avoid error with pandas >=0.22.0 when all rates are missing

This fixes a bug where genes can gen p=0.0 when no rates data is available for a gene.
Since pandas 0.22.0, whenever summing a list where all values are NA, pandas will by
by default give sum=0.0, which if all gene rates are missing, ends up giving an
expected number of mutation=0, and this results in a P=0. This change reverts to the
previous pandas behaviour, where summing a list of NAs results in NA.

mupit/mutation_rates.py

@@ -119,9 +119,14 @@ def get_expected_mutations(rates, male, female):
 
     expected = rates[["hgnc", "chrom"]].copy()
 
+    # account for how different pandas versions sum series with only NA
+    kwargs = {}
+    if pandas.__version__ >= '0.22.0':
+        kwargs = {'min_count': 1}
+
     # get the number of expected mutations, given the number of transmissions
     expected["lof_indel"] = rates["frameshift"] * autosomal
-    expected["lof_snv"] = (rates[["non", "splice_site"]].sum(axis=1, skipna=True)) * autosomal
+    expected["lof_snv"] = (rates[["non", "splice_site"]].sum(axis=1, skipna=True, **kwargs)) * autosomal
     expected["missense_indel"] = (rates["frameshift"] / 9) * autosomal
     expected["missense_snv"] = rates["mis"] * autosomal
     expected["synonymous_snv"] = rates["syn"] * autosomal