def read_file(session, fmt, handle, sample, v_germlines, j_germlines, props):
reader = csv.DictReader(handle, delimiter='\t')
uniques = {}
for i, line in enumerate(reader):
if fmt == 'adaptive':
try:
line = extract_adaptive_sequence(i, line, v_germlines,
j_germlines)
except (AlignmentException, KeyError) as e:
seq = VDJSequence('seq_{}'.format(i), '')
add_noresults_for_vdj(session, seq, sample, str(e))
continue
seq = VDJSequence(line['SEQUENCE_ID'],
line['SEQUENCE_IMGT'].replace('.', '-'))
if 'DUPCOUNT' in line:
seq.copy_number = int(line['DUPCOUNT'])
try:
alignment = create_alignment(seq, line, v_germlines, j_germlines)
for other in uniques.setdefault(len(alignment.sequence.sequence),
[]):
if dnautils.equal(other.sequence.sequence,
alignment.sequence.sequence):
other.sequence.copy_number += (
alignment.sequence.copy_number)
break
else:
uniques[len(alignment.sequence.sequence)].append(alignment)
except AlignmentException as e:
add_noresults_for_vdj(session, seq, sample, str(e))
uniques = [s for k in sorted(uniques.keys()) for s in uniques[k]]
lens = []
muts = []
for unique in uniques:
try:
props.validate(unique)
add_sequences(session, [unique], sample)
lens.append(unique.v_length)
muts.append(unique.v_mutation_fraction)
except AlignmentException as e:
add_noresults_for_vdj(session, seq, sample, str(e))
if len(lens) > 0:
sample.v_ties_len = sum(lens) / float(len(lens))
sample.v_ties_mutations = sum(muts) / float(len(muts))
session.commit()
def read_input(path):
vdjs = []
parser = SeqIO.parse(path, 'fasta' if path.endswith('.fasta') else 'fastq')
# Collapse identical sequences
logger.info('Parsing input')
for record in parser:
try:
vdjs.append(
VDJSequence(
seq_id=record.description,
sequence=str(record.seq),
quality=funcs.ord_to_quality(
record.letter_annotations.get('phred_quality'))))
except ValueError:
continue
logger.info('There are {} sequences'.format(len(vdjs)))
return vdjs
def process_sample(session, sample, indexes, temp, v_germlines, j_germlines,
nproc):
indels = session.query(Sequence.ai, Sequence.seq_id, Sequence.sample_id,
Sequence.sequence).filter(
Sequence.sample_id == sample.id,
Sequence.probable_indel_or_misalign == 1)
# Get the sequences that were not identifiable
noresults = session.query(NoResult).filter(NoResult.sample_id == sample.id)
if indels.count() == 0 and noresults.count() == 0:
logger.info('Sample {} has no indels or noresults'.format(sample.id))
return
logger.info('Sample {} has {} indels and {} noresults'.format(
sample.id, indels.count(), noresults.count()))
mut_bucket = v_germlines.mut_bucket(sample.v_ties_mutations)
len_bucket = v_germlines.length_bucket(sample.v_ties_len)
bucket = '{}_{}'.format(str(mut_bucket).replace('.', ''), len_bucket)
sample_v_germlines = get_formatted_ties(
v_germlines.all_ties(sample.v_ties_len, sample.v_ties_mutations))
sample_j_germlines = get_formatted_ties(
j_germlines.all_ties(sample.v_ties_len, sample.v_ties_mutations))
if bucket not in indexes:
indexes.add(bucket)
v_path = os.path.join(temp, 'v_genes_{}'.format(bucket))
j_path = os.path.join(temp, 'j_genes_{}'.format(bucket))
logger.info('Creating index for V-ties at {} length, {} '
'mutation'.format(len_bucket, mut_bucket))
build_index(sample_v_germlines, v_path)
build_index(sample_j_germlines, j_path)
seq_path = os.path.join(temp, 'll_{}.fasta'.format(sample.id))
with open(seq_path, 'w+') as fh:
fh.write(
get_fasta({
'tp=Sequence|ai={}|sample_id={}|seq_id={}'.format(
r.ai, r.sample_id, r.seq_id): r.sequence
for r in indels
}))
fh.write(
get_fasta({
'tp=NoResult|pk={}|sample_id={}|seq_id={}'.format(
r.pk, r.sample_id, r.seq_id): r.sequence
for r in noresults
}))
alignments = {}
logger.info('Running bowtie2 for V-gene sequences')
for line in get_reader(
align_reference(temp, 'v_genes_{}'.format(bucket), seq_path,
nproc)):
line['ref_offset'] = int(line['ref_offset']) - 1
ref_gene = line['reference']
ref, seq, rem_seqs = create_seqs(
ref_seq=sample_v_germlines[ref_gene].replace('-', ''),
min_size=CDR3_OFFSET,
**line)
if len(rem_seqs) == 0:
continue
ref, seq, seq_start = add_imgt_gaps(sample_v_germlines[ref_gene], ref,
seq, line['ref_offset'])
if len(ref) < CDR3_OFFSET:
continue
alignments[line['seq_id']] = {
'v_germline': ref,
'v_gene': line['reference'],
'seq_start': seq_start,
'v_sequence': seq,
'v_rem_seq': rem_seqs[-1],
'cdr3_start': len(ref)
}
seq_path = os.path.join(temp, 'll_j_{}.fasta'.format(sample.id))
with open(seq_path, 'w+') as fh:
seqs = {
k: v['v_rem_seq']
for k, v in alignments.iteritems() if len(v['v_rem_seq']) > 0
}
fh.write(get_fasta(seqs))
tasks = []
logger.info('Running bowtie2 for J-gene sequences')
for line in get_reader(
align_reference(temp, 'j_genes_{}'.format(bucket), seq_path,
nproc)):
line['ref_offset'] = int(line['ref_offset']) - 1
ref_gene = line['reference']
ref, seq, rem_seqs = create_seqs(
ref_seq=sample_j_germlines[ref_gene].replace('-', ''),
min_size=j_germlines.upstream_of_cdr3,
**line)
alignments[line['seq_id']]['j_gene'] = line['reference']
full_seq = (alignments[line['seq_id']]['v_sequence'] +
alignments[line['seq_id']]['v_rem_seq'])
if len(rem_seqs) > 0:
full_seq = full_seq[:-len(rem_seqs[-1])]
cdr3_end = len(full_seq)
if len(ref) < j_germlines.upstream_of_cdr3:
continue
for i in range(j_germlines.upstream_of_cdr3):
if ref[-i] != '-':
cdr3_end -= 1
alignments[line['seq_id']]['cdr3_end'] = cdr3_end
cdr3_length = cdr3_end - alignments[line['seq_id']]['cdr3_start']
full_germ = (alignments[line['seq_id']]['v_germline'] +
(GAP_PLACEHOLDER * cdr3_length))
j_length = len(full_seq) - len(full_germ)
if j_length <= 0 or cdr3_length <= 0:
continue
full_germ += ref[-j_length:]
r_type, pk, sample_id, seq_id = [
v.split('=', 1)[1] for v in line['seq_id'].split('|', 3)
]
insertions = gap_positions(full_germ)
deletions = gap_positions(full_seq)
alignment = VDJAlignment(
VDJSequence(seq_id, full_seq.replace(GAP_PLACEHOLDER, '-')))
alignment.germline = full_germ.replace(GAP_PLACEHOLDER, '-')
if len(alignment.germline) != len(alignment.sequence.sequence):
continue
alignment.v_gene.add(GeneName(alignments[line['seq_id']]['v_gene']))
alignment.j_gene.add(GeneName(alignments[line['seq_id']]['j_gene']))
alignment.seq_offset = alignments[line['seq_id']]['seq_start']
# TODO: This should really look for a streak like in anchoring
alignment.germline_cdr3 = '-' * cdr3_length
gaps_in_seq = alignment.sequence.sequence[
alignment.
seq_start:alignments[line['seq_id']]['cdr3_start']].count('-')
alignment.v_length = (alignments[line['seq_id']]['cdr3_start'] -
alignment.seq_offset) - gaps_in_seq
alignment.j_length = j_length
alignment.v_mutation_fraction = 1 - (alignment.v_match /
float(alignment.v_length))
alignment.cdr3_start = alignments[line['seq_id']]['cdr3_start']
alignment.cdr3_num_nts = cdr3_length
alignment.post_cdr3_length = j_length
alignment.insertions = insertions
alignment.deletions = deletions
alignment.locally_aligned = True
tasks.append({
'r_type': r_type,
'pk': int(pk),
'sample_id': int(sample_id),
'alignment': alignment
})
return tasks
def parse_airr(line, v_germlines, j_germlines):
seq = VDJSequence(
seq_id=line['sequence_id'].replace('reversed|', ''),
sequence=line['sequence_alignment'],
rev_comp=line['rev_comp'] == 'T',
)
if not all([line['v_call'], line['j_call'], line['junction_aa']]):
raise AlignmentException(seq, 'Missing v_gene, j_gene, or junction_aa')
seq.pad(int(line['v_germline_start']) - 1)
try:
v_germ_seq = v_germlines.get_ties(line['v_call'].split(','))
except KeyError:
raise AlignmentException(
seq,
'V-gene {} not in germline database'.format(line['v_call'])
)
aligned_germ = ''.join([
v_germ_seq.replace('-', '')[:int(line['v_germline_start']) - 1],
line['germline_alignment']
])
# Append the missing portion, if any, of the J to the germline
j_germ_seq = j_germlines.get_ties(line['j_call'].split(','))
append_j = len(j_germ_seq) - int(line['j_germline_end'])
if append_j > 0:
aligned_germ += j_germ_seq[-append_j:]
seq.pad_right(append_j)
aligned_seq, gaps_added = add_imgt_gaps(v_germ_seq, seq)
aligned_germ = add_imgt_gaps(
v_germ_seq, VDJSequence('', aligned_germ)
)[0].sequence
cdr3_start = int(line['cdr3_start']) - int(line['v_sequence_start'])
# Push the start of the CDR3 based on number of IMGT gaps added. Then add
# 3 because IgBLAST's CDR3 excludes the preserved Cysteine
cdr3_start += gaps_added - 3
cdr3_start += aligned_seq.sequence[:cdr3_start].count('-')
cdr3_start += int(line['v_germline_start']) - 1
cdr3_end = cdr3_start + len(line['cdr3']) + 6
# If there is an insertion in the CDR3 but not junction, increase CDR3
# length
junction_insertions = aligned_germ[cdr3_end - 3:cdr3_end].count('-')
cdr3_end += junction_insertions
cdr3_seq = aligned_seq.sequence[cdr3_start:cdr3_end]
germline_cdr3 = aligned_germ[cdr3_start:cdr3_end]
aligned_germ = ''.join([
aligned_germ[:cdr3_start],
'.' * (cdr3_end - cdr3_start),
aligned_germ[cdr3_end:]
])
aligned_seq = ''.join([
aligned_seq.sequence[:cdr3_start],
cdr3_seq,
aligned_seq.sequence[cdr3_end:]
])
total_insertions = line['v_germline_alignment'].count('-')
correct_cdr3_start = CDR3_OFFSET + total_insertions
if cdr3_start != correct_cdr3_start:
raise AlignmentException(
seq, 'CDR3 starts at {} instead of {} ({} insertions)'.format(
cdr3_start, correct_cdr3_start, total_insertions))
alignment = funcs.ClassProxy(VDJAlignment(
VDJSequence(line['sequence_id'], aligned_seq.replace('.', '-'))
))
alignment.germline = aligned_germ.replace('.', '-')
alignment.v_gene = set([GeneName(c) for c in line['v_call'].split(',')])
alignment.j_gene = set([GeneName(c) for c in line['j_call'].split(',')])
alignment.cdr3_start = cdr3_start
alignment.cdr3_num_nts = len(cdr3_seq)
alignment.locally_aligned = True
alignment.germline_cdr3 = germline_cdr3
alignment.seq_offset = int(line['v_germline_start']) - 1
alignment.v_length = int(line['v_alignment_end'])
alignment.j_length = (int(line['j_alignment_end']) -
int(line['j_alignment_start']))
alignment.v_mutation_fraction = (100 - float(line['v_identity'])) / 100
# Skipping the germline_cdr3 field and instead populating its dependencies
# via the proxy
alignment.j_match = float(line['j_identity']) * alignment.j_length / 100
alignment.post_cdr3_length = len(alignment.sequence.sequence) - cdr3_end
alignment.insertions = funcs.gap_positions(aligned_germ)
alignment.deletions = funcs.gap_positions(aligned_seq)
return alignment
def read_file(session, handle, sample, v_germlines, j_germlines, columns,
remaps):
seqs = _collapse_seqs(session, sample,
csv.DictReader(handle, delimiter='\t'), columns)
aligned_seqs = {}
missed = 0
total = 0
for total, seq in enumerate(seqs):
if total > 0 and total % 1000 == 0:
logger.info('Finished {}'.format(total))
session.commit()
orig_v_genes = set(
re.findall('IGHV[^ ,]+', seq['record'][columns.v_gene]))
orig_j_genes = set(
re.findall('IGHJ[^ ,]+', seq['record'][columns.j_gene]))
if remaps is not None:
remapped_j_genes = set([])
for j in orig_j_genes:
for remap_from, remap_to in remaps.iteritems():
if j.startswith(remap_from):
remapped_j_genes.add(remap_to)
break
else:
remapped_j_genes.add(j)
orig_j_genes = remapped_j_genes
v_genes = filter(lambda v: v in v_germlines, orig_v_genes)
j_genes = filter(lambda j: j in j_germlines, orig_j_genes)
vdj = VDJSequence(seq['seq_ids'],
seq['record'][columns.full_sequence],
v_germlines,
j_germlines,
force_vs=v_genes,
force_js=j_genes)
try:
if len(v_genes) == 0:
raise AlignmentException('No valid V germline for {}'.format(
','.join(sorted(orig_v_genes))))
if len(j_genes) == 0:
raise AlignmentException('No valid J germline for {}'.format(
','.join(sorted(orig_j_genes))))
vdj.analyze()
if vdj.sequence in aligned_seqs:
aligned_seqs[vdj.sequence].ids += vdj.ids
else:
aligned_seqs[vdj.sequence] = vdj
except AlignmentException as e:
add_as_noresult(session, vdj, sample, str(e))
missed += 1
logger.info('Aligned {} / {} sequences'.format(total - missed + 1, total))
logger.info('Collapsing ambiguous character sequences')
if len(aligned_seqs) > 0:
avg_mut = sum([v.mutation_fraction for v in aligned_seqs.values()
]) / float(len(aligned_seqs))
avg_len = sum([v.v_length for v in aligned_seqs.values()]) / float(
len(aligned_seqs))
sample.v_ties_mutations = avg_mut
sample.v_ties_len = avg_len
if columns.ties:
add_uniques(session,
sample,
aligned_seqs.values(),
realign_mut=avg_mut,
realign_len=avg_len,
trim_to=columns.trim_to,
max_padding=columns.max_padding)
else:
add_uniques(session, sample, aligned_seqs.values())
session.commit()
def do_task(self, args):
meta = args['meta']
self.info('Starting sample {}'.format(meta['sample_name']))
study, sample = self._setup_sample(meta)
vdjs = {}
parser = SeqIO.parse(
args['path'],
'fasta' if args['path'].endswith('.fasta') else 'fastq')
# Collapse identical sequences
self.info('\tCollapsing identical sequences')
for record in parser:
try:
seq = str(record.seq)
if seq not in vdjs:
vdjs[seq] = VDJSequence(
ids=[],
sequence=seq,
quality=funcs.ord_to_quality(
record.letter_annotations.get('phred_quality')))
vdjs[seq].ids.append(record.description)
except ValueError:
continue
alignments = {}
aligner = AnchorAligner(self._v_germlines, self._j_germlines)
self.info('\tAligning {} unique sequences'.format(len(vdjs)))
# Attempt to align all unique sequences
for sequence in funcs.periodic_commit(self._session,
sorted(vdjs.keys())):
vdj = vdjs[sequence]
del vdjs[sequence]
try:
# The alignment was successful. If the aligned sequence
# already exists, append the seq_ids. Otherwise add it as a
# new unique sequence.
alignment = aligner.get_alignment(vdj)
seq_key = alignment.sequence.sequence
if seq_key in alignments:
alignments[seq_key].sequence.ids.extend(
alignment.sequence.ids)
else:
alignments[seq_key] = alignment
except AlignmentException as e:
add_as_noresult(self._session, vdj, sample, str(e))
except Exception:
self.error(
'\tUnexpected error processing sequence {}\n\t{}'.format(
vdj.ids[0], traceback.format_exc()))
if len(alignments) > 0:
avg_len = (sum([v.v_length for v in alignments.values()]) /
float(len(alignments)))
avg_mut = (
sum([v.v_mutation_fraction
for v in alignments.values()]) / float(len(alignments)))
sample.v_ties_mutations = avg_mut
sample.v_ties_len = avg_len
self.info('\tRe-aligning {} sequences to V-ties, Mutations={}, '
'Length={}'.format(len(alignments), round(avg_mut, 2),
round(avg_len, 2)))
add_uniques(self._session, sample, alignments.values(),
self._props, aligner, avg_len, avg_mut)
self._session.commit()
self.info('Completed sample {}'.format(sample.name))
def do_task(self, args):
meta = args['meta']
self.info('Starting sample {}'.format(meta.get('sample_name')))
study, sample = self._setup_sample(meta)
vdjs = {}
parser = SeqIO.parse(
os.path.join(args['path'], args['fn']),
'fasta' if args['fn'].endswith('.fasta') else 'fastq')
# Collapse identical sequences
self.info('\tCollapsing identical sequences')
for record in parser:
seq = str(record.seq)
if seq not in vdjs:
vdjs[seq] = VDJSequence(
ids=[],
seq=seq,
v_germlines=self._v_germlines,
j_germlines=self._j_germlines,
quality=funcs.ord_to_quality(
record.letter_annotations.get('phred_quality')))
vdjs[seq].ids.append(record.description)
self.info('\tAligning {} unique sequences'.format(len(vdjs)))
# Attempt to align all unique sequences
for sequence in funcs.periodic_commit(self._session,
sorted(vdjs.keys())):
vdj = vdjs[sequence]
del vdjs[sequence]
try:
# The alignment was successful. If the aligned sequence
# already exists, append the seq_ids. Otherwise add it as a
# new unique sequence.
vdj.analyze()
if vdj.sequence in vdjs:
vdjs[vdj.sequence].ids += vdj.ids
else:
vdjs[vdj.sequence] = vdj
except AlignmentException as e:
add_as_noresult(self._session, vdj, sample, str(e))
except:
self.error(
'\tUnexpected error processing sequence {}\n\t{}'.format(
vdj.ids[0], traceback.format_exc()))
if len(vdjs) > 0:
avg_len = sum(map(lambda vdj: vdj.v_length,
vdjs.values())) / float(len(vdjs))
avg_mut = sum(map(lambda vdj: vdj.mutation_fraction,
vdjs.values())) / float(len(vdjs))
sample.v_ties_mutations = avg_mut
sample.v_ties_len = avg_len
self.info('\tRe-aligning {} sequences to V-ties, Mutations={}, '
'Length={}'.format(len(vdjs), round(avg_mut, 2),
round(avg_len, 2)))
add_uniques(self._session, sample, vdjs.values(), avg_len, avg_mut,
self._min_similarity, self._max_vties, self._trim_to,
self._max_padding)
self._session.commit()
self.info('Completed sample {}'.format(sample.name))