def read_hla_file(path, permissive_parsing=True):
"""
Read in HLA alleles and normalize them, returning a list of HLA allele
names.
"""
assert path.endswith(".hla"), \
"Expected HLA file %s to end with suffix .hla" % path
logging.info("Reading HLA file %s", path)
alleles = []
with open(path, 'r') as f:
contents = f.read()
for line in contents.split("\n"):
for raw_allele in line.split(","):
if permissive_parsing:
# get rid of surrounding whitespace
raw_allele = raw_allele.strip()
# sometimes we get extra columns with scores,
# ignore those
raw_allele = raw_allele.split(" ")[0]
raw_allele = raw_allele.split("\t")[0]
raw_allele = raw_allele.split("'")[0]
if len(raw_allele) > 0:
alleles.append(normalize_hla_allele_name(raw_allele))
return alleles
def read_hla_file(path, permissive_parsing=True):
"""
Read in HLA alleles and normalize them, returning a list of HLA allele
names.
"""
assert path.endswith(".hla"), \
"Expected HLA file %s to end with suffix .hla" % path
logging.info("Reading HLA file %s", path)
alleles = []
with open(path, 'r') as f:
contents = f.read()
for line in contents.split("\n"):
for raw_allele in line.split(","):
if permissive_parsing:
# get rid of surrounding whitespace
raw_allele = raw_allele.strip()
# sometimes we get extra columns with scores,
# ignore those
raw_allele = raw_allele.split(" ")[0]
raw_allele = raw_allele.split("\t")[0]
raw_allele = raw_allele.split("'")[0]
if len(raw_allele) > 0:
alleles.append(
normalize_hla_allele_name(
raw_allele))
return alleles
def __init__(
self,
hla_alleles,
netmhc_command = "netMHCcons"):
self.netmhc_command = netmhc_command
try:
subprocess.check_output([self.netmhc_command],
stderr=subprocess.STDOUT)
except:
assert False, "Failed to run %s" % self.netmhc_command
# normalize alleles and keep only unique names
normalized_alleles = {
normalize_hla_allele_name(allele.strip().upper()).replace("*", "")
for allele in hla_alleles
}
self.alleles = []
# try running "netMHCcons -a" with each allele name
# and check if it gives you back a "wrong format" error
for allele in normalized_alleles:
try:
subprocess.check_output(
[self.netmhc_command, '-a', allele],
stderr=subprocess.STDOUT)
except subprocess.CalledProcessError, e:
if "allele" in e.output and "wrong format" in e.output:
logging.warning(
"Allele %s not recognized by NetMHCcons", allele)
continue
except:
def __init__(
self,
hla_alleles,
netmhc_command = "netMHCpan"):
self.netmhc_command = netmhc_command
try:
run_command([self.netmhc_command])
except:
assert False, "Failed to run %s" % self.netmhc_command
try:
valid_alleles_str = check_output([self.netmhc_command, "-listMHC"])
assert len(valid_alleles_str) > 0, \
"%s returned empty allele list" % self.self.netmhc_command
valid_alleles = set([])
for line in valid_alleles_str.split("\n"):
if not line.startswith("#"):
valid_alleles.add(line)
except:
logging.warning("Failed to run %s -listMHC", self.netmhc_command)
valid_alleles = None
self.alleles = []
for allele in hla_alleles:
allele = normalize_hla_allele_name(allele.strip().upper())
# for some reason netMHCpan drop the "*" in names
# such as "HLA-A*03:01" becomes "HLA-A03:01"
if valid_alleles and allele.replace("*", "") not in valid_alleles:
print "Skipping %s (not available in NetMHCpan)" % allele
else:
self.alleles.append(allele)
# don't run the MHC predictor twice for homozygous alleles,
# only run it for unique alleles
self.alleles = set(self.alleles)
def run_pipeline(patient_id, score_epitopes):
"""Run the pipeline for this patient, and save the output to the DB as a
Run."""
hla_types = HLAType.query.with_entities(HLAType.allele,
HLAType.mhc_class).filter_by(patient_id=patient_id).all()
peptide_length = 31
alleles = [normalize_hla_allele_name(
allele) for allele, mhc_class in hla_types]
vcf_df = get_vcf_df(patient_id)
transcripts_df, vcf_df, variant_report = expand_transcripts(
vcf_df,
patient_id,
min_peptide_length = peptide_length,
max_peptide_length = peptide_length)
scored_epitopes = score_epitopes(transcripts_df, alleles)
imm = ImmunogenicityPredictor(alleles=alleles)
scored_epitopes = imm.predict(scored_epitopes)
# TODO(tavi) Make this expansion more robust. It breaks the IEDB predictor,
# for example.
short_transcripts_df = transcripts_df[['chr', 'pos', 'ref',
'alt', 'TranscriptId']]
scored_epitopes = merge(scored_epitopes, short_transcripts_df,
on='TranscriptId', how='left')
peptides = group_epitopes_dataframe(
scored_epitopes, use_transcript_name = True)
run = Run(patient_id=patient_id, output=dumps(peptides))
db.session.add(run)
def __init__(self, hla_alleles, netmhc_command="netMHCpan"):
self.netmhc_command = netmhc_command
try:
run_command([self.netmhc_command])
except:
assert False, "Failed to run %s" % self.netmhc_command
try:
valid_alleles_str = check_output([self.netmhc_command, "-listMHC"])
assert len(valid_alleles_str) > 0, \
"%s returned empty allele list" % self.self.netmhc_command
valid_alleles = set([])
for line in valid_alleles_str.split("\n"):
if not line.startswith("#"):
valid_alleles.add(line)
except:
logging.warning("Failed to run %s -listMHC", self.netmhc_command)
valid_alleles = None
self.alleles = []
for allele in hla_alleles:
allele = normalize_hla_allele_name(allele.strip().upper())
# for some reason netMHCpan drop the "*" in names
# such as "HLA-A*03:01" becomes "HLA-A03:01"
if valid_alleles and allele.replace("*", "") not in valid_alleles:
print "Skipping %s (not available in NetMHCpan)" % allele
else:
self.alleles.append(allele)
# don't run the MHC predictor twice for homozygous alleles,
# only run it for unique alleles
self.alleles = set(self.alleles)
def create_binding_result_row(mutation_entry,
allele,
pos,
epitope,
log_ic50,
ic50,
rank,
mutation_window_size=None):
# if we have a bad IC50 score we might still get a salvageable
# log of the score. Strangely, this is necessary sometimes!
if invalid_binding_score(ic50):
ic50 = 50000**(-log_ic50 + 1)
# if IC50 is still NaN or otherwise invalid, abort
if invalid_binding_score(ic50):
logging.warn("Invalid IC50 value %0.4f for %s w/ allele %s", ic50,
epitope, allele)
return None
if invalid_binding_score(rank) or rank > 100:
logging.warn("Invalid percentile rank %s for %s w/ allele %s", rank,
epitope, allele)
return None
if mutation_window_size:
# if we clipped parts of the amino acid sequence which don't
# overlap mutations then we have to offset epitope positions by
# however much was removed from the beginning of the sequence
original_start = max(
0, mutation_entry.MutationStart - mutation_window_size)
pos += original_start
# keep track of original genetic variant that
# gave rise to this epitope
new_row = {}
# fields shared by all epitopes from this sequence
new_row['chr'] = mutation_entry.chr
new_row['pos'] = mutation_entry.pos
new_row['ref'] = mutation_entry.ref
new_row['alt'] = mutation_entry.alt
new_row['SourceSequence'] = mutation_entry.SourceSequence
new_row['MutationStart'] = mutation_entry.MutationStart
new_row['MutationEnd'] = mutation_entry.MutationEnd
new_row['GeneInfo'] = mutation_entry.GeneInfo
new_row['Gene'] = mutation_entry.Gene
new_row["GeneMutationInfo"] = mutation_entry.GeneMutationInfo
new_row['PeptideMutationInfo'] = mutation_entry.PeptideMutationInfo
new_row['TranscriptId'] = mutation_entry.TranscriptId
# fields specific to this epitope
new_row['Allele'] = normalize_hla_allele_name(allele)
new_row['EpitopeStart'] = pos
new_row['EpitopeEnd'] = pos + len(epitope)
new_row['Epitope'] = epitope
new_row[IC50_FIELD_NAME] = ic50
new_row[PERCENTILE_RANK_FIELD_NAME] = rank
return new_row
def create_hla_types(file, patient_id):
filename = secure_filename(file.filename)
filepath = join(app.config['UPLOAD_FOLDER'], filename)
file.save(filepath)
alleles = read_hla_file(filepath)
hla_types = []
for allele in alleles:
allele_normalized = normalize_hla_allele_name(allele)
mhc_class = mhc_class_from_normalized_allele_name(allele_normalized)
hla_type = HLAType(patient_id=patient_id, allele=allele_normalized,
mhc_class=mhc_class)
hla_types.append(hla_type)
return hla_types
def create_binding_result_row(
mutation_entry,
allele,
pos,
epitope,
log_ic50,
ic50,
rank,
mutation_window_size = None):
# if we have a bad IC50 score we might still get a salvageable
# log of the score. Strangely, this is necessary sometimes!
if invalid_binding_score(ic50):
ic50 = 50000 ** (-log_ic50 + 1)
# if IC50 is still NaN or otherwise invalid, abort
if invalid_binding_score(ic50):
logging.warn(
"Invalid IC50 value %0.4f for %s w/ allele %s",
ic50,
epitope,
allele)
return None
if invalid_binding_score(rank) or rank > 100:
logging.warn(
"Invalid percentile rank %s for %s w/ allele %s",
rank, epitope, allele)
return None
if mutation_window_size:
# if we clipped parts of the amino acid sequence which don't
# overlap mutations then we have to offset epitope positions by
# however much was removed from the beginning of the sequence
original_start = max(
0,
mutation_entry.MutationStart - mutation_window_size)
pos += original_start
# keep track of original genetic variant that
# gave rise to this epitope
new_row = {}
# fields shared by all epitopes from this sequence
new_row['chr'] = mutation_entry.chr
new_row['pos'] = mutation_entry.pos
new_row['ref'] = mutation_entry.ref
new_row['alt'] = mutation_entry.alt
new_row['SourceSequence'] = mutation_entry.SourceSequence
new_row['MutationStart'] = mutation_entry.MutationStart
new_row['MutationEnd'] = mutation_entry.MutationEnd
new_row['GeneInfo'] = mutation_entry.GeneInfo
new_row['Gene'] = mutation_entry.Gene
new_row["GeneMutationInfo"] = mutation_entry.GeneMutationInfo
new_row['PeptideMutationInfo'] = mutation_entry.PeptideMutationInfo
new_row['TranscriptId'] = mutation_entry.TranscriptId
# fields specific to this epitope
new_row['Allele'] = normalize_hla_allele_name(allele)
new_row['EpitopeStart'] = pos
new_row['EpitopeEnd'] = pos + len(epitope)
new_row['Epitope'] = epitope
new_row[IC50_FIELD_NAME] = ic50
new_row[PERCENTILE_RANK_FIELD_NAME] = rank
return new_row
predictor = ConsensusBindingPredictor(alleles)
return predictor.predict(mutated_regions)
else:
predictor = PanBindingPredictor(alleles)
return predictor.predict(mutated_regions)
if __name__ == '__main__':
args = parser.parse_args()
init_logging(args.quiet)
peptide_length = int(args.vaccine_peptide_length)
# get rid of gene descriptions if they're in the dataframe
if args.hla_file:
alleles = [normalize_hla_allele_name(l) for l in open(args.hla_file)]
elif args.hla:
alleles = [normalize_hla_allele_name(l) for l in args.hla.split(",")]
else:
alleles = [normalize_hla_allele_name(DEFAULT_ALLELE)]
# stack up the dataframes and later concatenate in case we
# want both commandline strings (for weird mutations like translocations)
# and files
mutated_region_dfs = []
if args.string:
df = load_comma_string(args.string)
mutated_region_dfs.append(df)
# loop over all the input files and
