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

Better genotype concordance module. Code refactoring for clarity (please see below/after for educational purposes). Now reports variant sensitivity, concordance, and genotype error rate by default. Also aggregates this data across all samples, so you get a per sample and overall stats for each of these in the allSamples row. 

git-svn-id: file:///humgen/gsa-scr1/gsa-engineering/svn_contents/trunk@3265 348d0f76-0448-11de-a6fe-93d51630548a
This commit is contained in:
depristo 2010-04-28 13:10:11 +00:00
parent 64c5f287c5
commit 5dce16a8f1
2 changed files with 118 additions and 82 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

178
java/src/org/broadinstitute/sting/gatk/walkers/varianteval/GenotypeConcordance.java
View File

@ -52,7 +52,6 @@ public class GenotypeConcordance extends VariantEvaluator {
@DataPoint(name="summary", description = "the concordance statistics summary for each sample")
SampleSummaryStats sampleSummaryStats = null;
// two histograms of variant quality scores, for true positive and false positive calls
@DataPoint(description = "the variant quality score histograms for true positive and false positive calls")
QualityScoreHistograms qualityScoreHistograms = null;
@ -225,32 +224,7 @@ public class GenotypeConcordance extends VariantEvaluator {
}
public String toString() {
return getName() + ": " + getTableRows();
}
private static List<String> SAMPLE_HEADER =
Arrays.asList("sample",
"total_true_ref", "n_ref/ref", "%_ref/ref",
"n_ref/no-call", "n_ref/het", "n_ref/hom",
"total_true_het", "n_het/het", "%_het/het",
"n_het/no-call", "n_het/ref", "n_het/hom",
"total_true_hom", "n_hom/hom", "%_hom/hom",
"n_hom/no-call", "n_hom/ref", "n_hom/het");
private static List<String> FREQUENCY_HEADER =
Arrays.asList("alleleCount", "n_found", "n_missed", "%_found");
// making it a table
public List<String> getTableHeader() {
ArrayList<String> header = new ArrayList<String>();
header.addAll(SAMPLE_HEADER);
header.addAll(FREQUENCY_HEADER);
return header;
}
public List<List<String>> getTableRows() {
return null;
return getName() + ": <table>";
}
private boolean warnedAboutValidationData = false;
@ -449,6 +423,15 @@ class SampleStats implements TableType {
* a table of sample names to genotype concordance summary statistics
*/
class SampleSummaryStats implements TableType {
private final static String ALL_SAMPLES_KEY = "allSamples";
private final static String[] COLUMN_KEYS = new String[]{
"percent_comp_het_called_het",
"percent_comp_het_called_var",
"percent_comp_hom_called_hom",
"percent_comp_hom_called_var",
"percent_variant_sensitivity",
"percent_genotype_concordance",
"percent_genotype_error_rate"};
// sample to concordance stats object
private final HashMap<String, double[]> concordanceSummary = new HashMap<String, double[]>();
@ -466,26 +449,77 @@ class SampleSummaryStats implements TableType {
* @return a list of objects, in this case strings, that are the column names
*/
public Object[] getColumnKeys() {
return new String[]{"%het_called_het","%hom_called_hom","%nonref_called_nonref","calledGenotypeConcordance","calledNonRefConcordance"};
return COLUMN_KEYS;
}
public SampleSummaryStats(final VariantContext vc) {
concordanceSummary.put(ALL_SAMPLES_KEY, new double[COLUMN_KEYS.length]);
for( final String sample : vc.getSampleNames() ) {
concordanceSummary.put(sample, new double[5]); // There are five summary values to report out
concordanceSummary.put(sample, new double[COLUMN_KEYS.length]);
}
}
public Object getCell(int x, int y) {
final Object[] rowKeys = getRowKeys();
return String.format("%.3f",concordanceSummary.get(rowKeys[x])[y]);
return String.format("%.2f",concordanceSummary.get(rowKeys[x])[y]);
}
/**
* Helper routine that sums up all columns / rows found in stats specified by all pairs in d1 x d2
*
* @param stats
* @param d1
* @param d2
* @return
*/
private long sumStatsAllPairs( final long[][] stats, EnumSet<Genotype.Type> d1, EnumSet<Genotype.Type> d2 ) {
long sum = 0L;
for(final Genotype.Type e1 : d1 ) {
for(final Genotype.Type e2 : d2 ) {
sum += stats[e1.ordinal()][e2.ordinal()];
}
}
return sum;
}
private long sumStatsDiag( final long[][] stats, EnumSet<Genotype.Type> d1) {
long sum = 0L;
for(final Genotype.Type e1 : d1 ) {
sum += stats[e1.ordinal()][e1.ordinal()];
}
return sum;
}
private double ratio(long numer, long denom) {
return denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0;
}
final long[] allSamplesNumerators = new long[COLUMN_KEYS.length];
final long[] allSamplesDenominators = new long[COLUMN_KEYS.length];
private void updateSummaries(int i, double[] summary, long numer, long denom ) {
allSamplesNumerators[i] += numer;
allSamplesDenominators[i] += denom;
summary[i] = ratio(numer, denom);
}
/**
* Calculate the five summary stats per sample
* @param sampleStats The Map which holds concordance values per sample
*/
public void generateSampleSummaryStats( final SampleStats sampleStats ) {
EnumSet<Genotype.Type> allVariantGenotypes = EnumSet.of(Genotype.Type.HOM_VAR, Genotype.Type.HET);
EnumSet<Genotype.Type> allCalledGenotypes = EnumSet.of(Genotype.Type.HOM_VAR, Genotype.Type.HET, Genotype.Type.HOM_REF);
EnumSet<Genotype.Type> allGenotypes = EnumSet.allOf(Genotype.Type.class);
for( final String sample : concordanceSummary.keySet() ) {
if ( sample.equals(ALL_SAMPLES_KEY) ) continue;
final long[][] stats = sampleStats.concordanceStats.get(sample);
final double[] summary = concordanceSummary.get(sample);
if( stats == null ) { throw new StingException( "SampleStats and SampleSummaryStats contain different samples! sample = " + sample ); }
@ -494,57 +528,51 @@ class SampleSummaryStats implements TableType {
// Summary 0: % het called as het
numer = stats[Genotype.Type.HET.ordinal()][Genotype.Type.HET.ordinal()];
denom = 0L;
for(final Genotype.Type type : Genotype.Type.values()) {
denom += stats[Genotype.Type.HET.ordinal()][type.ordinal()];
}
summary[0] = ( denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0 );
denom = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HET), allGenotypes);
updateSummaries(0, summary, numer, denom);
// Summary 1: % homVar called as homVar
// Summary 1: % het called as var
numer = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HET), allVariantGenotypes);
denom = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HET), allGenotypes);
updateSummaries(1, summary, numer, denom);
// Summary 2: % homVar called as homVar
numer = stats[Genotype.Type.HOM_VAR.ordinal()][Genotype.Type.HOM_VAR.ordinal()];
denom = 0L;
for(final Genotype.Type type : Genotype.Type.values()) {
denom += stats[Genotype.Type.HOM_VAR.ordinal()][type.ordinal()];
}
summary[1] = ( denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0 );
denom = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HOM_VAR), allGenotypes);
updateSummaries(2, summary, numer, denom);
// Summary 2: % non-ref called as non-ref
numer = 0L;
numer += stats[Genotype.Type.HOM_VAR.ordinal()][Genotype.Type.HOM_VAR.ordinal()];
numer += stats[Genotype.Type.HOM_VAR.ordinal()][Genotype.Type.HET.ordinal()];
numer += stats[Genotype.Type.HET.ordinal()][Genotype.Type.HOM_VAR.ordinal()];
numer += stats[Genotype.Type.HET.ordinal()][Genotype.Type.HET.ordinal()];
denom = 0L;
for(final Genotype.Type type : Genotype.Type.values()) {
denom += stats[Genotype.Type.HOM_VAR.ordinal()][type.ordinal()];
denom += stats[Genotype.Type.HET.ordinal()][type.ordinal()];
}
summary[2] = ( denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0 );
// Summary 3: % homVars called as var
numer = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HOM_VAR), allVariantGenotypes);
denom = sumStatsAllPairs(stats, EnumSet.of(Genotype.Type.HOM_VAR), allGenotypes);
updateSummaries(3, summary, numer, denom);
// Summary 3: genotype concordance of sites called in eval track
numer = 0L;
numer += stats[Genotype.Type.HOM_VAR.ordinal()][Genotype.Type.HOM_VAR.ordinal()];
numer += stats[Genotype.Type.HET.ordinal()][Genotype.Type.HET.ordinal()];
numer += stats[Genotype.Type.HOM_REF.ordinal()][Genotype.Type.HOM_REF.ordinal()];
denom = 0L;
for(final Genotype.Type type : new Genotype.Type[]{Genotype.Type.HOM_VAR, Genotype.Type.HET, Genotype.Type.HOM_REF} ) { // excluding no calls here
denom += stats[Genotype.Type.HOM_VAR.ordinal()][type.ordinal()];
denom += stats[Genotype.Type.HET.ordinal()][type.ordinal()];
denom += stats[Genotype.Type.HOM_REF.ordinal()][type.ordinal()];
}
summary[3] = ( denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0 );
// Summary 4: % non-ref called as non-ref
// MAD: this is known as the variant sensitivity (# non-ref according to comp found in eval / # non-ref in comp)
numer = sumStatsAllPairs(stats, allVariantGenotypes, allVariantGenotypes);
denom = sumStatsAllPairs(stats, allVariantGenotypes, allGenotypes);
updateSummaries(4, summary, numer, denom);
// Summary 4: genotype concordance of sites called non-ref in eval track
numer = 0L;
numer += stats[Genotype.Type.HOM_VAR.ordinal()][Genotype.Type.HOM_VAR.ordinal()];
numer += stats[Genotype.Type.HET.ordinal()][Genotype.Type.HET.ordinal()];
denom = 0L;
for(final Genotype.Type type : new Genotype.Type[]{Genotype.Type.HOM_VAR, Genotype.Type.HET, Genotype.Type.HOM_REF} ) { // excluding no calls here
denom += stats[Genotype.Type.HOM_VAR.ordinal()][type.ordinal()];
denom += stats[Genotype.Type.HET.ordinal()][type.ordinal()];
}
summary[4] = ( denom != 0L ? 100.0 * ( ((double)numer) / ((double)denom) ) : 0.0 );
// Summary 5: genotype concordance of sites called in eval track
// MAD: this is the tradition genotype concordance
numer = sumStatsDiag(stats, allCalledGenotypes);
denom = sumStatsAllPairs(stats, allCalledGenotypes, allCalledGenotypes);
updateSummaries(5, summary, numer, denom);
// Summary 6: genotype concordance of sites called non-ref in eval track
long homrefConcords = stats[Genotype.Type.HOM_REF.ordinal()][Genotype.Type.HOM_REF.ordinal()];
long diag = sumStatsDiag(stats, allVariantGenotypes);
long allNoHomRef = sumStatsAllPairs(stats, allCalledGenotypes, allCalledGenotypes) - homrefConcords;
numer = allNoHomRef - diag;
denom = allNoHomRef;
updateSummaries(6, summary, numer, denom);
}
// update the final summary stats
final double[] allSamplesSummary = concordanceSummary.get(ALL_SAMPLES_KEY);
for ( int i = 0; i < allSamplesSummary.length; i++) {
allSamplesSummary[i] = ratio(allSamplesNumerators[i], allSamplesDenominators[i]);
}
}
public String getName() {

22
java/test/org/broadinstitute/sting/gatk/walkers/varianteval/VariantEvalIntegrationTest.java
View File

@ -10,18 +10,26 @@ import java.util.Map;
public class
VariantEvalIntegrationTest extends WalkerTest {
private static String cmdRoot = "-T VariantEval" +
" -R " + oneKGLocation + "reference/human_b36_both.fasta -reportType Grep";
" -R " + oneKGLocation + "reference/human_b36_both.fasta";
private static String root = cmdRoot +
" -D " + GATKDataLocation + "dbsnp_129_b36.rod" +
" -B eval,VCF," + validationDataLocation + "yri.trio.gatk_glftrio.intersection.annotated.filtered.chr1.vcf" +
" -B comp_genotypes,VCF," + validationDataLocation + "yri.trio.gatk.ug.head.vcf";
" -B comp_genotypes,VCF," + validationDataLocation + "yri.trio.gatk.ug.head.vcf -reportType Grep";
@Test
public void testVEGenotypeConcordance() {
WalkerTestSpec spec = new WalkerTestSpec( cmdRoot + " -B eval,VCF," + validationDataLocation + "GenotypeConcordanceEval.vcf -B comp,VCF," + validationDataLocation + "GenotypeConcordanceComp.vcf -E GenotypeConcordance -reportType CSV -o %s",
1, // just one output file
Arrays.asList("608b32189818df7271c546dd5f257033"));
executeTest("testVEGenotypeConcordance", spec);
}
@Test
public void testVESimple() {
HashMap<String, String> expectations = new HashMap<String, String>();
expectations.put("-L 1:1-10,000,000", "f2ce1e0e9bbe81b482a9448cb88e2263");
expectations.put("-L 1:1-10,000,000 -family NA19238+NA19239=NA19240 -MVQ 0", "b39229f55b726c2f61a17463a3c883be");
expectations.put("-L 1:1-10,000,000", "b2fcb4a3e852c43acb93f7720a3c4b76");
expectations.put("-L 1:1-10,000,000 -family NA19238+NA19239=NA19240 -MVQ 0", "63e9b9db244f4a593e643d2d7431219e");
for ( Map.Entry<String, String> entry : expectations.entrySet() ) {
String extraArgs = entry.getKey();
@ -42,10 +50,10 @@ public class
" -B comp_hapmap,VCF," + validationDataLocation + "CEU_hapmap_nogt_23.vcf";
String matchingMD5 = "f8c7bd3ce4499cff485d2170ce896af9";
String matchingMD5 = "fc1d4d2aca0667605a6b4d486fcbedf2";
expectations.put("", matchingMD5);
expectations.put(" -known comp_hapmap -known dbsnp", matchingMD5);
expectations.put(" -known comp_hapmap", "99c687b95b1a4614cd76570bfd24e7a6");
expectations.put(" -known comp_hapmap", "fd9cd9970f7e509ca476502869063929");
for ( Map.Entry<String, String> entry : expectations.entrySet() ) {
String extraArgs2 = entry.getKey();
@ -63,7 +71,7 @@ public class
String extraArgs = "-L 1:1-10,000,000 -family NA19238+NA19239=NA19240 -MVQ 30";
WalkerTestSpec spec = new WalkerTestSpec( root + " " + extraArgs + " -o %s -outputVCF %s",
2,
Arrays.asList("bfbad4053c89c264cec16040df0bcc4c", "b4a42c90318adc88361691ece50426f2"));
Arrays.asList("f7a06a988573c5b1b69e52fb8e0edc06", "b4a42c90318adc88361691ece50426f2"));
executeTest("testVEWriteVCF", spec);
}
}