Merge pull request #266 from dib-lab/filter/lowlik

Apply filter to variant calls with LIKESCORE < 0. Closes #253.

kevlar/simlike.py

@@ -229,6 +229,8 @@ def simlike(variants, case, controls, refr, mu=30.0, sigma=8.0, epsilon=0.001,
 
     allcalls.sort(key=lambda c: c.attribute('LIKESCORE'), reverse=True)
     for call in allcalls:
+        if call.attribute('LIKESCORE') < 0.0:
+            call.filter(kevlar.vcf.VariantFilter.LikelihoodFail)
         yield call
 
 

kevlar/tests/test_alac.py

@@ -73,7 +73,7 @@ def test_pico_partitioned(capsys):
 
     out, err = capsys.readouterr()
     lines = out.strip().split('\n')
-    assert len(lines) == 33
+    assert len(lines) == 34
     lines = [l for l in lines if not l.startswith('#')]
     assert len(lines) == 10
     numnocalls = sum([1 for line in lines if '\t.\t.\t.\t.\t' in line])

kevlar/vcf.py

@@ -29,6 +29,7 @@ class VariantFilter(Enum):
     PassengerVariant = 3
     MateFail = 4
     PartitionScore = 5
+    LikelihoodFail = 6
 
 
 class FormattedList(list):
@@ -233,7 +234,10 @@ class VCFWriter(object):
             'call',
         VariantFilter.PartitionScore:
             'Expectation is 1 variant call per partition, so all call(s) with '
-            'suboptimal likelihood scores are filtered'
+            'suboptimal likelihood scores are filtered',
+        VariantFilter.LikelihoodFail:
+            'Variant calls with a likelihood score < 0.0 are unlikely to be'
+            'real',
     }
 
     info_metadata = {