fixed sequence in MAGE test, added Titin test

.travis.yml

@@ -47,9 +47,9 @@ script:
   - >
       mhcflurry-train-class1-allele-specific-models.py
       --alleles HLA-A0101 HLA-A0201
-      --embedding-size 4
-      --hidden-layer-size 4
-      --training-epochs 5
+      --embedding-size 10
+      --hidden-layer-size 10
+      --training-epochs 20
   # run tests
   - nosetests test --with-coverage --cover-package=mhcflurry  && ./lint.sh
 after_success:

mhcflurry/class1_allele_specific_hyperparameters.py

@@ -16,8 +16,8 @@ N_EPOCHS = 250
 ACTIVATION = "tanh"
 INITIALIZATION_METHOD = "lecun_uniform"
 EMBEDDING_DIM = 32
-HIDDEN_LAYER_SIZE = 200
-DROPOUT_PROBABILITY = 0.25
+HIDDEN_LAYER_SIZE = 100
+DROPOUT_PROBABILITY = 0.1
 MAX_IC50 = 50000.0
 LEARNING_RATE = 0.001
 

mhcflurry/data.py

@@ -28,6 +28,7 @@ from .peptide_encoding import (
     indices_to_hotshot_encoding,
     fixed_length_from_many_peptides
 )
+from .class1_allele_specific_hyperparameters import MAX_IC50
 
 index_encoding = common_amino_acids.index_encoding
 
@@ -67,7 +68,7 @@ def _infer_csv_separator(filename):
 def load_dataframe(
         filename,
         peptide_length=None,
-        max_ic50=50000.0,
+        max_ic50=MAX_IC50,
         sep=None,
         species_column_name="species",
         allele_column_name="mhc",
@@ -148,7 +149,7 @@ def load_dataframe(
 
 def load_allele_dicts(
         filename,
-        max_ic50=50000.0,
+        max_ic50=MAX_IC50,
         regression_output=False,
         sep=None,
         species_column_name="species",
@@ -251,7 +252,7 @@ def load_allele_datasets(
         filename,
         peptide_length=9,
         use_multiple_peptide_lengths=True,
-        max_ic50=50000.0,
+        max_ic50=MAX_IC50,
         flatten_binary_encoding=True,
         sep=None,
         species_column_name="species",

script/mhcflurry-train-class1-allele-specific-models.py

@@ -128,9 +128,10 @@ if __name__ == "__main__":
         n_allele = len(allele_data.Y)
         assert len(X) == n_allele
 
-        print("\n=== Training predictor for %s: %d samples" % (
+        print("\n=== Training predictor for %s: %d samples, %d unique" % (
             allele_name,
-            n_allele))
+            n_allele,
+            len(set(allele_data.original_peptides))))
 
         model = Class1BindingPredictor.from_hyperparameters(
             name=allele_name,

test/test_known_class1_epitopes.py

 
 from mhcflurry import Class1BindingPredictor
 
-def test_MAGE_epitope():
-    # Test the A1 MAGE epitope ESDPIVAQY from
+def test_A1_Titin_epitope():
+    # Test the A1 Titin epitope ESDPIVAQY from
     #   Identification of a Titin-Derived HLA-A1-Presented Peptide
     #   as a Cross-Reactive Target for Engineered MAGE A3-Directed
     #   T Cells
@@ -13,7 +13,19 @@ def test_MAGE_epitope():
     ic50 = ic50s[0]
     assert ic50 <= 500
 
-def test_HIV_epitope():
+def test_A1_MAGE_epitope():
+    # Test the A1 MAGE epitope EVDPIGHLY from
+    #   Identification of a Titin-Derived HLA-A1-Presented Peptide
+    #   as a Cross-Reactive Target for Engineered MAGE A3-Directed
+    #   T Cells
+    model = Class1BindingPredictor.from_allele_name("HLA-A*01:01")
+    ic50s = model.predict_peptides_ic50(["EVDPIGHLY"])
+    print(ic50s)
+    assert len(ic50s) == 1
+    ic50 = ic50s[0]
+    assert ic50 <= 500
+
+def test_A2_HIV_epitope():
     # Test the A2 HIV epitope SLYNTVATL from
     #    The HIV-1 HLA-A2-SLYNTVATL Is a Help-Independent CTL Epitope
     model = Class1BindingPredictor.from_allele_name("HLA-A*02:01")
@@ -25,5 +37,6 @@ def test_HIV_epitope():
 
 
 if __name__ == "__main__":
-    test_MAGE_epitope()
-    test_HIV_epitope()
+    test_A1_Titin_epitope()
+    test_A1_MAGE_epitope()
+    test_A2_HIV_epitope()