Merge remote-tracking branch 'origin/hotfix'

docs/conf.py

@@ -70,7 +70,7 @@
 # The short X.Y version.
 version = '4.0'
 # The full version, including alpha/beta/rc tags.
-release = '4.0.4'
+release = '4.0.5'
 
 
 # The language for content autogenerated by Sphinx. Refer to documentation

docs/index.rst

@@ -56,8 +56,15 @@ New in Release |release|
 
 This is a bugfix release. It fixes the following problem(s):
 
-- This release makes various fixes to allow pVACtools to run with non-human
-  data.
+- In recent releases, users have noticed that at some point during pipeline
+  runs, predictions to MHCflurry would hang or get killed. We were able to
+  determine that the cause was related to
+  `PR 988 <https://github.com/griffithlab/pVACtools/pull/988>`_.
+  This PR originally updated calls to MHCflurry to happen by instantiating
+  their predictor within Python instead of calling it on the command line.
+  However, we suspect that this causes a substantial increase in memory usage
+  resulting in the observed behavior. This release reverts the change from PR
+  988.
 
 New in Version |version|
 ------------------------

docs/releases/4_0.rst

@@ -114,3 +114,18 @@ This is a bugfix release. It fixes the following problem(s):
 
 - This release makes various fixes to allow pVACtools to run with non-human
   data.
+
+New in Version 4.0.5
+--------------------
+
+This is a bugfix release. It fixes the following problem(s):
+
+- In recent releases, users have noticed that at some point during pipeline
+  runs, predictions to MHCflurry would hang or get killed. We were able to
+  determine that the cause was related to
+  `PR 988 <https://github.com/griffithlab/pVACtools/pull/988>`_.
+  This PR originally updated calls to MHCflurry to happen by instantiating
+  their predictor within Python instead of calling it on the command line.
+  However, we suspect that this causes a substantial increase in memory usage
+  resulting in the observed behavior. This release reverts the change from PR
+  988.

pvactools/lib/prediction_class.py

@@ -13,11 +13,6 @@
 from Bio import SeqIO
 import random
 import uuid
-from mhcflurry.downloads import get_default_class1_presentation_models_dir
-from mhcflurry.class1_presentation_predictor import Class1PresentationPredictor
-import numpy
-
-os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'
 
 class IEDB(metaclass=ABCMeta):
     @classmethod
@@ -323,28 +318,29 @@ def predict(self, input_file, allele, epitope_length, iedb_executable_path, iedb
 
         all_epitopes = list(set(all_epitopes))
         if len(all_epitopes) > 0:
-            models_dir = get_default_class1_presentation_models_dir(test_exists=True)
-            predictor = Class1PresentationPredictor.load(models_dir)
-            df = predictor.predict(
-                peptides=numpy.array(all_epitopes, dtype='object'),
-                n_flanks=None,
-                c_flanks=None,
-                alleles={allele: [allele]},
-                throw=True,
-                include_affinity_percentile=True,
-                verbose=0
-            )
+            tmp_output_file = tempfile.NamedTemporaryFile('r', dir=tmp_dir, delete=False)
+            arguments = ["mhcflurry-predict", "--alleles", allele, "--out", tmp_output_file.name, "--peptides"]
+            arguments.extend(all_epitopes)
+            stderr_fh = tempfile.NamedTemporaryFile('w', dir=tmp_dir, delete=False)
+            try:
+                response = run(arguments, check=True, stdout=DEVNULL, stderr=stderr_fh)
+            except:
+                stderr_fh.close()
+                with open(stderr_fh.name, 'r') as fh:
+                    err = fh.read()
+                os.unlink(stderr_fh.name)
+                raise Exception("An error occurred while calling MHCflurry:\n{}".format(err))
+            stderr_fh.close()
+            os.unlink(stderr_fh.name)
+            tmp_output_file.close()
+            df = pd.read_csv(tmp_output_file.name)
+            os.unlink(tmp_output_file.name)
             df.rename(columns={
-                'prediction': 'ic50',
-                'affinity': 'ic50',
-                'prediction_percentile': 'percentile',
-                'affinity_percentile': 'percentile',
-                'processing_score': 'mhcflurry_processing_score',
-                'presentation_score': 'mhcflurry_presentation_score',
-                'presentation_percentile': 'mhcflurry_presentation_percentile',
-                'best_allele': 'allele',
+                'mhcflurry_prediction': 'ic50',
+                'mhcflurry_affinity': 'ic50',
+                'mhcflurry_prediction_percentile': 'percentile',
+                'mhcflurry_affinity_percentile': 'percentile'
             }, inplace=True)
-            df.drop(labels='peptide_num', axis=1, inplace=True)
             for record in SeqIO.parse(input_file, "fasta"):
                 seq_num = record.id
                 peptide = str(record.seq)

setup.py

@@ -51,7 +51,7 @@
 
 setup(
     name="pvactools",
-    version="4.0.4",
+    version="4.0.5",
     packages=[
         "pvactools.tools",
         "pvactools.tools.pvacbind",

tests/test_call_iedb.py

@@ -96,8 +96,8 @@ def test_mhcflurry_method_generates_expected_files(self):
         ])
         if sys.platform == 'darwin':
             expected_output_file = os.path.join(self.test_data_dir, 'output_mhcflurry_osx.tsv')
-            expected_df = pd.read_csv(expected_output_file, sep="\t", index_col=[0,8,9])
-            actual_df = pd.read_csv(call_iedb_output_file.name, sep="\t", index_col=[0,8,9])
+            expected_df = pd.read_csv(expected_output_file, sep="\t", index_col=[1,7,8])
+            actual_df = pd.read_csv(call_iedb_output_file.name, sep="\t", index_col=[1,7,8])
             pd.testing.assert_frame_equal(expected_df, actual_df, check_like=True, check_exact=False, rtol=0.05)
 
     def test_mhcnuggetsi_method_generates_expected_files(self):