def check_netmhc_pan(program_name, fail_if_no_such_program=True):
try:
predictor = NetMHCpan(alleles=[DEFAULT_ALLELE],
program_name=program_name)
except FileNotFoundError:
if fail_if_no_such_program:
raise
print("Skipping because no such program: %s" % program_name)
return
binding_predictions = predictor.predict_subsequences(protein_sequence_dict,
peptide_lengths=[9])
assert len(binding_predictions) == 4, \
"Expected 4 binding predictions from %s" % (binding_predictions,)
for x in binding_predictions:
# recompute the peptide from the offset and starting sequence, and make sure it matches.
# this is currently wrong in netMHCpan-3.0 and we want to test our wrapper fix to that
offset = x.offset
length = x.length
seq_name = x.source_sequence_name
expected_peptide = protein_sequence_dict[seq_name][offset:offset +
length]
eq_(
expected_peptide, x.peptide,
"Peptide mismatch: expected %s but got %s in binding prediction '%s'"
% (
expected_peptide,
x.peptide,
x,
))
def test_netmhc_pan():
alleles = [normalize_allele_name(DEFAULT_ALLELE)]
pan_predictor = NetMHCpan(
alleles=alleles,
epitope_lengths=[9])
fasta_dictionary = {
"SMAD4-001": "ASIINFKELA",
"TP53-001": "ASILLLVFYW"
}
epitope_collection = pan_predictor.predict(
fasta_dictionary=fasta_dictionary)
assert len(epitope_collection) == 4, \
"Expected 4 epitopes from %s" % (epitope_collection,)
def run_wt_pep_single_netMHCpan():
size = 8
hlas = ["A*02:01"]
MUT_peps = {
"chr3_52533601_ACGCCGGGGCAGTGGG/A-L6-WT25-MUT25-M1": "NLYIRSSG",
}
for allele in hlas:
predictor = NetMHCpan(alleles=allele)
for _id, pep in MUT_peps.items():
mut_binding_prediction = \
predictor.predict_subsequences({_id : pep}, peptide_lengths = [size])
for binding_prediction in mut_binding_prediction:
print(binding_prediction.affinity)
def run_single_netMHCpan():
predictor = NetMHCpan(alleles=["A*02:01"])
wt_protein_sequence = {
"chr3_52533601_ACGCCGGGGCAGTGGG/A-L6-WT25-MUT25-M1": "NLYIQSSG",
}
binding_predictions = predictor.predict_subsequences(wt_protein_sequence,
peptide_lengths=[8])
for binding_prediction in binding_predictions:
print("Binder: %s" % (binding_prediction, ))
mut_protein_sequence = {
"chr3_52533601_ACGCCGGGGCAGTGGG/A-L6-WT25-MUT25-M1": "NLYIRSSG",
}
binding_predictions = predictor.predict_subsequences(mut_protein_sequence,
peptide_lengths=[8])
for binding_prediction in binding_predictions:
print("Binder: %s" % binding_prediction)
print("Binder allele: %s" % binding_prediction.allele)
def run_netMHCpan():
# Run NetMHCpan for alleles HLA-A*01:01 and HLA-A*02:01
predictor = NetMHCpan(alleles=["A*02:01", "hla-a0101"])
predictor = NetMHC(alleles=["A*02:01", "hla-a0101"])
# scan the short proteins 1L2Y and 1L3Y for epitopes
protein_sequences = {
"chr3_52533601_ACGCCGGGGCAGTGGG/A-L6-WT25-MUT25-M1":
"NLYIQSSGRPPPSWLKDGGP",
"chr8_127738959_G/A-1": "NLYIQSSGRPPPSWLKDGGP",
"1L2Y": "NLYIQWLKDGGPSSGRPPPS",
"1L3Y": "ECDTINCERYNGQVCGGPGRGLCFCGKCRCHPGFEGSACQA"
}
binding_predictions = predictor.predict_subsequences(protein_sequences,
peptide_lengths=[9])
# flatten binding predictions into a Pandas DataFrame
df = binding_predictions.to_dataframe()
# epitope collection is sorted by percentile rank
# of binding predictions
for binding_prediction in binding_predictions:
print("Binder: %s" % (binding_prediction, ))
start=32861682,
ref='G',
alt='A',
ensembl=ensembl_grch37)])
alleles = [
'A02:01',
'a0204',
'B*07:02',
'HLA-B14:02',
'HLA-C*07:02',
'hla-c07:01'
]
mhc_model = NetMHCpan(
alleles=alleles,
default_peptide_lengths=[9])
def test_epitope_prediction_without_padding():
output_without_padding = TopiaryPredictor(
mhc_model=mhc_model,
only_novel_epitopes=True).predict_from_variants(variants=variants)
# one prediction for each variant * number of alleles
strong_binders = output_without_padding[output_without_padding.affinity <= 500]
eq_(len(strong_binders), 5)
@raises(ValueError)
def test_epitope_prediction_with_invalid_padding():
TopiaryPredictor(
mhc_model=mhc_model,
# Downloaded from https://github.com/hammerlab/mhctools#Example
from mhctools import NetMHCpan
# Run NetMHCpan for alleles HLA-A*01:01 and HLA-A*02:01
predictor = NetMHCpan(alleles=["A*02:01", "hla-a0101"])
# scan the short proteins 1L2Y and 1L3Y for epitopes
protein_sequences = {
"1L2Y": "NLYIQWLKDGGPSSGRPPPS",
"1L3Y": "ECDTINCERYNGQVCGGPGRGLCFCGKCRCHPGFEGSACQA"
}
epitope_collection = predictor.predict(protein_sequences)
# flatten binding predictions into a Pandas DataFrame
df = epitope_collection.dataframe()
# epitope collection is sorted by percentile rank
# of binding predictions
strongest_predicted_binder = epitope_collection[0]
# should be NLYIQWLKDGGPSSGRPPPS
print strongest_predicted_binder.source_sequence
alt='T',
ensembl=ensembl_grch37),
Variant(contig=11,
start=32861682,
ref='G',
alt='A',
ensembl=ensembl_grch37)
])
alleles = [
'A02:01', 'a0204', 'B*07:02', 'HLA-B14:02', 'HLA-C*07:02', 'hla-c07:01'
]
epitope_lengths = [9]
mhc_model = NetMHCpan(alleles=alleles, epitope_lengths=epitope_lengths)
def test_epitope_prediction_without_padding():
output_without_padding = predict_epitopes_from_variants(
variants=variants,
mhc_model=mhc_model,
transcript_expression_dict=None,
only_novel_epitopes=True)
# one prediction for each variant * number of alleles
strong_binders = [
epitope_prediction for epitope_prediction in output_without_padding
if epitope_prediction.value <= 500.0
]
eq_(len(strong_binders), 4)
def predict_binding(hla_alleles, WT_peps, MUT_peps, size, bind_pred_software, extra_flag=""):
logging.info("Starting binding prediction for '%s', size: %i.", \
bind_pred_software + extra_flag, size)
logging.info("Alleles: '%s'.", hla_alleles)
hlas = hla_alleles.split(",")
data_netMHC = []
for allele in hlas:
if bind_pred_software.lower() == "netmhcpan" and extra_flag == "":
predictor = NetMHCpan(alleles = allele)
elif bind_pred_software.lower() == "netmhcpan" and extra_flag.lower() == "-ba":
predictor = NetMHCpan(alleles = allele, extra_flags=[extra_flag])
elif bind_pred_software.lower() == "netmhc":
try:
predictor = NetMHC(alleles = allele)
except UnsupportedAllele:
logging.warning("Unsupported Allele: '%s'", allele)
break
else:
raise Exception("Unknown binding prediction software: {}".format(bind_pred_software))
for var_id, MUT_pep in MUT_peps.items():
is_frameshift = helper.check_if_frameshift(var_id.split("-")[0])
mut_binding_prediction = \
predictor.predict_subsequences({var_id : MUT_pep}, peptide_lengths = [size])
# A sequence of the form "X" * size comes from the case of an empty_WT
if not is_frameshift and not WT_peps[var_id] == "X" * size:
wt_binding_prediction = \
predictor.predict_subsequences({var_id : WT_peps[var_id]}, peptide_lengths = [size])
for wt_bp, mut_bp in zip(wt_binding_prediction, mut_binding_prediction):
bind_level = "N/A"
if mut_bp.percentile_rank <= 2.0:
bind_level = "WB"
if mut_bp.percentile_rank <= 0.5:
bind_level = "SB"
if extra_flag == "-BA" or bind_pred_software.lower() == "netmhc":
dai = wt_bp.affinity - mut_bp.affinity
data_netMHC.append((var_id, mut_bp.percentile_rank, \
bind_level, WT_peps[var_id], MUT_pep, \
wt_bp.affinity, mut_bp.affinity, dai, \
allele, size, bind_pred_software + extra_flag))
else:
data_netMHC.append((var_id, mut_bp.percentile_rank, \
bind_level, WT_peps[var_id], MUT_pep, \
"N/A", "N/A", "N/A", \
allele, size, bind_pred_software + extra_flag))
else:
for mut_bp in mut_binding_prediction:
bind_level = "N/A"
if mut_bp.percentile_rank <= 2.0:
bind_level = "WB"
if mut_bp.percentile_rank <= 0.5:
bind_level = "SB"
if extra_flag == "-BA" or bind_pred_software.lower() == "netmhc":
data_netMHC.append((var_id, mut_bp.percentile_rank, \
bind_level, "N/A", MUT_pep, \
"N/A", mut_bp.affinity, "N/A", \
allele, size, bind_pred_software + extra_flag))
else:
data_netMHC.append((var_id, mut_bp.percentile_rank, \
bind_level, "N/A", MUT_pep, \
"N/A", "N/A", "N/A", \
allele, size, bind_pred_software + extra_flag))
write_netMHC_type_file(data_netMHC, size, bind_pred_software, extra_flag)
logging.info("Finished binding prediction for '%s', size: %i.", \
bind_pred_software + extra_flag, size)
