From 5a002dba1628a6546255a6a1c0b8ab4b2e5dbdad Mon Sep 17 00:00:00 2001
From: Tim O'Donnell <timodonnell@gmail.com>
Date: Mon, 5 Feb 2018 23:13:50 -0500
Subject: [PATCH] fixes for pan
---
mhcflurry/allele_encoding.py | 9 ++--
mhcflurry/class1_affinity_predictor.py | 69 ++++++++++++++++----------
mhcflurry/class1_neural_network.py | 14 ++----
3 files changed, 52 insertions(+), 40 deletions(-)
diff --git a/mhcflurry/allele_encoding.py b/mhcflurry/allele_encoding.py
index a2bc8715..d95d2ce2 100644
--- a/mhcflurry/allele_encoding.py
+++ b/mhcflurry/allele_encoding.py
@@ -37,7 +37,7 @@ class AlleleEncoding(object):
self.encoding_cache = {}
- def fixed_length_sequences(self, vector_encoding_name):
+ def fixed_length_vector_encoded_sequences(self, vector_encoding_name):
"""
Encode alleles.
@@ -57,14 +57,13 @@ class AlleleEncoding(object):
"fixed_length_vector_encoding",
vector_encoding_name)
if cache_key not in self.encoding_cache:
- index_encoded_matrix = amino_acid.fixed_vectors_encoding(
+ index_encoded_matrix = amino_acid.index_encoding(
self.fixed_length_sequences.values,
- amino_acid.COMMON_AMINO_ACIDS_WITH_UNKNOWN)
+ amino_acid.AMINO_ACID_INDEX)
vector_encoded = amino_acid.fixed_vectors_encoding(
index_encoded_matrix,
amino_acid.ENCODING_DATA_FRAMES[vector_encoding_name])
- vector_encoded_df = pandas.DataFrame(vector_encoded)
- result = vector_encoded_df.iloc[self.indices]
+ result = vector_encoded[self.indices]
self.encoding_cache[cache_key] = result
return self.encoding_cache[cache_key]
diff --git a/mhcflurry/class1_affinity_predictor.py b/mhcflurry/class1_affinity_predictor.py
index fad3aa11..5f82e982 100644
--- a/mhcflurry/class1_affinity_predictor.py
+++ b/mhcflurry/class1_affinity_predictor.py
@@ -24,6 +24,7 @@ from .percent_rank_transform import PercentRankTransform
from .regression_target import to_ic50
from .version import __version__
from .ensemble_centrality import CENTRALITY_MEASURES
+from .allele_encoding import AlleleEncoding
class Class1AffinityPredictor(object):
@@ -39,7 +40,7 @@ class Class1AffinityPredictor(object):
self,
allele_to_allele_specific_models=None,
class1_pan_allele_models=None,
- allele_to_pseudosequence=None,
+ allele_to_fixed_length_sequence=None,
manifest_df=None,
allele_to_percent_rank_transform=None):
"""
@@ -51,7 +52,7 @@ class Class1AffinityPredictor(object):
class1_pan_allele_models : list of `Class1NeuralNetwork`
Ensemble of pan-allele models.
- allele_to_pseudosequence : dict of string -> string
+ allele_to_fixed_length_sequence : dict of string -> string
Required only if class1_pan_allele_models is specified.
manifest_df : `pandas.DataFrame`, optional
@@ -70,11 +71,11 @@ class Class1AffinityPredictor(object):
class1_pan_allele_models = []
if class1_pan_allele_models:
- assert allele_to_pseudosequence, "Pseudosequences required"
+ assert allele_to_fixed_length_sequence, "Allele sequences required"
self.allele_to_allele_specific_models = allele_to_allele_specific_models
self.class1_pan_allele_models = class1_pan_allele_models
- self.allele_to_pseudosequence = allele_to_pseudosequence
+ self.allele_to_fixed_length_sequence = allele_to_fixed_length_sequence
if manifest_df is None:
rows = []
@@ -140,7 +141,7 @@ class Class1AffinityPredictor(object):
allele_to_allele_specific_models = collections.defaultdict(list)
class1_pan_allele_models = []
- allele_to_pseudosequence = predictors[0].allele_to_pseudosequence
+ allele_to_fixed_length_sequence = predictors[0].allele_to_fixed_length_sequence
for predictor in predictors:
for (allele, networks) in (
@@ -152,7 +153,7 @@ class Class1AffinityPredictor(object):
return Class1AffinityPredictor(
allele_to_allele_specific_models=allele_to_allele_specific_models,
class1_pan_allele_models=class1_pan_allele_models,
- allele_to_pseudosequence=allele_to_pseudosequence
+ allele_to_fixed_length_sequence=allele_to_fixed_length_sequence
)
@property
@@ -165,8 +166,8 @@ class Class1AffinityPredictor(object):
list of string
"""
result = set(self.allele_to_allele_specific_models)
- if self.allele_to_pseudosequence:
- result = result.union(self.allele_to_pseudosequence)
+ if self.allele_to_fixed_length_sequence:
+ result = result.union(self.allele_to_fixed_length_sequence)
return sorted(result)
@property
@@ -196,7 +197,7 @@ class Class1AffinityPredictor(object):
The serialization format consists of a file called "manifest.csv" with
the configurations of each Class1NeuralNetwork, along with per-network
files giving the model weights. If there are pan-allele predictors in
- the ensemble, the allele pseudosequences are also stored in the
+ the ensemble, the allele sequences are also stored in the
directory. There is also a small file "index.txt" with basic metadata:
when the models were trained, by whom, on what host.
@@ -250,6 +251,15 @@ class Class1AffinityPredictor(object):
pandas.DataFrame(rows).to_csv(
info_path, sep="\t", header=False, index=False)
+ if self.allele_to_fixed_length_sequence is not None:
+ allele_to_sequence_df = pandas.DataFrame(
+ list(self.allele_to_fixed_length_sequence.items()),
+ columns=['allele', 'sequence']
+ )
+ allele_to_sequence_df.to_csv(
+ join(models_dir, "allele_sequences.csv"), index=False)
+ logging.info("Wrote: %s" % join(models_dir, "allele_sequences.csv"))
+
if self.allele_to_percent_rank_transform:
percent_ranks_df = None
for (allele, transform) in self.allele_to_percent_rank_transform.items():
@@ -310,10 +320,10 @@ class Class1AffinityPredictor(object):
manifest_df["model"] = all_models
- pseudosequences = None
- if exists(join(models_dir, "pseudosequences.csv")):
- pseudosequences = pandas.read_csv(
- join(models_dir, "pseudosequences.csv"),
+ allele_to_fixed_length_sequence = None
+ if exists(join(models_dir, "allele_sequences.csv")):
+ allele_to_fixed_length_sequence = pandas.read_csv(
+ join(models_dir, "allele_sequences.csv"),
index_col="allele").to_dict()
allele_to_percent_rank_transform = {}
@@ -325,10 +335,10 @@ class Class1AffinityPredictor(object):
PercentRankTransform.from_series(percent_ranks_df[allele]))
logging.info(
- "Loaded %d class1 pan allele predictors, %d pseudosequences, "
+ "Loaded %d class1 pan allele predictors, %d allele sequences, "
"%d percent rank distributions, and %d allele specific models: %s" % (
len(class1_pan_allele_models),
- len(pseudosequences) if pseudosequences else 0,
+ len(allele_to_fixed_length_sequence) if allele_to_fixed_length_sequence else 0,
len(allele_to_percent_rank_transform),
sum(len(v) for v in allele_to_allele_specific_models.values()),
", ".join(
@@ -339,7 +349,7 @@ class Class1AffinityPredictor(object):
result = Class1AffinityPredictor(
allele_to_allele_specific_models=allele_to_allele_specific_models,
class1_pan_allele_models=class1_pan_allele_models,
- allele_to_pseudosequence=pseudosequences,
+ allele_to_fixed_length_sequence=allele_to_fixed_length_sequence,
manifest_df=manifest_df,
allele_to_percent_rank_transform=allele_to_percent_rank_transform,
)
@@ -516,6 +526,7 @@ class Class1AffinityPredictor(object):
alleles,
peptides,
affinities,
+ inequalities,
models_dir_for_save=None,
verbose=1,
progress_preamble=""):
@@ -534,12 +545,15 @@ class Class1AffinityPredictor(object):
architecture_hyperparameters : dict
alleles : list of string
- Allele names (not pseudosequences) corresponding to each peptide
+ Allele names (not sequences) corresponding to each peptide
peptides : `EncodableSequences` or list of string
affinities : list of float
nM affinities
+
+ inequalities : list of string, each element one of ">", "<", or "="
+ See Class1NeuralNetwork.fit for details.
models_dir_for_save : string, optional
If specified, the Class1AffinityPredictor is (incrementally) written
@@ -557,7 +571,9 @@ class Class1AffinityPredictor(object):
"""
alleles = pandas.Series(alleles).map(mhcnames.normalize_allele_name)
- allele_pseudosequences = alleles.map(self.allele_to_pseudosequence)
+ allele_encoding = AlleleEncoding(
+ alleles,
+ allele_to_fixed_length_sequence=self.allele_to_fixed_length_sequence)
encodable_peptides = EncodableSequences.create(peptides)
models = []
@@ -567,7 +583,8 @@ class Class1AffinityPredictor(object):
model.fit(
encodable_peptides,
affinities,
- allele_pseudosequences=allele_pseudosequences,
+ inequalities=inequalities,
+ allele_encoding=allele_encoding,
verbose=verbose,
progress_preamble=progress_preamble)
@@ -777,27 +794,27 @@ class Class1AffinityPredictor(object):
unsupported_alleles = [
allele for allele in
df.normalized_allele.unique()
- if allele not in self.allele_to_pseudosequence
+ if allele not in self.allele_to_fixed_length_sequence
]
if unsupported_alleles:
msg = (
- "No pseudosequences for allele(s): %s.\n"
+ "No sequences for allele(s): %s.\n"
"Supported alleles: %s" % (
" ".join(unsupported_alleles),
- " ".join(sorted(self.allele_to_pseudosequence))))
+ " ".join(sorted(self.allele_to_fixed_length_sequence))))
logging.warning(msg)
if throw:
raise ValueError(msg)
mask = df.supported_peptide_length
if mask.sum() > 0:
- masked_allele_pseudosequences = (
- df.ix[mask].normalized_allele.map(
- self.allele_to_pseudosequence))
+ masked_allele_encoding = AlleleEncoding(
+ df.loc[mask].normalized_allele,
+ allele_to_fixed_length_sequence=self.allele_to_fixed_length_sequence)
masked_peptides = peptides.sequences[mask]
for (i, model) in enumerate(self.class1_pan_allele_models):
df.loc[mask, "model_pan_%d" % i] = model.predict(
masked_peptides,
- allele_pseudosequences=masked_allele_pseudosequences)
+ allele_encoding=masked_allele_encoding)
if self.allele_to_allele_specific_models:
query_alleles = df.normalized_allele.unique()
diff --git a/mhcflurry/class1_neural_network.py b/mhcflurry/class1_neural_network.py
index 5cf87569..c48b051f 100644
--- a/mhcflurry/class1_neural_network.py
+++ b/mhcflurry/class1_neural_network.py
@@ -370,7 +370,7 @@ class Class1NeuralNetwork(object):
-------
numpy.array
"""
- return allele_encoding.fixed_length_sequences("BLOSUM62")
+ return allele_encoding.fixed_length_vector_encoded_sequences("BLOSUM62")
def fit(
self,
@@ -693,9 +693,8 @@ class Class1NeuralNetwork(object):
'peptide': self.peptides_to_network_input(peptides)
}
if allele_encoding is not None:
- pseudosequences_input = self.pseudosequence_to_network_input(
- allele_pseudosequences)
- x_dict['pseudosequence'] = pseudosequences_input
+ allele_input = self.allele_encoding_to_network_input(allele_encoding)
+ x_dict['allele'] = allele_input
network = self.network(borrow=True)
raw_predictions = network.predict(x_dict, batch_size=batch_size)
@@ -787,8 +786,8 @@ class Class1NeuralNetwork(object):
if allele_encoding_dims:
allele_input = Input(
shape=allele_encoding_dims,
- dtype='int32',
- name='peptide')
+ dtype='float32',
+ name='allele')
inputs.append(allele_input)
allele_embedding_layer = Flatten(name="allele_flat")(allele_input)
@@ -807,9 +806,6 @@ class Class1NeuralNetwork(object):
current_layer = keras.layers.multiply([
current_layer, allele_embedding_layer
], name="allele_peptide_merged")
-
- current_layer = keras.layers.concatenate(
- [current_layer, allele_embedding_layer], name="concatenated_0")
else:
raise ValueError(
"Unsupported peptide_allele_encoding_merge_method: %s"
--
GitLab