Removed experimental feature (aka hack) that was meant for 1000G consensus but remained in VQSR data manager - QD was being scaled by indel length. There's no evidence any more that QD is length-dependent, neither in CEU trio data nor in latest 1000G P2 calls

This commit is contained in:
Guillermo del Angel 2012-04-20 10:58:34 -04:00
parent a1561a97c4
commit de68363c23
2 changed files with 3 additions and 11 deletions

View File

@ -241,14 +241,6 @@ public class VariantDataManager {
value += -0.25 + 0.5 * GenomeAnalysisEngine.getRandomGenerator().nextDouble();
}
if (vc.isIndel() && annotationKey.equalsIgnoreCase("QD")) {
// normalize QD by event length for indel case
int eventLength = Math.abs(vc.getAlternateAllele(0).getBaseString().length() - vc.getReference().getBaseString().length()); // ignore multi-allelic complication here for now
if (eventLength > 0) { // sanity check
value /= (double)eventLength;
}
}
if( jitter && annotationKey.equalsIgnoreCase("HaplotypeScore") && MathUtils.compareDoubles(value, 0.0, 0.0001) == 0 ) { value = -0.2 + 0.4*GenomeAnalysisEngine.getRandomGenerator().nextDouble(); }
if( jitter && annotationKey.equalsIgnoreCase("FS") && MathUtils.compareDoubles(value, 0.0, 0.001) == 0 ) { value = -0.2 + 0.4*GenomeAnalysisEngine.getRandomGenerator().nextDouble(); }
} catch( Exception e ) {

View File

@ -73,9 +73,9 @@ public class VariantRecalibrationWalkersIntegrationTest extends WalkerTest {
}
VRTest indel = new VRTest("combined.phase1.chr20.raw.indels.sites.vcf",
"6d7ee4cb651c8b666e4a4523363caaff", // tranches
"ee5b408c8434a594496118875690c438", // recal file
"5d7e07d8813db96ba3f3dfe4737f83d1"); // cut VCF
"da4458d05f6396f5c4ab96f274e5ccdc", // tranches
"cf380d9b0ae04c8918be8425f82035b4", // recal file
"b00e5e5a6807df8ed1682317948e8a6d"); // cut VCF
@DataProvider(name = "VRIndelTest")
public Object[][] createData2() {