zzh
/
gatk-3.8
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

More tables output by CovariateCounterWalker AND made CovariateCounterWalker and LogisticRecalibration aware of positive and negative strandedness of data which changes the regression output significantly. 

git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@568 348d0f76-0448-11de-a6fe-93d51630548a
This commit is contained in:
andrewk 2009-04-30 01:22:50 +00:00
parent f7a877bfeb
commit b630f2f2f1
2 changed files with 107 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

109
java/src/org/broadinstitute/sting/playground/gatk/walkers/CovariateCounterWalker.java
View File

@ -21,8 +21,11 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
@Argument(fullName="MAX_QUAL_SCORE", shortName="mqs", required=false,defaultValue="63")
public int MAX_QUAL_SCORE;
@Argument(fullName="LOG_REG_TRAINING_FILEROOT", shortName="lrtf", required=false, defaultValue="dinuc", doc="Filename root for the outputted logistic regression training files")
public String LOG_REG_TRAINING_FILEROOT;
@Argument(fullName="OUTPUT_FILEROOT", shortName="outroot", required=false, defaultValue="output", doc="Filename root for the outputted logistic regression training files")
public String OUTPUT_FILEROOT;
@Argument(fullName="CREATE_TRAINING_DATA", shortName="trainingdata", required=false, defaultValue="false", doc="Create training data files for logistic regression")
public boolean CREATE_TRAINING_DATA;
int NDINUCS = 16;
RecalData[][][] data = new RecalData[MAX_READ_LENGTH+1][MAX_QUAL_SCORE+1][NDINUCS];
@ -35,7 +38,9 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
String dinuc_root = "dinuc";
ArrayList<PrintStream> dinuc_outs = new ArrayList<PrintStream>();
PrintStream covars_out = new PrintStream("covars.out");
PrintStream ByQual = new PrintStream("quality_empirical_vs_observed.csv");
PrintStream ByQualFile; // = new PrintStream("quality_empirical_vs_observed.csv");
PrintStream ByCycleFile;
PrintStream ByDinucFile;
private class RecalData {
long N;
@ -93,18 +98,23 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
List<SAMRecord> reads = context.getReads();
List<Integer> offsets = context.getOffsets();
for (int i =0; i < reads.size(); i++ ) {
SAMRecord read = reads.get(i);
int offset = offsets.get(i);
if ( offset > 0 ) {
int numBases = read.getReadLength();
if ( offset > 0 && offset < (numBases-1) ) { // skip first and last bases because they suck and they don't have a dinuc count
char base = (char)read.getReadBases()[offset];
char prevBase = (char)read.getReadBases()[offset-1];
int qual = (int)read.getBaseQualities()[offset];
//out.printf("%d %d %d\n", offset, qual, bases2dinucIndex(prevBase,base));
// previous base is the next base in terms of machine chemistry if this is a negative strand
char prevBase = (char)read.getReadBases()[offset + (read.getReadNegativeStrandFlag() ? 1 : -1)];
int dinuc_index = bases2dinucIndex(prevBase,base);
if (qual > 0 && qual <= MAX_QUAL_SCORE) {
RecalData datum = data[offset][qual][dinuc_index];
datum.inc(base,ref);
dinuc_outs.get(dinuc_index).format("%d,%d,%d\n", qual, offset, nuc2num[base]==nuc2num[ref] ? 0 : 1);
// Convert offset into cycle position which means reversing the position of reads on the negative strand
int cycle = read.getReadNegativeStrandFlag() ? numBases - offset - 1 : offset;
data[cycle][qual][dinuc_index].inc(base,ref);
if (CREATE_TRAINING_DATA)
dinuc_outs.get(dinuc_index).format("%d,%d,%d\n", qual, cycle, nuc2num[base]==nuc2num[ref] ? 0 : 1);
}
}
}
@ -120,17 +130,77 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
}
qualityEmpiricalObserved();
qualityDiffVsCycle();
qualityDiffVsDinucleotide();
// Close dinuc filehandles
for ( PrintStream dinuc_stream : this.dinuc_outs)
dinuc_stream.close();
if (CREATE_TRAINING_DATA)
for ( PrintStream dinuc_stream : this.dinuc_outs)
dinuc_stream.close();
}
class MeanReportedQuality {
double Qn = 0;
long n = 0;
void inc(double Q, long n) {
this.n += n;
Qn += Q * n;
}
double result() {
return Qn / n;
}
}
public void qualityDiffVsCycle() {
ArrayList<RecalData> ByCycle = new ArrayList<RecalData>();
ArrayList<MeanReportedQuality> ByCycleReportedQ = new ArrayList<MeanReportedQuality>();
ByCycleFile.printf("cycle,Qemp-obs,Qemp,Qobs,B,N%n");
for (int c=0; c < MAX_READ_LENGTH+1; c++) {
ByCycle.add(new RecalData(c, -1, "-"));
ByCycleReportedQ.add(new MeanReportedQuality());
}
for ( RecalData datum: flattenData ) {
ByCycle.get(datum.pos).inc(datum.N, datum.B);
ByCycleReportedQ.get(datum.pos).inc(datum.qual, datum.N);
}
for (int c=0; c < MAX_READ_LENGTH+1; c++) {
double empiricalQual = -10 * Math.log10((double)ByCycle.get(c).B / ByCycle.get(c).N);
double reportedQual = ByCycleReportedQ.get(c).result();
ByCycleFile.printf("%d, %f, %f, %f, %d, %d%n", c, empiricalQual-reportedQual, empiricalQual, reportedQual, ByCycle.get(c).B, ByCycle.get(c).N);
}
}
public void qualityDiffVsDinucleotide() {
ArrayList<RecalData> ByCycle = new ArrayList<RecalData>();
ArrayList<MeanReportedQuality> ByCycleReportedQ = new ArrayList<MeanReportedQuality>();
ByDinucFile.printf("dinuc,Qemp-obs,Qemp,Qobs,B,N%n");
for (int c=0; c < NDINUCS; c++) {
ByCycle.add(new RecalData(-1, -1, dinucIndex2bases(c)));
ByCycleReportedQ.add(new MeanReportedQuality());
}
for ( RecalData datum: flattenData ) {
int dinucIndex = bases2dinucIndex(datum.dinuc.charAt(0), datum.dinuc.charAt(1));
ByCycle.get(dinucIndex).inc(datum.N, datum.B);
ByCycleReportedQ.get(dinucIndex).inc(datum.qual, datum.N);
}
for (int c=0; c < NDINUCS; c++) {
double empiricalQual = -10 * Math.log10((double)ByCycle.get(c).B / ByCycle.get(c).N);
double reportedQual = ByCycleReportedQ.get(c).result();
ByDinucFile.printf("%s, %f, %f, %f, %d, %d%n", ByCycle.get(c).dinuc, empiricalQual-reportedQual, empiricalQual, reportedQual, ByCycle.get(c).B, ByCycle.get(c).N);
}
}
public void qualityEmpiricalObserved() {
ArrayList<RecalData> ByQ = new ArrayList<RecalData>();
ByQual.printf("Qcal, Qobs, B, N%n");
ByQualFile.printf("Qrep,Qemp,B,N%n");
for (int q=0; q<MAX_QUAL_SCORE+1; q++)
ByQ.add(new RecalData(-1,q,"-"));
@ -140,7 +210,7 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
for (int q=0; q<MAX_QUAL_SCORE; q++) {
double empiricalQual = -10 * Math.log10((double)ByQ.get(q).B / ByQ.get(q).N);
ByQual.printf("%d, %f, %d, %d%n", q, empiricalQual, ByQ.get(q).B, ByQ.get(q).N);
ByQualFile.printf("%d, %f, %d, %d%n", q, empiricalQual, ByQ.get(q).B, ByQ.get(q).N);
//out.printf("%3d,%s,%3d,%5.1f,%5.1f,%6d,%6d", pos, dinuc, qual, empiricalQual, qual-empiricalQual, N, B);
}
}
@ -182,11 +252,14 @@ public class CovariateCounterWalker extends LocusWalker<Integer, Integer> {
// Print out data for regression
public CovariateCounterWalker() throws FileNotFoundException {
for ( int i=0; i<NDINUCS; i++) {
PrintStream next_dinuc = new PrintStream( dinuc_root+"."+dinucIndex2bases(i)+".csv");
next_dinuc.println("logitQ,pos,indicator");
dinuc_outs.add(next_dinuc);
}
ByQual = new PrintStream("by_quality.csv");
if (CREATE_TRAINING_DATA)
for ( int i=0; i<NDINUCS; i++) {
PrintStream next_dinuc = new PrintStream( dinuc_root+"."+dinucIndex2bases(i)+".csv");
next_dinuc.println("logitQ,pos,indicator");
dinuc_outs.add(next_dinuc);
}
ByQualFile = new PrintStream(OUTPUT_FILEROOT+".empirical_v_reported_quality.csv");
ByCycleFile = new PrintStream(OUTPUT_FILEROOT+".quality_difference_v_cycle.csv");
ByDinucFile = new PrintStream(OUTPUT_FILEROOT+".quality_difference_v_dinucleotide.csv");
}
}

25
java/src/org/broadinstitute/sting/playground/gatk/walkers/LogisticRecalibrationWalker.java
View File

@ -96,25 +96,32 @@ public class LogisticRecalibrationWalker extends ReadWalker<SAMRecord, SAMFileWr
byte[] quals = read.getBaseQualities();
byte[] recalQuals = new byte[quals.length];
int numBases = read.getReadLength();
recalQuals[0] = quals[0]; // can't change the first -- no dinuc
for ( int i = 1; i < bases.length; i++ ) {
int cycle = i;
recalQuals[numBases-1] = quals[numBases-1]; // can't change last -- no dinuc
for ( int i = 1; i < numBases-1; i++ ) { // skip first and last base, qual already set because no dinuc
// Take into account that previous base is the next base in terms of machine chemistry if this is a negative strand
int cycle = read.getReadNegativeStrandFlag() ? numBases - i - 1 : i;
String dinuc = String.format("%c%c", bases[i + (read.getReadNegativeStrandFlag() ? 1 : -1)], bases[i]);
byte qual = quals[i];
String dinuc = String.format("%c%c", bases[i-1], bases[i]);
//System.out.printf("dinuc %c %c%n", bases[i-1], bases[i]);
LogisticRegressor regressor = regressors.get(dinuc);
double P = -1;
byte newQual = qual;
byte newQual;
if ( regressor != null ) { // no N or some other unexpected bp in the stream
double logPOver1minusP = regressor.regress((double)cycle+1, (double)qual);
double gamma = regressor.regress((double)cycle+1, (double)qual);
double expGamma = Math.exp(gamma);
double finalP = expGamma / (1+expGamma);
newQual = QualityUtils.probToQual(1-finalP);
//newQual = -10 * Math.round(logPOver1minusP)
/*double POver1minusP = Math.pow(10, logPOver1minusP);
P = POver1minusP / (1 + POver1minusP);
newQual = QualityUtils.probToQual(P);*/
P = POver1minusP / (1 + POver1minusP);*/
//newQual = QualityUtils.probToQual(P);
newQual = (byte)Math.min(Math.round(-10*logPOver1minusP),63);
//newQual = (byte)Math.min(Math.round(-10*logPOver1minusP),63);
//System.out.printf("Recal %s %d %d => %f => %f leads to %d%n", dinuc, cycle, qual, logPOver1minusP, P, newQual);
}else{
newQual = qual;
}
recalQuals[i] = newQual;