From a942fa38ef87c1e1565d6a1cff041d8a62aaeb0a Mon Sep 17 00:00:00 2001
From: Guillermo del Angel <delangel@broadinstitute.org>
Date: Thu, 15 Sep 2011 10:22:28 -0400
Subject: [PATCH] Refine the way we merge records in CombineVariants of
 different types. As of before, two records of different types were not
 combined and were kept separate. This is still the case, except when the
 alleles of one record are a strict subset of alleles of another record. For
 example, a SNP with alleles {A*,T} and a mixed record with alleles {A*,T,
 AAT} are now combined when start position matches.

---
 .../walkers/variantutils/CombineVariants.java | 35 +++++++++++++++++--
 .../variantcontext/VariantContextUtils.java   | 12 +++++++
 2 files changed, 45 insertions(+), 2 deletions(-)

diff --git a/public/java/src/org/broadinstitute/sting/gatk/walkers/variantutils/CombineVariants.java b/public/java/src/org/broadinstitute/sting/gatk/walkers/variantutils/CombineVariants.java
index 7062f17e5..3e3b29a7f 100755
--- a/public/java/src/org/broadinstitute/sting/gatk/walkers/variantutils/CombineVariants.java
+++ b/public/java/src/org/broadinstitute/sting/gatk/walkers/variantutils/CombineVariants.java
@@ -234,16 +234,47 @@ public class CombineVariants extends RodWalker<Integer, Integer> {
         if (minimumN > 1 && (vcs.size() - numFilteredRecords < minimumN))
             return 0;
         
-        List<VariantContext> mergedVCs = new ArrayList<VariantContext>();
+        List<VariantContext> preMergedVCs = new ArrayList<VariantContext>();
         Map<VariantContext.Type, List<VariantContext>> VCsByType = VariantContextUtils.separateVariantContextsByType(vcs);
         // iterate over the types so that it's deterministic
         for ( VariantContext.Type type : VariantContext.Type.values() ) {
             if ( VCsByType.containsKey(type) )
-                mergedVCs.add(VariantContextUtils.simpleMerge(getToolkit().getGenomeLocParser(), VCsByType.get(type),
+                preMergedVCs.add(VariantContextUtils.simpleMerge(getToolkit().getGenomeLocParser(), VCsByType.get(type),
                         priority, filteredRecordsMergeType, genotypeMergeOption, true, printComplexMerges,
                         SET_KEY, filteredAreUncalled, MERGE_INFO_WITH_MAX_AC));
         }
 
+        List<VariantContext> mergedVCs = new ArrayList<VariantContext>();
+        // se have records merged but separated by type. If a particular record is for example a snp but all alleles are a subset of an existing mixed record,
+        // we will still merge those records.
+        if (preMergedVCs.size() > 1) {
+            for (VariantContext vc1 : preMergedVCs) {
+                VariantContext newvc = vc1;
+                boolean merged = false;
+                for (int k=0; k < mergedVCs.size(); k++) {
+                    VariantContext vc2 = mergedVCs.get(k);
+
+                    if (VariantContextUtils.allelesAreSubset(vc1,vc2) || VariantContextUtils.allelesAreSubset(vc2,vc1)) {
+                        // all alleles of vc1 are contained in vc2 but they are of different type (say, vc1 is snp, vc2 is complex): try to merget v1 into v2
+                        List<VariantContext> vcpair = new ArrayList<VariantContext>();
+                        vcpair.add(vc1);
+                        vcpair.add(vc2);
+                        newvc = VariantContextUtils.simpleMerge(getToolkit().getGenomeLocParser(), vcpair,
+                                               priority, filteredRecordsMergeType, genotypeMergeOption, true, printComplexMerges,
+                                               SET_KEY, filteredAreUncalled, MERGE_INFO_WITH_MAX_AC);
+                        mergedVCs.set(k,newvc);
+                        merged = true;
+                        break;
+                    }
+                }
+                if (!merged)
+                    mergedVCs.add(vc1);
+            }
+        }
+        else {
+            mergedVCs = preMergedVCs;
+        }
+
         for ( VariantContext mergedVC : mergedVCs ) {
             // only operate at the start of events
             if ( mergedVC == null )
diff --git a/public/java/src/org/broadinstitute/sting/utils/variantcontext/VariantContextUtils.java b/public/java/src/org/broadinstitute/sting/utils/variantcontext/VariantContextUtils.java
index 986d6305c..506bb3b33 100755
--- a/public/java/src/org/broadinstitute/sting/utils/variantcontext/VariantContextUtils.java
+++ b/public/java/src/org/broadinstitute/sting/utils/variantcontext/VariantContextUtils.java
@@ -663,6 +663,18 @@ public class VariantContextUtils {
         return merged;
     }
 
+    public static boolean allelesAreSubset(VariantContext vc1, VariantContext vc2) {
+        // if all alleles of vc1 are a contained in alleles of vc2, return true
+        if (!vc1.getReference().equals(vc2.getReference()))
+            return false;
+
+        for (Allele a :vc1.getAlternateAlleles()) {
+            if (!vc2.getAlternateAlleles().contains(a))
+                return false;
+        }
+
+        return true;
+    }
     public static VariantContext createVariantContextWithTrimmedAlleles(VariantContext inputVC) {
         // see if we need to trim common reference base from all alleles
         boolean trimVC;