def parse_external_result(self, output):
"""
Searches within the defined dir _file for the newest dir and reads
the prediction file from there
:param str output: The path to the output dir
:return: The predicted HLA genotype
:rtype: list(:class:`~Fred2.Core.Allele.Allele`)
"""
alleles = []
if os.path.isdir(output):
_file = os.path.join(output, ".typing.txt")
else:
_file = output + ".typing.txt"
typing = False
with open(_file, "r") as f:
for l in f:
if typing and l.strip() != "":
a1, a2, _ = l.split()
alleles.append(Allele("HLA-" +
":".join(a1.split(":")[:2])))
alleles.append(Allele("HLA-" +
":".join(a2.split(":")[:2])))
if "------------ Inferred Allelic Pairs -------------" in l:
typing = True
return alleles
def setUp(self):
self.peptides_mhcI = [Peptide("SYFPEITHI"), Peptide("IHTIEPFYS")]
self.peptides_mhcII = [Peptide("AAAAAASYFPEITHI"), Peptide("IHTIEPFYSAAAAAA")]
self.mhcI = [Allele("HLA-B*07:02"), Allele("HLA-A*02:01")]
self.mhcII = [Allele("HLA-DRB1*07:01"), Allele("HLA-DRB1*15:01")]
self.mhcII_combined_alleles = [CombinedAllele("DPA1*01:03-DPB1*01:01"), CombinedAllele("DQA1*06:02-DQB1*06:31")]
self.transcript = Transcript("")
def test_allele_factory(self):
a = Allele("HLA-DPA1*01:03-DPB1*01:01", prob=1)
b = Allele("HLA-A*02:01", prob=2)
self.assertIsInstance(a, CombinedAllele)
self.assertEqual(a.prob, 1)
self.assertIsInstance(b, Allele)
self.assertEqual(b.prob, 2)
def read_hla_input(hla_file):
"""
reads in the hla file
header are defined as:
A1 - first HLA-A allele in 4-digit notation
A2 - second HLA-A allele in 4-digit notation
B1 - first HLA-B allele in 4-digit notation
B2 - second HLA-B allele in 4-digit notation
C1 - first HLA-C allele in 4-digit notation
C2 - second HLA-C allele in 4-digit notation
A_expression - expression of HLA A gene
B_expression - expression of HLA B gene
C_expression - expression of HLA C gene
:param hla_file:
:return: list(Allele)
"""
alleles = []
with open(hla_file, "rU") as f:
reader = csv.DictReader(f, delimiter='\t')
for row in reader:
for n, hla in itr.product([1,2],["A","B","C"]):
a = Allele(row[hla+str(n)])
a.log_metadata("abundance",float(row[hla+"_expression"]))
alleles.append(a)
return alleles
def parse_external_result(self, output):
"""
Searches within the defined dir _file for the newest dir and reads
the prediction file from there
:param str output: The path to the output dir
:return: The predicted HLA genotype
:rtype: list(:class:`~Fred2.Core.Allele.Allele`)
"""
alleles = []
try:
with open(os.path.join(output, "winner.hla.txt"), "r") as f:
for l in f:
try:
_, a1, a2 = l.replace("-n",
"").replace("-e",
"").strip().split()
a1 = a1.split("_")
a2 = a2.split("_")
alleles.extend([
Allele("HLA-" + a1[1].upper() + "*" + a1[2] + ":" +
a1[3]),
Allele("HLA-" + a2[1].upper() + "*" + a2[2] + ":" +
a2[3])
])
except ValueError:
IOError(
"Output format seems incorrect:\n{line}\n. Please check if Polysolver ran correctly."
.format(lines=l))
return alleles
except IOError:
raise IOError(
"File {out} could not be found. Please check your specified output folder"
.format(out=os.path.join(output, "winner.hla.txt")))
def read_hla_input(hla_file):
"""
reads in the hla file
header are defined as:
A1 - first HLA-A allele in 4-digit notation
A2 - second HLA-A allele in 4-digit notation
B1 - first HLA-B allele in 4-digit notation
B2 - second HLA-B allele in 4-digit notation
C1 - first HLA-C allele in 4-digit notation
C2 - second HLA-C allele in 4-digit notation
A_expression - expression of HLA A gene
B_expression - expression of HLA B gene
C_expression - expression of HLA C gene
:param hla_file:
:return: list(Allele)
"""
alleles = []
with open(hla_file, "rU") as f:
reader = csv.DictReader(f, delimiter='\t')
for row in reader:
for n, hla in itr.product([1, 2], ["A", "B", "C"]):
a = Allele(row[hla + str(n)])
a.log_metadata("abundance", float(row[hla + "_expression"]))
alleles.append(a)
return alleles
def affinities_from_csv(bindings_file,
allele_data=None,
peptide_coverage=None):
''' Loads binding affinities from a csv file. Optionally, augments alleles with probability
and peptides with protein coverage. Discards all peptides for which coverage is not provided.
'''
df = pd.read_csv(bindings_file)
df['Seq'] = df.Seq.apply(Peptide)
if peptide_coverage is not None:
keep = []
for pep in df.Seq:
if pep not in peptide_coverage:
keep.append(False)
continue
keep.append(True)
for prot in peptide_coverage[str(pep)]:
pep.proteins[prot] = prot
df = df[keep]
df = df.set_index(['Seq', 'Method'])
if allele_data is not None:
df.columns = [
Allele(c, allele_data[c]['frequency'] / 100) for c in df.columns
]
else:
df.columns = [Allele(c) for c in df.columns]
return EpitopePredictionResult(df)
def parse_external_result(self, output):
"""
Searches within the defined dir _file for the newest dir and reads
the prediction file from there
:param str output: The path to the output dir
:return: The predicted HLA genotype
:rtype: list(:class:`~Fred2.Core.Allele.Allele`)
"""
alleles = []
try:
with open(output+"-ClassI.HLAgenotype4digits") as c1:
for row in csv.DictReader(c1, delimiter="\t"):
alleles.extend([Allele("HLA-"+row["Allele 1"]), Allele("HLA-"+row["Allele 2"])])
except IOError as e:
warnings.warn("Output file {c1} for HLA-I could not be found. {error}".format(
c1=output + "-ClassI.HLAgenotype4digits"), error=e)
try:
with open(output+"-ClassII.HLAgenotype4digits") as c2:
for row in csv.DictReader(c2, delimiter="\t"):
alleles.extend([Allele("HLA-"+row["Allele 1"]), Allele("HLA-"+row["Allele 2"])])
except IOError as e:
warnings.warn("Output file {c2} for HLA-I could not be found. {error}".format(
c2=output + "-ClassII.HLAgenotype4digits"), error=e)
return alleles
def test_consistency(self):
"""
tests all __*__ (including init)
test has several asserts! If one fails, the following will not be evaluated!
"""
self.assertTrue(repr(self.simple) == "HLA-A*02:01")
self.assertEqual(self.simple, Allele("HLA-A*02:01"))
self.assertNotEqual(repr(self.simple), Allele("HLA-A*02:01:666"))
def setUp(self):
#Peptides of different length 9,10,11,12,13,14,15
self.peptides_mhcI = [Peptide("SYFPEITHI"), Peptide("IHTIEPFYS")]
self.peptides_mhcII = [
Peptide("SYFPEITHI"),
Peptide("IHTIEPFYSAAAAAA")
]
self.mhcI = [Allele("HLA-B*15:01"), Allele("HLA-A*02:01")]
self.mhcII = [Allele("HLA-DRB1*07:01"), Allele("HLA-DRB1*15:01")]
def compute_affinities(input_alleles, input_peptides, output_affinities,
processes, predictor):
''' Computes the binding affinities between the given peptides and HLA alleles
'''
alleles = [
Allele(a.replace('HLA-', ''))
for a in utilities.get_alleles_and_thresholds(input_alleles).index
]
LOGGER.info('Loaded %d alleles', len(alleles))
with open(input_peptides) as f:
reader = csv.DictReader(f)
peptides = [(Peptide(r['peptide']), len(r['proteins'].split(';')))
for r in reader]
peptides.sort(key=lambda p: p[1], reverse=True)
LOGGER.info('Loaded %d peptides', len(peptides))
results = utilities.parallel_apply(
get_binding_affinity_process,
((predictor.lower(), batch, alleles)
for batch in utilities.batches((p for p, _ in peptides), bsize=256)),
processes)
count = 0
for bindings in results:
bindings.to_csv(output_affinities,
header=(count == 0),
mode=('w' if count == 0 else 'a'))
count += len(bindings)
LOGGER.debug('Processed %d peptides (%.2f%%)...', count,
100 * count / len(peptides))
def parse_external_result(self, output):
"""
Searches within the defined dir _file for the newest dir and reads
the prediction file from there
:param str output: The path to the output dir
:return: The predicted HLA genotype
:rtype: list(:class:`~Fred2.Core.Allele.Allele`)
"""
alleles = []
try:
with open(output + "-ClassI.HLAgenotype4digits") as c1:
for row in csv.DictReader(c1, delimiter="\t"):
alleles.extend([
Allele("HLA-" + row["Allele 1"].replace("'", "")),
Allele("HLA-" + row["Allele 2"].replace("'", ""))
])
except IOError as e:
warnings.warn(
"Output file {c1} for HLA-I could not be found. {error}".
format(c1=output + "-ClassI.HLAgenotype4digits"),
error=e)
try:
with open(output + "-ClassII.HLAgenotype4digits") as c2:
DQA = []
DQB = []
for row in csv.DictReader(c2, delimiter="\t"):
a1, a2 = row["Allele 1"], row["Allele 2"]
if "DRB" in a1 or "DRB" in a2:
alleles.extend([
Allele("HLA-" + a1.replace("'", "")),
Allele("HLA-" + a2.replace("'", ""))
])
elif "DQA" in a1 or "DQA" in a2:
DQA.extend([a1.replace("'", ""), a2.replace("'", "")])
else:
DQB.extend([a1.replace("'", ""), a2.replace("'", "")])
for dq in itertools.product(DQA, DQB):
alleles.append(CombinedAllele("HLA-" + "-".join(dq)))
except IOError as e:
warnings.warn(
"Output file {c2} for HLA-I could not be found. {error}".
format(c2=output + "-ClassII.HLAgenotype4digits"),
error=e)
return alleles
def predict_peptide_effects(peptides, alleles=None):
"""
Predict the peptide effect for all the available methods on the machine
Args:
peptides (list of Peptides): Usually an output from read_fasta
alleles (list of chars): Alleles for which to run the predictors
Returns:
pd.DataFrame: Tidy pd.DataFrame. If the method is unable to predict
for a particular value the rows are not present.
Example:
>>> peptides = [Peptide("SYFPEITHI"), Peptide("FIASNGVKL"), Peptide("LLGATCMFV")]
>>> alleles = ['A*02:16', 'B*45:01']
>>> predict_peptide_effects(peptides, alleles = alleles).head()
Seq Method allele score
0 (F, I, A, S, N, G, V, K, L) arb A*02:16 594.691144
1 (F, I, A, S, N, G, V, K, L) smm A*02:16 159.768074
2 (F, I, A, S, N, G, V, K, L) smmpmbec A*02:16 211.977614
4 (F, I, A, S, N, G, V, K, L) unitope A*02:16 0.527849
5 (L, L, G, A, T, C, M, F, V) arb A*02:16 6.784222
"""
dt = valid_predictors()
results = []
for i in range(len(dt)):
# subset to valid alleles
if alleles is not None:
valid_alleles = dt.iloc[i]["supportedAlleles"].intersection(
alleles)
if len(valid_alleles) == 0:
continue
valid_alleles = [Allele(al) for al in valid_alleles]
else:
valid_alleles = None
method = dt.iloc[i]["name"]
print("method: ", method)
# TODO - use try, except
t0 = time.time()
try:
results.append(
EpitopePredictorFactory(method).predict(peptides,
alleles=valid_alleles))
except:
print("Error! Unable to run ", method, ": ", sys.exc_info())
t1 = time.time()
print(" - runtime: ", str(t1 - t0))
df = results[0].merge_results(results[1:]).reset_index()
dfm = pd.melt(df,
id_vars=["Seq", "Method"],
var_name="allele",
value_name="score")
dfm = dfm[dfm["score"].notnull()]
dfm.rename(columns={'Seq': 'peptide', 'Method': 'method'}, inplace=True)
return dfm
def generate_alleles(allele_file, generated=None):
"""
generate allele objects from input
"""
result=[]
with open(allele_file, "r") as f:
for l in f:
al,freq = l.replace(","," ").replace(";"," ").replace("\n","").split()
if al.split("HLA-")[-1][0] in ["A","B","C"]:
result.append(Allele(al,prob=float(freq)))
return result
def test_pareto_front_assembly(self):
cl_pred = CleavageSitePredictorFactory("PCM")
ep_pred = EpitopePredictorFactory("SMM")
allele = [Allele("HLA-A*02:01")]
thresh = {a.name:10000 for a in allele}
comp = lambda a,b: a <= b
assembler = ParetoEpitopeAssembly(self.peptides,cl_pred, ep_pred, allele, thresh, comp, solver="cbc", verbosity=0)
r = assembler.paretosolve()
print(r)
#print assembler.solve(eps=2.0)
def run_predictor(pred, dataset):
predictor = EpitopePredictorFactory(pred)
results = ()
try:
results = predictor.predict(dataset,
alleles=[Allele(a) for a in args.allele])
print(results)
print(results.describe())
except ValueError:
pass
return (len(results), len(dataset))
def test_pareto_assembly(self):
cl_pred = CleavageSitePredictorFactory("PCM")
ep_pred = EpitopePredictorFactory("SMM")
allele = [Allele("HLA-A*02:01")]
thresh = {a.name:10000 for a in allele}
comp = lambda a,b: a <= b
print(ep_pred.predict(self.peptides,alleles=allele))
#cl_pred, ep_pred, alleles, threshold, comparator, length=9
assembler = ParetoEpitopeAssembly(self.peptides,cl_pred, ep_pred, allele, thresh, comp, solver="cbc", verbosity=1)
r = assembler.solve(eps=1e10, order=(1,0))
print(r)
def design_spacers(input_epitopes, input_alleles, top_proteins, top_immunogen,
top_alleles, solver, pssm_cleavage, alpha, beta,
spacer_length, pssm_epitope, processes, output_spacers):
all_epitopes = list(
utilities.load_epitopes(input_epitopes, top_immunogen, top_alleles,
top_proteins).keys())
epitopes = [e for e in all_epitopes if 'X' not in e]
LOGGER.debug('Removed %d epitopes with unknown amino acids',
len(all_epitopes) - len(epitopes))
LOGGER.info('Loaded %d epitopes', len(epitopes))
alleles_df = utilities.get_alleles_and_thresholds(input_alleles)
allele_list = [
Allele(a.replace('HLA-', ''), prob=row.frequency / 100)
for a, row in alleles_df.iterrows()
]
threshold = {
a.replace('HLA-', ''): row.threshold
for a, row in alleles_df.iterrows()
}
LOGGER.info('Loaded %d alleles', len(allele_list))
if pssm_cleavage != 'PCM':
raise ValueError('Only PCM supported as cleavage predictor')
cleavage_predictor = PCM() # TODO use factory when it works
if pssm_epitope != 'BIMAS':
raise ValueError('Only BIMAS supported as epitope predictor')
epitope_predictor = BIMAS() # TODO use factory when it works
designer = OptimalSpacerDesign(
epitopes,
cleavage_predictor,
epitope_predictor,
allele_list,
threshold=threshold,
solver=solver,
k=spacer_length,
alpha=alpha,
beta=beta,
).solve(threads=processes)
LOGGER.info('Writing results...')
with open(output_spacers, 'w') as f:
writer = csv.writer(f)
writer.writerow(('from', 'to', 'score', 'spacer'))
writer.writerows(
(ei, ej, designer.adj_matrix[ei, ej], designer.spacer[ei, ej])
for ei in epitopes for ej in epitopes if ei != ej)
def parse_external_result(self, output):
"""
Searches within the defined dir _file for the newest dir and reads
the prediction file from there
:param str output: The path to the output dir
:return: The predicted HLA genotype
:rtype: list(:class:`~Fred2.Core.Allele.Allele`)
"""
all_subdirs = [os.path.join(output,d) for d in os.listdir(output) if os.path.isdir(os.path.join(output,d))]
latest_subdir = max(all_subdirs, key=os.path.getmtime)
result_file = latest_subdir+"/"+os.path.basename(os.path.normpath(latest_subdir))+"_result.tsv"
with open(result_file, "r") as f:
row = csv.DictReader(f, delimiter="\t").next()
return map(lambda x: Allele("HLA-"+x), [ row[k] for k in ["A1","A2","B1","B2","C1","C2"]])
def generate_epitope_result(input, allele_file):
"""
generates EpitopePredictionResult from output of epitopeprediction and neoepitopeprediction
"""
#first generate alleles in allele file
alleles = {}
with open(allele_file, "r") as af:
for l in af:
allele, freq = l.split("\t")
alleles[allele] = Allele(allele, prob=float(freq))
r_raw = pandas.read_csv(input, sep="\t")
res_dic = {}
method = r_raw.loc[0, "Method"]
columns = set(["Sequence", "Method", "Antigen ID", "Variant"])
alleles_raw = [c for c in r_raw.columns if c not in columns]
for k, row in r_raw.iterrows():
seq = row["Sequence"]
protPos = collections.defaultdict(list)
try:
protPos = {Protein(p, gene_id=p, transcript_id=p): [0] for p in str(row["Antigen ID"]).split(",")}
except KeyError:
pass
pep = Peptide(seq, protein_pos=protPos)
for a in alleles_raw:
if a in alleles:
if alleles[a] not in res_dic:
res_dic[alleles[a]] = {}
res_dic[alleles[a]][pep] = float(row[a])
if not res_dic:
sys.stderr.write("HLA alleles of population and HLA used for prediction did not overlap.")
sys.exit(-1)
df_result = EpitopePredictionResult.from_dict(res_dic)
df_result.index = pandas.MultiIndex.from_tuples([tuple((i, method)) for i in df_result.index],
names=['Seq', 'Method'])
return df_result, method
def string_of_beads(input_proteins, input_alleles, input_epitopes,
input_cleavages, output_vaccine, cocktail, greedy_subtour,
max_aminoacids, max_epitopes, min_alleles, min_proteins,
min_avg_prot_conservation, min_avg_alle_conservation):
program_start_time = time.time()
# load proteins
LOGGER.info('Reading sequences...')
proteins = FileReader.read_fasta(input_proteins, in_type=Protein)
LOGGER.info('%d proteins read', len(proteins))
# load alleles
alleles = [
Allele(a)
for a in utilities.get_alleles_and_thresholds(input_alleles).index
]
LOGGER.info('Loaded %d alleles', len(alleles))
# load epitopes
epitopes = utilities.load_epitopes(input_epitopes)
LOGGER.info('Loaded %d epitopes', len(epitopes))
# read cleavage scores
cleavage_epitopes = set()
with open(input_cleavages) as f:
cleavages = {}
for row in csv.DictReader(f):
cleavages[(row['from'], row['to'])] = float(row['score'])
cleavage_epitopes.add(row['from'])
cleavage_epitopes.add(row['to'])
LOGGER.info('Loaded %d cleavage scores', len(cleavages))
# compute edge cost
edge_cost, vertices, vertices_rewards = [], [], []
vertex_to_epitope = [''] + list(cleavage_epitopes)
for ep_from in vertex_to_epitope:
vertices.append(ep_from)
vertices_rewards.append(0 if ep_from ==
'' else epitopes[ep_from]['immunogen'])
edge_cost.append([
cleavages[(ep_from,
ep_to)] if ep_from != '' and ep_to != '' else 0.0
for ep_to in vertex_to_epitope
])
LOGGER.info('Kept %d epitopes with available clevages', len(vertices) - 1)
type_coverage, min_type_coverage, min_avg_type_conservation = utilities.compute_coverage_matrix(
[epitopes[e] for e in vertex_to_epitope[1:]], min_alleles,
min_proteins, min_avg_prot_conservation, min_avg_alle_conservation,
len(proteins), len(alleles))
# find optimal design
solver_build_time = time.time()
solver = TeamOrienteeringIlp(
num_teams=cocktail,
vertex_reward=vertices_rewards,
edge_cost=edge_cost,
type_coverage=type_coverage,
min_type_coverage=min_type_coverage,
min_avg_type_conservation=min_avg_type_conservation,
max_edge_cost=max_aminoacids,
max_vertices=max_epitopes,
lazy_subtour_elimination=not greedy_subtour)
solver.build_model()
solver_start_time = time.time()
result = solver.solve()
solver_end_time = time.time()
# print info and save
with open(output_vaccine, 'w') as f:
writer = csv.writer(f)
writer.writerow(('cocktail', 'index', 'epitope'))
for i, mosaic in enumerate(result):
LOGGER.info('Mosaic #%d', i + 1)
for j, (_, vertex) in enumerate(mosaic[:-1]):
epitope = epitopes[vertex_to_epitope[vertex]]
writer.writerow((i, j, epitope['epitope']))
LOGGER.info(' %s - IG: %.2f', epitope['epitope'],
epitope['immunogen'])
LOGGER.info('==== Stopwatch')
LOGGER.info(' Total time : %.2f s',
solver_end_time - program_start_time)
LOGGER.info(' Pre-processing : %.2f s',
solver_build_time - program_start_time)
LOGGER.info(' Model creation time : %.2f s',
solver_start_time - solver_build_time)
LOGGER.info(' Solving time : %.2f s',
solver_end_time - solver_start_time)
def run_sequential(input_epitopes, input_alleles, input_affinities,
output_vaccine, num_epitopes, min_alleles, min_proteins,
solver, **kwargs):
epitope_data = {
k: v
for k, v in utilities.load_epitopes(input_epitopes).items()
if 'X' not in k
}
LOGGER.info('Loaded %d epitopes', len(epitope_data))
peptide_coverage = {
# we don't really need the actual protein sequence, just fill it with the id to make it unique
Peptide(r['epitope']):
set(Protein(gid, gene_id=gid) for gid in r['proteins'])
for r in epitope_data.values()
}
allele_data = utilities.get_alleles_and_thresholds(input_alleles).to_dict(
'index')
alleles = [
Allele(allele.replace('HLA-', ''), prob=data['frequency'] / 100)
for allele, data in allele_data.items()
]
threshold = {
allele.replace('HLA-', ''): data['threshold']
for allele, data in allele_data.items()
}
LOGGER.info('Loaded %d alleles', len(threshold))
affinities = affinities_from_csv(input_affinities,
allele_data,
peptide_coverage=peptide_coverage)
LOGGER.info('Loaded %d affinities', len(affinities))
LOGGER.info('Selecting epitopes...')
model = OptiTope(affinities, threshold, k=num_epitopes, solver=solver)
if min_alleles is not None:
model.activate_allele_coverage_const(min_alleles)
if min_proteins is not None:
model.activate_antigen_coverage_const(min_proteins)
selected_epitopes = model.solve()
LOGGER.info('Creating spacers...')
vaccine = EpitopeAssemblyWithSpacer(selected_epitopes,
PCM(),
BIMAS(),
alleles,
threshold=threshold,
solver=solver).solve()
immunogen = sum(epitope_data[str(e)]['immunogen'] for e in vaccine[::2])
sequence = ''.join(map(str, vaccine))
cleavage = pcm.DoennesKohlbacherPcm().cleavage_per_position(sequence)
with open(output_vaccine, 'w') as f:
writer = csv.DictWriter(
f, ('immunogen', 'vaccine', 'spacers', 'cleavage'))
writer.writeheader()
writer.writerow({
'immunogen': immunogen,
'vaccine': sequence,
'spacers': ';'.join(str(e) for e in vaccine[1::2]),
'cleavage': ';'.join('%.3f' % c for c in cleavage)
})
def get_mosaic_solver_instance(logger, input_proteins, input_alleles,
input_epitopes, input_overlaps, **kwargs):
top_immunogen = kwargs.pop('top_immunogen')
top_alleles = kwargs.pop('top_alleles')
top_proteins = kwargs.pop('top_proteins')
min_overlap = kwargs.get('min_overlap', 0)
cocktail = kwargs.get('cocktail', 1)
greedy_subtour = kwargs.get('greedy_subtour')
max_epitopes = kwargs.get('max_epitopes')
max_aminoacids = kwargs.get('max_aminoacids')
min_alleles = kwargs.get('min_alleles', 0)
min_proteins = kwargs.get('min_proteins', 0)
min_avg_prot_conservation = kwargs.get('min_avg_prot_conservation', 0)
min_avg_alle_conservation = kwargs.get('min_avg_alle_conservation', 0)
# load proteins
logger.info('Reading sequences...')
proteins = FileReader.read_fasta(input_proteins, in_type=Protein)
logger.info('%d proteins read', len(proteins))
# load alleles
alleles = [
Allele(a) for a in get_alleles_and_thresholds(input_alleles).index
]
logger.info('Loaded %d alleles', len(alleles))
# load epitopes
epitope_data = list(
load_epitopes(input_epitopes, top_immunogen, top_alleles,
top_proteins).values())
logger.info('Loaded %d epitopes', len(epitope_data))
# load edge cost
logger.info('Loading overlaps...')
vertex_rewards = [0] + [b['immunogen'] for b in epitope_data]
edges = load_edges_from_overlaps(input_overlaps, min_overlap,
[b['epitope'] for b in epitope_data])
logger.info('Kept %d edges (from %d)', len(edges),
len(epitope_data) * (len(epitope_data) + 1))
# compute hla and protein coverage
logger.info('Computing coverage matrix...')
type_coverage, min_type_coverage, min_avg_type_conservation = compute_coverage_matrix(
epitope_data, min_alleles, min_proteins, min_avg_prot_conservation,
min_avg_alle_conservation, len(proteins), len(alleles))
# find optimal design
solver = TeamOrienteeringIlp(
num_teams=cocktail,
vertex_reward=vertex_rewards,
edge_cost=edges,
max_edge_cost=0,
max_vertices=0,
lazy_subtour_elimination=not greedy_subtour,
type_coverage=type_coverage,
min_type_coverage=min_type_coverage,
min_avg_type_conservation=min_avg_type_conservation,
)
if isinstance(max_epitopes, (int, float)):
solver.update_max_vertices(max_epitopes)
if isinstance(max_aminoacids, (int, float)):
solver.update_max_edge_cost(max_aminoacids)
return solver, {
'proteins': proteins,
'alleles': alleles,
'epitope_data': epitope_data,
}
from Fred2.EpitopePrediction import EpitopePredictorFactory
####################################################################################################
# Convert raw peptide sequences to Fred2.Core.Peptide objects
all_peptides = [Peptide(row.strip()) for row in args.peptides]
# Separate peptides by length
peptides_by_length = {}
for peptide in all_peptides:
if not len(peptide) in peptides_by_length:
peptides_by_length[len(peptide)] = []
peptides_by_length[len(peptide)].append(peptide)
# Convert raw allele strings to Fred2.Core.Allele objects
alleles = [Allele(allele) for allele in args.alleles]
# Instatiate predictor
predictor = EpitopePredictorFactory("Syfpeithi")
def matrix_max(matrix):
"""Returns the maximum attainable score for a pssm"""
return sum([max(value.values()) for _, value in matrix.items()])
def load_allele_model(allele, length):
"""Returns the SYFPEITHI pssm for a given allele"""
allele_model = "%s_%i" % (allele, length)
try:
return matrix_max(
def setUp(self):
self.simple = Allele("HLA-A*02:01")
file_in = arguments["--input"]
if not file_in:
file_in = "./data/binders.csv"
file_out = arguments["--output"]
dt = pd.read_csv(file_in)
dt = dt[dt["Sequence"].notnull()]
dt = dt[dt["Sequence"].str.len() == 9]
peptides = [Peptide(peptide) for peptide in dt["Sequence"]]
dt["allele"] = dt["allele"].str.replace("\*","").\
str.replace("(-[a-zA-Z]+)([0-9]{2})([0-9]{2})","\\1*\\2:\\3").\
str.replace("w","").\
str.replace("HLA-","")
dt.rename(columns={"Sequence": "peptide"}, inplace=True)
alleles = [Allele(allele) for allele in dt["allele"].unique().tolist()]
res = fred2wrap.predict_peptide_effects(
peptides, alleles=dt["allele"].unique().tolist())
res["peptide"] = [peptide.tostring() for peptide in res["peptide"]]
res["allele"] = [str(allele) for allele in res["allele"]]
res = res.pivot_table(index=["peptide", "allele"],
columns='method',
values='score').reset_index(None)
dt_merge = pd.merge(dt, res, on=["peptide", "allele"], how="left")
dt_merge.to_csv(file_out, index=False)
all_peptides = dt["mutant_sequence"].append(dt["wt_sequence"]).unique()
peptides = [Peptide(peptide) for peptide in all_peptides]
dt["allele"] = dt["allele"].str.replace("\*", "").\
str.replace(":", "").\
str.replace("(-[a-zA-Z]+)([0-9]{2})([0-9]{2})", "\\1*\\2:\\3").\
str.replace("w", "").\
str.replace("HLA-", "")
# TODO
# dt.rename(columns = {"Sequence": "peptide"}, inplace = True)
alleles = []
valid_alleles = []
for allele in dt["allele"].tolist():
try:
a = Allele(allele)
valid_alleles.append(True)
except:
a = None
valid_alleles.append(False)
alleles.append(a)
# subset invalid allele names
dt = dt[pd.Series(valid_alleles)]
res = fred2wrap.predict_peptide_effects(peptides, alleles=dt["allele"].unique().tolist())
res["peptide"] = [str(peptide) for peptide in res["peptide"]]
res["allele"] = [str(allele) for allele in res["allele"]]
# TODO - change melt order
res = res.pivot_table(index=["peptide", "allele"],