Fixes for running HaplotypeCaller with reduced reads: a) minor refactoring, pulled out code to compute mean representative count to ReadUtils, b) Don't use min representative count over kmer when constructing de Bruijn graph - this creates many paths with multiplicity=1 and makes us lose a lot of SNP's at edge of capture targets. Use mean instead

protected/java/src/org/broadinstitute/sting/gatk/walkers/haplotypecaller/LikelihoodCalculationEngine.java

@@ -30,6 +30,7 @@ import com.google.java.contract.Requires;
 import org.broadinstitute.sting.utils.*;
 import org.broadinstitute.sting.utils.collections.Pair;
 import org.broadinstitute.sting.utils.sam.GATKSAMRecord;
+import org.broadinstitute.sting.utils.sam.ReadUtils;
 import org.broadinstitute.sting.utils.variantcontext.Allele;
 import org.broadinstitute.sting.utils.variantcontext.VariantContext;
 
@@ -92,7 +93,7 @@ public class LikelihoodCalculationEngine {
         final int[][] readCounts = new int[numHaplotypes][numReads];
         for( int iii = 0; iii < numReads; iii++ ) {
             final GATKSAMRecord read = reads.get(iii);
-            final int readCount = getRepresentativeReadCount(read);
+            final int readCount = ReadUtils.getMeanRepresentativeReadCount(read);
 
             final byte[] overallGCP = new byte[read.getReadLength()];
             Arrays.fill( overallGCP, constantGCP ); // Is there a way to derive empirical estimates for this from the data?
@@ -123,15 +124,6 @@ public class LikelihoodCalculationEngine {
         }
     }
 
-    private static int getRepresentativeReadCount(GATKSAMRecord read) {
-        if (!read.isReducedRead())
-            return 1;
-
-        // compute mean representative read counts
-        final byte[] counts = read.getReducedReadCounts();
-        return MathUtils.sum(counts)/counts.length;
-    }
-
     private static int computeFirstDifferingPosition( final byte[] b1, final byte[] b2 ) {
         for( int iii = 0; iii < b1.length && iii < b2.length; iii++ ){
             if( b1[iii] != b2[iii] ) {

protected/java/src/org/broadinstitute/sting/gatk/walkers/haplotypecaller/SimpleDeBruijnAssembler.java

@@ -198,8 +198,10 @@ public class SimpleDeBruijnAssembler extends LocalAssemblyEngine {
                     }
                     int countNumber = 1;
                     if (read.isReducedRead()) {
-                        // compute min (?) number of reduced read counts in current kmer span
+                        // compute mean number of reduced read counts in current kmer span
-                        countNumber = MathUtils.arrayMin(Arrays.copyOfRange(reducedReadCounts,iii,iii+KMER_LENGTH+1));
+                        final byte[] counts = Arrays.copyOfRange(reducedReadCounts,iii,iii+KMER_LENGTH+1);
+                        // precise rounding can make a difference with low consensus counts
+                        countNumber = (int)Math.round((double)MathUtils.sum(counts)/counts.length);
                     }
 
                     if( !badKmer ) {

public/java/src/org/broadinstitute/sting/utils/sam/ReadUtils.java

@@ -56,6 +56,15 @@ public class ReadUtils {
     private static int DEFAULT_ADAPTOR_SIZE = 100;
     public static int CLIPPING_GOAL_NOT_REACHED = -1;
 
+    public static int getMeanRepresentativeReadCount(GATKSAMRecord read) {
+        if (!read.isReducedRead())
+            return 1;
+
+        // compute mean representative read counts
+        final byte[] counts = read.getReducedReadCounts();
+        return (int)Math.round((double)MathUtils.sum(counts)/counts.length);
+    }
+
     /**
      * A marker to tell which end of the read has been clipped
      */