
      System.out.print((Integer) pointPowerF.get(key) + "\t");
      System.out.print((Integer) pointPowerWald.get(key) + "\t");
      System.out.print((Integer) pointPowerLOD.get(key) + "\t");
      System.out.print((Integer) pointPowerAPvalue.get(key) + "\t");
      System.out.print((Integer) pointPowerEmpiricalAPvalue.get(key) + "\t");
      if (weight.length > 1) {
        System.out.print(pointPowerDPvalue.get(key) + "\t");
        System.out.println(pointPowerEmpiricalDPvalue.get(key) + "\t");
      } else {
        System.out.println("--\t--\t");
      }
    }
  }