def go(output_stream=sys.stdout, input_stream=sys.stdin,
verbose=False, report_multiplier=1.2,
alignment_count_to_report=1, tie_margin=0):
""" Processes Bowtie 2 alignments, emitting filtered SAM output.
Only max(# tied alignments, alignment_count_to_report) alignments
are printed. This way, the compare_alignments step always has enough
information to fill the XS field.
output_stream: where to emit exon and junction tuples; typically, this
is sys.stdout.
input_stream: where to find input to process
verbose: True if alignments should occasionally be written to stderr.
report_multiplier: if verbose is True, the line number of an
alignment written to stderr increases exponentially with base
report_multiplier.
alignment_count_to_report: argument of Bowtie 2's -k field
tie_margin: allowed score difference per 100 bases among ties in
max alignment score.
"""
output_line_count, next_report_line = 0, 0
threshold_alignment_count = max(2, alignment_count_to_report)
for (qname,), xpartition in xstream(input_stream, 1):
max_score, alignments_output, current_tie_margin = None, 0, None
for rest_of_line in xpartition:
# Note Bowtie 2 outputs alignments in order of descending score
try:
score = int([field[5:] for field in rest_of_line
if field[:5] == 'AS:i:'][0])
except IndexError:
# Unmapped read; flag should be 4. Print only essentials.
assert int(rest_of_line[0]) == 4
print >>output_stream, ('%s\t4\t\x1c\t\x1c\t\x1c\t\x1c'
'\t\x1c\t\x1c\t\x1c\t%s\t%s') % (
qname,
rest_of_line[8],
rest_of_line[9]
)
output_line_count += 1
else:
if current_tie_margin is None:
current_tie_margin = round(
tie_margin * float(len(rest_of_line[8])) / 100
)
if score + current_tie_margin >= max_score:
max_score = max(max_score, score)
elif alignments_output >= threshold_alignment_count:
break
print >>output_stream, '\t'.join((qname,) + rest_of_line)
alignments_output += 1
output_line_count += 1
if verbose and next_report_line == output_line_count:
print >>sys.stderr, \
'SAM output record %d: rdname="%s", flag=%d' \
% (output_line_count, qname, int(rest_of_line[0]))
next_report_line = max(int(next_report_line
* report_multiplier), next_report_line + 1)
output_stream.flush()
print >>sys.stderr, ('realign_reads_delegate.py reports %d output lines.'
% output_line_count)
def go(output_stream=sys.stdout, input_stream=sys.stdin,
verbose=False, report_multiplier=1.2,
alignment_count_to_report=1, tie_margin=0):
""" Processes Bowtie 2 alignments, emitting filtered SAM output.
Only max(# tied alignments, alignment_count_to_report) alignments
are printed. This way, the compare_alignments step always has enough
information to fill the XS field.
output_stream: where to emit exon and intron tuples; typically, this is
sys.stdout.
input_stream: where to find input to process
verbose: True if alignments should occasionally be written to stderr.
report_multiplier: if verbose is True, the line number of an
alignment written to stderr increases exponentially with base
report_multiplier.
alignment_count_to_report: argument of Bowtie 2's -k field
tie_margin: allowed score difference per 100 bases among ties in
max alignment score.
"""
output_line_count, next_report_line = 0, 0
threshold_alignment_count = max(2, alignment_count_to_report)
for (qname,), xpartition in xstream(input_stream, 1):
max_score, alignments_output, current_tie_margin = None, 0, None
for rest_of_line in xpartition:
# Note Bowtie 2 outputs alignments in order of descending score
try:
score = int([field[5:] for field in rest_of_line
if field[:5] == 'AS:i:'][0])
except IndexError:
# Unmapped read; flag should be 4. Print only essentials.
assert int(rest_of_line[0]) == 4
print >>output_stream, ('%s\t4\t\x1c\t\x1c\t\x1c\t\x1c'
'\t\x1c\t\x1c\t\x1c\t%s\t%s') % (
qname,
rest_of_line[8],
rest_of_line[9]
)
output_line_count += 1
else:
if current_tie_margin is None:
current_tie_margin = round(
tie_margin * float(len(rest_of_line[8])) / 100
)
if score + current_tie_margin >= max_score:
max_score = max(max_score, score)
elif alignments_output >= threshold_alignment_count:
break
print >>output_stream, '\t'.join((qname,) + rest_of_line)
alignments_output += 1
output_line_count += 1
if verbose and next_report_line == output_line_count:
print >>sys.stderr, \
'SAM output record %d: rdname="%s", flag=%d' \
% (output_line_count, qname, int(rest_of_line[0]))
next_report_line = max(int(next_report_line
* report_multiplier), next_report_line + 1)
output_stream.flush()
print >>sys.stderr, ('realign_reads_delegate.py reports %d output lines.'
% output_line_count)
def go(input_stream=sys.stdin, output_stream=sys.stdout, bowtie2_exe='bowtie2',
bowtie_index_base='genome', bowtie2_index_base='genome2',
manifest_file='manifest', bowtie2_args=None, bin_size=10000, verbose=False,
exon_differentials=True, exon_intervals=False, report_multiplier=1.2,
min_exon_size=8, search_filter=1, min_readlet_size=15, max_readlet_size=25,
readlet_interval=12, capping_multiplier=1.5, drop_deletions=False,
gzip_level=3, scratch=None, index_count=1, output_bam_by_chr=False,
tie_margin=0, no_realign=False, no_polyA=False):
""" Runs Rail-RNA-align_reads.
A single pass of Bowtie is run to find end-to-end alignments. Unmapped
reads are saved for readletizing to determine junctions in sucessive
reduce steps as well as for realignment in a later map step.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns in a mix of any of the following
three formats:
Format 1 (single-end, 3-column):
1. Nucleotide sequence or its reversed complement, whichever is first
in alphabetical order
2. 1 if sequence was reverse-complemented else 0
3. Name
4. Quality sequence or its reverse, whichever corresponds to field 1
Format 2 (paired, 2 lines, 3 columns each)
(so this is the same as single-end)
1. Nucleotide sequence for mate 1 or its reversed complement,
whichever is first in alphabetical order
2. 1 if sequence was reverse-complemented else 0
3. Name for mate 1
4. Quality sequence for mate 1 or its reverse, whichever corresponds
to field 1
(new line)
1. Nucleotide sequence for mate 2 or its reversed complement,
whichever is first in alphabetical order
2. 1 if sequence was reverse complemented else 0
3. Name for mate 2
4. Quality sequence for mate 2 or its reverse, whichever corresponds
to field 1
Input is partitioned and sorted by field 1, the read sequence.
Hadoop output (written to stdout)
----------------------------
A given RNAME sequence is partitioned into intervals ("bins") of some
user-specified length (see partition.py).
Exonic chunks (aka ECs; three formats, any or all of which may be
emitted):
Format 1 (exon_ival); tab-delimited output tuple columns:
1. Reference name (RNAME in SAM format) + ';' + bin number
2. Sample index
3. EC start (inclusive) on forward strand
4. EC end (exclusive) on forward strand
Format 2 (exon_diff); tab-delimited output tuple columns:
1. Reference name (RNAME in SAM format) + ';' + bin number
2. max(EC start, bin start) (inclusive) on forward strand IFF diff is
positive and EC end (exclusive) on forward strand IFF diff is
negative
3. Sample index
4. '1' if alignment from which diff originates is "unique" according to
--tie-margin criterion; else '0'
5. +1 or -1 * count, the number of instances of a read sequence for
which to print exonic chunks
Note that only unique alignments are currently output as ivals and/or
diffs.
Format 3 (sam); tab-delimited output tuple columns:
Standard SAM output except fields are in different order, and the first
field corresponds to sample label. (Fields are reordered to facilitate
partitioning by sample name/RNAME and sorting by POS.) Each line
corresponds to a spliced alignment. The order of the fields is as
follows.
1. Sample index if outputting BAMs by sample OR
sample-rname index if outputting BAMs by chr
2. (Number string representing RNAME; see BowtieIndexReference
class in bowtie_index for conversion information) OR
'0' if outputting BAMs by chr
3. POS
4. QNAME
5. FLAG
6. MAPQ
7. CIGAR
8. RNEXT
9. PNEXT
10. TLEN
11. SEQ
12. QUAL
... + optional fields
Insertions/deletions (indel_bed)
tab-delimited output tuple columns:
1. 'I' or 'D' insertion or deletion line
2. Number string representing RNAME
3. Start position (Last base before insertion or
first base of deletion)
4. End position (Last base before insertion or last base of deletion
(exclusive))
5. Inserted sequence for insertions or deleted sequence for deletions
6. Sample index
----Next fields are for junctions only; they are '\x1c' for indels----
7. '\x1c'
8. '\x1c'
--------------------------------------------------------------------
9. Number of instances of insertion or deletion in sample; this is
always +1 * count before bed_pre combiner/reducer
Read whose primary alignment is not end-to-end
Tab-delimited output tuple columns (unmapped):
1. Transcriptome Bowtie 2 index group number
2. SEQ
3. 1 if SEQ is reverse-complemented, else 0
4. QNAME
5. QUAL
Tab-delimited output tuple columns (readletized):
1. Readlet sequence or its reversed complement, whichever is first in
alphabetical order
2. read sequence ID + ('-' if readlet
sequence is reverse-complemented; else '+') + '\x1e' + displacement
of readlet's 5' end from read's 5' end + '\x1e' + displacement of
readlet's 3' end from read's 3' end (+, for EXACTLY one readlet of
a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence)) + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective sample in list
A) + '\x1e' + (an '\x1f'-separated list of the number of instances
of the read sequence's reversed complement for each respective
sample in list B). Here, a read sequence ID takes the form X:Y,
where X is the "mapred_task_partition" environment variable -- a
unique index for a task within a job -- and Y is the index of the
read sequence relative to the beginning of the input stream.
Tab-delimited tuple columns (postponed_sam):
Standard 11+ -column raw SAM output
Single column (unique):
1. A unique read sequence
Two columns, exactly one line (dummy); ensures creation of junction
index:
1. character "-"
2. the word "dummy"
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input reads.
output_stream: where to emit exonic chunks and junctions.
bowtie2_exe: filename of Bowtie2 executable; include path if not in
$PATH.
bowtie_index_base: the basename of the Bowtie1 index files associated
with the reference.
bowtie2_index_base: the basename of the Bowtie2 index files associated
with the reference.
manifest_file: filename of manifest
bowtie2_args: string containing precisely extra command-line arguments
to pass to first-pass Bowtie2.
bin_size: genome is partitioned in units of bin_size for later load
balancing.
verbose: True iff more informative messages should be written to
stderr.
exon_differentials: True iff EC differentials are to be emitted.
exon_intervals: True iff EC intervals are to be emitted.
report_multiplier: if verbose is True, the line number of an alignment
or read written to stderr increases exponentially with base
report_multiplier.
min_exon_size: minimum exon size searched for in junction_search.py
later in pipeline; used to determine how large a soft clip on one
side of a read is necessary to pass it on to junction search
pipeline
search_filter: how large a soft clip on one side of a read is necessary
to pass it on to junction search pipeline
min_readlet_size: "capping" readlets (that is, readlets that terminate
at a given end of the read) are never smaller than this value
max_readlet_size: size of every noncapping readlet
readlet_interval: number of bases separating successive readlets along
the read
capping_multiplier: successive capping readlets on a given end of a
read are increased in size exponentially with base
capping_multiplier
drop_deletions: True iff deletions should be dropped from coverage
vector
gzip_level: compression level to use for temporary files
scratch: scratch directory for storing temporary files or None if
securely created temporary directory
index_count: number of transcriptome Bowtie 2 indexes to which to
assign unmapped reads for later realignment
output_bam_by_chr: True iff final output BAMs will be by chromosome
tie_margin: allowed score difference per 100 bases among ties in
max score. For example, 150 and 144 are tied alignment scores
for a 100-bp read when --tie-margin is 6.
no_realign: True iff job flow does not need more than readlets: this
usually means only a transcript index is being constructed
no_polyA: kill noncapping readlets that are all As and write as
unmapped all reads with polyA prefixes whose suffixes are <
min_exon_size
No return value.
"""
global _input_line_count
reference_index = bowtie_index.BowtieIndexReference(bowtie_index_base)
manifest_object = manifest.LabelsAndIndices(manifest_file)
alignment_printer = AlignmentPrinter(
manifest_object,
reference_index,
bin_size=bin_size,
output_stream=output_stream,
exon_ivals=exon_intervals,
exon_diffs=exon_differentials,
drop_deletions=drop_deletions,
output_bam_by_chr=output_bam_by_chr,
tie_margin=tie_margin
)
# Get task partition to pass to align_reads_delegate.py
try:
task_partition = os.environ['mapred_task_partition']
except KeyError:
# Hadoop 2.x?
try:
task_partition = os.environ['mapreduce_task_partition']
except KeyError:
# A unit test is probably being run
task_partition = '0'
temp_dir = make_temp_dir(scratch)
register_cleanup(tempdel.remove_temporary_directories, [temp_dir])
align_file = os.path.join(temp_dir, 'first_pass_reads.temp.gz')
other_reads_file = os.path.join(temp_dir, 'other_reads.temp.gz')
second_pass_file = os.path.join(temp_dir, 'second_pass_reads.temp.gz')
k_value, _, _ = bowtie.parsed_bowtie_args(bowtie2_args)
nothing_doing = True
# Required length of prefix after poly(A) is trimmed
remaining_seq_size = max(min_exon_size - 1, 1)
with xopen(True, align_file, 'w', gzip_level) as align_stream, \
xopen(True, other_reads_file, 'w', gzip_level) as other_stream:
for seq_number, ((seq,), xpartition) in enumerate(
xstream(sys.stdin, 1)
):
seq_length = len(seq)
if no_polyA and (
all(seq[i] == 'A'
for i in xrange(seq_length - remaining_seq_size))
or all(seq[i] == 'T'
for i in xrange(remaining_seq_size, seq_length))
or all(seq[i] == 'A'
for i in xrange(remaining_seq_size, seq_length))
or all(seq[i] == 'T'
for i in xrange(seq_length - remaining_seq_size))
):
if not no_realign:
'''If a sequence is too short without its poly(A) tail,
make all reads with that sequence unmapped. Technically,
this also kills poly(A)s at 5' ends, but we probably
couldn't align those sequences anyway.'''
reversed_complement_seq = seq[::-1].translate(
_reversed_complement_translation_table
)
for is_reversed, name, qual in xpartition:
if is_reversed == '0':
alignment_printer.print_unmapped_read(
name,
seq,
qual
)
else:
alignment_printer.print_unmapped_read(
name,
reversed_complement_seq,
qual[::-1]
)
continue
nothing_doing = False
'''Select highest-quality read with alphabetically last qname
for first-pass alignment.'''
best_name, best_mean_qual, best_qual_index, i = None, None, 0, 0
others_to_print = dlist()
for is_reversed, name, qual in xpartition:
_input_line_count += 1
others_to_print.append(
'\t'.join([
str(seq_number), is_reversed, name, qual
])
)
mean_qual = (
float(sum([ord(score) for score in qual])) / len(qual)
)
if (mean_qual > best_mean_qual
or mean_qual == best_mean_qual and name > best_name):
best_qual_index = i
best_mean_qual = mean_qual
best_name = name
to_align = '\t'.join([
'%s\x1d%s' % (is_reversed, name),
seq, qual
])
i += 1
assert i >= 1
if i == 1:
print >>other_stream, str(seq_number)
else:
for j, other_to_print in enumerate(others_to_print):
if j != best_qual_index:
print >>other_stream, other_to_print
print >>align_stream, to_align
# Print dummy line
print 'dummy\t-\tdummy'
sys.stdout.flush() # this is REALLY important b/c called script will stdout
if nothing_doing:
# No input
sys.exit(0)
input_command = 'gzip -cd %s' % align_file
bowtie_command = ' '.join([bowtie2_exe,
bowtie2_args if bowtie2_args is not None else '',
' --sam-no-qname-trunc --local -t --no-hd --mm -x',
bowtie2_index_base, '--12 -'])
delegate_command = ''.join(
[sys.executable, ' ', os.path.realpath(__file__)[:-3],
('_delegate.py --task-partition {task_partition} '
'--other-reads {other_reads} --second-pass-reads '
'{second_pass_reads} --min-readlet-size '
'{min_readlet_size} {drop_deletions} '
'--max-readlet-size {max_readlet_size} '
'--readlet-interval {readlet_interval} '
'--capping-multiplier {capping_multiplier:1.12f} '
'{verbose} --report-multiplier {report_multiplier:1.12f} '
'--k-value {k_value} '
'--bowtie-idx {bowtie_index_base} '
'--partition-length {bin_size} '
'--manifest {manifest_file} '
'{exon_differentials} {exon_intervals} '
'--gzip-level {gzip_level} '
'--search-filter {search_filter} '
'--index-count {index_count} '
'--tie-margin {tie_margin} '
'{no_realign} '
'{no_polyA} '
'{output_bam_by_chr}').format(
task_partition=task_partition,
other_reads=other_reads_file,
second_pass_reads=second_pass_file,
min_readlet_size=min_readlet_size,
drop_deletions=('--drop-deletions' if drop_deletions
else ''),
max_readlet_size=max_readlet_size,
readlet_interval=readlet_interval,
capping_multiplier=capping_multiplier,
verbose=('--verbose' if verbose else ''),
report_multiplier=report_multiplier,
k_value=k_value,
bowtie_index_base=bowtie_index_base,
bin_size=bin_size,
manifest_file=manifest_file,
exon_differentials=('--exon-differentials'
if exon_differentials else ''),
exon_intervals=('--exon-intervals'
if exon_intervals else ''),
gzip_level=gzip_level,
search_filter=search_filter,
index_count=index_count,
tie_margin=tie_margin,
no_realign=('--no-realign' if no_realign else ''),
no_polyA=('--no-polyA' if no_polyA else ''),
output_bam_by_chr=('--output-bam-by-chr'
if output_bam_by_chr
else '')
)]
)
full_command = ' | '.join([input_command,
bowtie_command, delegate_command])
print >>sys.stderr, \
'Starting first-pass Bowtie 2 with command: ' + full_command
bowtie_process = subprocess.Popen(' '.join(
['set -exo pipefail;', full_command]
),
bufsize=-1, stdout=sys.stdout, stderr=sys.stderr, shell=True,
executable='/bin/bash')
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError('Error occurred while reading first-pass Bowtie 2 '
'output; exitlevel was %d.' % return_code)
os.remove(align_file)
os.remove(other_reads_file)
if not no_realign:
input_command = 'gzip -cd %s' % second_pass_file
bowtie_command = ' '.join([bowtie2_exe,
bowtie2_args if bowtie2_args is not None else '',
' --sam-no-qname-trunc --local -t --no-hd --mm -x',
bowtie2_index_base, '--12 -'])
delegate_command = ''.join(
[sys.executable, ' ', os.path.realpath(__file__)[:-3],
('_delegate.py --task-partition {task_partition} '
'--min-readlet-size {min_readlet_size} '
'{drop_deletions} '
'--max-readlet-size {max_readlet_size} '
'--readlet-interval {readlet_interval} '
'--capping-multiplier {capping_multiplier:012f} '
'{verbose} '
'--report-multiplier {report_multiplier:012f} '
'--k-value {k_value} '
'--bowtie-idx {bowtie_index_base} '
'--partition-length {bin_size} '
'--manifest {manifest_file} '
'{exon_differentials} {exon_intervals} '
'--gzip-level {gzip_level} '
'--search-filter {search_filter} '
'--index-count {index_count} '
'--tie-margin {tie_margin} '
'{output_bam_by_chr}').format(
task_partition=task_partition,
min_readlet_size=min_readlet_size,
drop_deletions=('--drop-deletions'
if drop_deletions else ''),
readlet_interval=readlet_interval,
max_readlet_size=max_readlet_size,
capping_multiplier=capping_multiplier,
verbose=('--verbose' if verbose else ''),
report_multiplier=report_multiplier,
k_value=k_value,
bowtie_index_base=bowtie_index_base,
bin_size=bin_size,
manifest_file=manifest_file,
exon_differentials=('--exon-differentials'
if exon_differentials else ''),
exon_intervals=('--exon-intervals'
if exon_intervals else ''),
gzip_level=gzip_level,
search_filter=search_filter,
index_count=index_count,
tie_margin=tie_margin,
output_bam_by_chr=('--output-bam-by-chr'
if output_bam_by_chr
else '')
)]
)
full_command = ' | '.join([input_command,
bowtie_command, delegate_command])
print >>sys.stderr, \
'Starting second-pass Bowtie 2 with command: ' + full_command
bowtie_process = subprocess.Popen(' '.join(
['set -exo pipefail;', full_command]
),
bufsize=-1, stdout=sys.stdout, stderr=sys.stderr, shell=True,
executable='/bin/bash')
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError('Error occurred while reading second-pass '
'Bowtie 2 output; exitlevel was %d.'
% return_code)
sys.stdout.flush()
def go(input_stream=sys.stdin,
output_stream=sys.stdout,
fudge=5,
stranded=False,
verbose=False,
report_multiplier=1.2):
""" Emits intron combinations associated with reads.
Soft-clipped Bowtie 2 alignments of read sequences to the transcript
fragment index are used infer which cointrons could possibly be
overlapped by reads. Then maximal cliques of the graph described in
the maximal_cliques() function are enumerated to obtain which
intron combinations could possibly be overlapped by reads.
input_stream: where to retrieve Bowtie 2 output
output_stream: where to emit exon and intron tuples; typically, this is
sys.stdout.
verbose: True if alignments should occasionally be written to stderr.
stranded: True iff input reads are strand-specific; this affects
whether an output partition has a terminal '+' or '-' indicating
the sense strand. Further, if stranded is True, an alignment is
returned only if its strand agrees with the intron's strand.
fudge: by how many bases to extend left and right extend sizes
to accommodate potential indels
report_multiplier: if verbose is True, the line number of an
alignment written to stderr increases exponentially with base
report_multiplier.
"""
output_line_count, next_report_line, i = 0, 0, 0
for (qname, ), xpartition in xstream(input_stream, 1):
'''While labeled multireadlet, this list may end up simply a
unireadlet.'''
multiread = []
for tokens in xpartition:
flag = int(tokens[0])
if verbose and next_report_line == i:
print >>sys.stderr, \
'SAM output record %d: rdname="%s", flag=%d' % (i,
qname,
flag)
next_report_line = int(
(next_report_line + 1) * report_multiplier + 1) - 1
i += 1
multiread.append((qname, ) + tokens)
if flag & 4: continue
corrected_multiread = multiread_with_introns(multiread, stranded)
all_introns = {}
for alignment in multiread_with_introns(multiread, stranded):
cigar = alignment[5]
md = [field for field in alignment if field[:5] == 'MD:Z:'][0][5:]
pos = int(alignment[3])
seq = alignment[9]
reversed_complement_seq = seq[::-1].translate(
_reversed_complement_translation_table)
if seq < reversed_complement_seq:
seq_to_print = seq
else:
seq_to_print = reversed_complement_seq
seq_size = len(seq)
rname = alignment[2]
sense = [field for field in alignment
if field[:5] == 'XS:A:'][0][5:]
if (rname, sense) not in all_introns:
all_introns[(rname, sense)] = defaultdict(list)
_, _, introns, _ = indels_introns_and_exons(cigar, md, pos, seq)
for intron in introns:
if (intron[0], intron[1]) \
not in all_introns[(rname, sense)]:
all_introns[(rname, sense)][(intron[0], intron[1])] \
= [intron[2], intron[3]]
else:
all_introns[(rname,
sense)][(intron[0], intron[1])][0] = max(
all_introns[(rname,
sense)][(intron[0],
intron[1])][0],
intron[2])
all_introns[(rname,
sense)][(intron[0], intron[1])][1] = max(
all_introns[(rname,
sense)][(intron[0],
intron[1])][1],
intron[3])
for rname, sense in all_introns:
to_write = set()
# Grab maximal cliques
for clique in \
maximal_cliques(all_introns[(rname, sense)].keys()):
for cointrons in separated_introns(clique,
separation=(seq_size +
fudge)):
cointrons.sort()
left_extend_size = all_introns[(rname, sense)][(
cointrons[0][0], cointrons[0][1])][0]
right_extend_size = all_introns[(rname, sense)][(
cointrons[-1][0], cointrons[-1][1])][1]
to_write.add(
('{rname}{sense}\t{starts}'
'\t{ends}\t{left_size}'
'\t{right_size}\t{seq}').format(
rname=rname,
sense=sense,
starts=','.join(
[str(intron[0]) for intron in cointrons]),
ends=','.join(
[str(intron[1]) for intron in cointrons]),
left_size=(left_extend_size + fudge),
right_size=(right_extend_size + fudge),
seq=seq_to_print))
for line_to_write in to_write:
print line_to_write
output_line_count += 1
output_stream.flush()
print >> sys.stderr, (
'cointron_enum_delegate.py reports %d output lines.' %
output_line_count)
output_url = Url(args.out) if args.out is not None \
else Url(os.getcwd())
input_line_count = 0
if output_url.is_local:
# Set up destination directory
try:
os.makedirs(output_url.to_url())
except:
pass
else:
mover = filemover.FileMover(args=args)
# Set up temporary destination
import tempfile
temp_dir_path = make_temp_dir(tempdel.silentexpandvars(args.scratch))
register_cleanup(tempdel.remove_temporary_directories, [temp_dir_path])
for (line_type, sample_label), xpartition in xstream(sys.stdin, 2):
assert line_type in 'NID'
sample_label = manifest_object.index_to_label[sample_label]
type_string = ('insertions' if line_type == 'I' else
('deletions' if line_type == 'D' else 'junctions'))
output_filename = (
(args.bed_basename + '.' if args.bed_basename != '' else '') +
type_string + '.' + sample_label + '.bed')
if output_url.is_local:
output_path = os.path.join(args.out, output_filename)
else:
output_path = os.path.join(temp_dir_path, output_filename)
with open(output_path, 'w') as output_stream:
print >>output_stream, 'track name="%s_%s" description="' \
'Rail-RNA v%s %s for sample %s"' \
% (sample_label,
instance=args.read_instance,
)
write_read_introns_from_sam_stream(
sys.stdin, combined_stream, retrieved_intron_counts, instance=args.read_instance
)
import subprocess
sorted_combined_file = os.path.join(temp_dir_path, "combined.sorted.temp")
subprocess.check_call(
" ".join(["sort -T %s -k1,1" % temp_dir_path, combined_file, ">", sorted_combined_file]), bufsize=-1, shell=True
)
relevant = 0
retrieved = 0
relevant_and_retrieved = 0
with open(sorted_combined_file) as sorted_combined_stream:
for (name,), xpartition in xstream(sorted_combined_stream, 1):
relevant_and_retrieved_instances = list(xpartition)
ts = [
instance[:-1]
for instance in relevant_and_retrieved_instances
if instance[-1] == "t"
and (
args.coverage_threshold is None
or any([intron_counts[intron] <= args.coverage_threshold for intron in instance[:-1]])
)
]
rs = [
instance[:-1]
for instance in relevant_and_retrieved_instances
if instance[-1] == "r"
and (
output_url = Url(args.out) if args.out is not None \
else Url(os.getcwd())
input_line_count = 0
if output_url.is_local:
# Set up destination directory
try: os.makedirs(output_url.to_url())
except: pass
else:
mover = filemover.FileMover(args=args)
# Set up temporary destination
import tempfile
temp_dir_path = make_temp_dir(tempdel.silentexpandvars(args.scratch))
register_cleanup(tempdel.remove_temporary_directories, [temp_dir_path])
input_line_count = 0
for (line_type,), xpartition in xstream(sys.stdin, 1):
type_string = ('insertions' if line_type == '0' else
('deletions' if line_type == '1' else
('junctions' if line_type == '2' else
'normalization')))
output_filename = ((args.tsv_basename + '.'
if args.tsv_basename != '' else '')
+ type_string + '.tsv.gz')
if output_url.is_local:
output_path = os.path.join(args.out, output_filename)
else:
output_path = os.path.join(temp_dir_path, output_filename)
with xopen(True, output_path, 'w', args.gzip_level) as output_stream:
if line_type != '3':
'''Print all labels in the order in which they appear in the
manifest file.'''
def go(
input_stream=sys.stdin,
output_stream=sys.stdout,
bowtie_exe="bowtie",
bowtie_index_base="genome",
bowtie_args="",
gzip_level=3,
verbose=False,
report_multiplier=1.2,
scratch=None,
):
""" Runs Rail-RNA-align_readlets.
Aligns input readlet sequences and writes a single output line per
readlet belonging to a distinct read sequence.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns:
1. Readlet sequence or its reversed complement, whichever is first in
alphabetical order
2. read sequence ID + ('-' if readlet
sequence is reverse-complemented; else '+') + '\x1e' + displacement
of readlet's 5' end from read's 5' end + '\x1e' + displacement of
readlet's 3' end from read's 3' end (+, for EXACTLY one readlet of
a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence)) + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective sample in list
A) + '\x1e' + (an '\x1f'-separated list of the number of instances
of the read sequence's reversed complement for each respective
sample in list B). Here, a read sequence ID takes the form X:Y,
where X is the "mapred_task_partition" environment variable -- a
unique index for a task within a job -- and Y is the index of the
read sequence relative to the beginning of the input stream.
Input is partitioned by field 1, the readlet sequence or its reversed
complement.
Hadoop output (written to stdout)
----------------------------
Tab-delimited output tuple columns, where each line corresponds to a
readlet from a distinct read rather than a unique readlet sequence:
1. Read sequence ID
2. Displacement of readlet's 5' end from read's 5' end + '\x1e' +
displacement of readlet's 3' end from read's 3' end (+, for EXACTLY
one readlet of a read sequence, '\x1e' + read sequence + '\x1e' +
number of instances of read sequence + '\x1e' + number of instances
of read sequence's reversed complement + '\x1e' (+, for EXACTLY one
readlet of a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence))] + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective
sample in list A) + '\x1e' + (an '\x1f'-separated list of the
number of instances of the read sequence's reversed complement for
each respective sample in list B)
3. '\x1f'-separated list of alignment RNAMEs or '\x1c' if no alignments
4. '\x1f'-separated list of alignment FLAGs or '\x1c' if no alignments
5. '\x1f-separated list of alignment POSes or '\x1c' if no alignments
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input reads.
output_stream: where to emit exonic chunks and introns.
bowtie_exe: filename of Bowtie executable; include path if not in
$PATH.
bowtie_index_base: the basename of the Bowtie index files associated
with the reference.
bowtie_args: string containing precisely extra command-line arguments
to pass to first-pass Bowtie, e.g., "--tryhard --best"; or None.
gzip_level: level of gzip compression to use for qname file
verbose: True iff more informative messages should be written to
stderr.
report_multiplier: if verbose is True, the line number of an alignment
written to stderr increases exponentially with base
report_multiplier.
scratch: scratch directory for storing temporary files or None if
securely created temporary directory
No return value.
"""
global _input_line_count
# For storing long qnames
temp_dir = make_temp_dir(scratch)
register_cleanup(tempdel.remove_temporary_directories, [temp_dir])
qnames_file = os.path.join(temp_dir, "qnames.temp.gz")
readlet_file = os.path.join(temp_dir, "readlets.temp.gz")
with xopen(True, qnames_file, "w", gzip_level) as qname_stream:
with xopen(True, readlet_file, "w", gzip_level) as readlet_stream:
for (seq_count, ((seq,), xpartition)) in enumerate(xstream(input_stream, 1)):
print >> readlet_stream, "\t".join([str(seq_count), seq, "I" * len(seq)])
print >> qname_stream, next(iter(xpartition))[0]
for (qname,) in xpartition:
_input_line_count += 1
print >> qname_stream, qname
# Separate qnames with single + character
print >> qname_stream, "+"
input_command = "gzip -cd %s" % readlet_file
bowtie_command = " ".join([bowtie_exe, bowtie_args, "-S -t --sam-nohead --mm", bowtie_index_base, "--12 -"])
delegate_command = "".join(
[
sys.executable,
" ",
os.path.realpath(__file__)[:-3],
"_delegate.py --report-multiplier %08f --qnames-file %s %s"
% (report_multiplier, qnames_file, "--verbose" if verbose else ""),
]
)
full_command = " | ".join([input_command, bowtie_command, delegate_command])
print >>sys.stderr, "Starting Bowtie with command: " + full_command
bowtie_process = subprocess.Popen(
" ".join(["set -exo pipefail;", full_command]),
bufsize=-1,
stdout=sys.stdout,
stderr=sys.stderr,
shell=True,
executable="/bin/bash",
)
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError("Error occurred while reading Bowtie output; " "exitlevel was %d." % return_code)
unique_mapped_read_counts[sample_index]) = [
int(token) for token
in tokens[-2].split(',')
]
try:
mean_weight = 1. / len([_ for _ in mapped_read_counts.values() if _])
except ZeroDivisionError:
mean_weight = 0.0
try:
unique_mean_weight = 1. / len(
[_ for _ in unique_mapped_read_counts.values() if _]
)
except ZeroDivisionError:
unique_mean_weight = 0.0
for (partition_id,), xpartition in xstream(sys.stdin, 1):
bin_count += 1
bin_start_time, bin_diff_count = time.time(), 0
rname = partition_id.rpartition(';')[0]
rname_index = reference_index.l_rname_to_string[rname]
coverages, unique_coverages = defaultdict(int), defaultdict(int)
for (pos, sample_indexes_and_diffs) in itertools.groupby(
xpartition, lambda val: val[0]
):
input_line_count += 1
pos = int(pos)
for sample_index, diffs in itertools.groupby(
sample_indexes_and_diffs, lambda val: val[1]
):
for _, _, uniqueness, diff in diffs:
coverages[sample_index] += int(diff)
import bowtie_index
from dooplicity.tools import xstream
# Print file's docstring if -h is invoked
parser = argparse.ArgumentParser(
description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
bowtie.add_args(parser)
args = parser.parse_args()
input_line_count, output_line_count = 0, 0
start_time = time.time()
reference_index = bowtie_index.BowtieIndexReference(
os.path.expandvars(args.bowtie_idx))
for (_, rname_string, intron_pos, intron_end_pos, sense,
sample_index), xpartition in xstream(sys.stdin, 6):
coverage = 0
for value in xpartition:
input_line_count += 1
try:
# Assume intron line
_, _, instance_count = value
except ValueError:
# Alignment line
print('\t'.join((value[2], value[3],
reference_index.string_to_rname[rname_string],
str(int(value[1]))) + value[4:]) +
('\tXC:i:%d' % coverage))
output_line_count += 1
else:
coverage += int(instance_count)
from dooplicity.tools import xstream
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument(\
'--verbose', action='store_const', const=True, default=False,
help='Print out extra debugging statements')
bowtie.add_args(parser)
args = parser.parse_args()
start_time = time.time()
input_line_count = 0
reference_index = bowtie_index.BowtieIndexReference(
os.path.expandvars(args.bowtie_idx)
)
for key, xpartition in xstream(sys.stdin, 3, skip_duplicates=True):
'''For computing maximum left and right extend sizes for every key --
that is, every intron combo (fields 1-3 of input).'''
left_extend_size, right_extend_size = None, None
left_size, right_size = None, None
for value in xpartition:
assert len(value) == 4
input_line_count += 1
left_extend_size = max(left_extend_size, int(value[-4]))
right_extend_size = max(right_extend_size, int(value[-3]))
try:
left_size = max(left_size, int(value[-2]))
except ValueError:
left_size = 'NA'
try:
right_size = max(right_size, int(value[-1]))
sample_index = manifest_object.label_to_index[tokens[0]]
(mapped_read_counts[sample_index],
unique_mapped_read_counts[sample_index]) = [
int(token) for token in tokens[-2].split(',')
]
try:
mean_weight = 1. / len([_ for _ in mapped_read_counts.values() if _])
except ZeroDivisionError:
mean_weight = 0.0
try:
unique_mean_weight = 1. / len(
[_ for _ in unique_mapped_read_counts.values() if _])
except ZeroDivisionError:
unique_mean_weight = 0.0
for (partition_id, ), xpartition in xstream(sys.stdin, 1):
bin_count += 1
bin_start_time, bin_diff_count = time.time(), 0
rname = partition_id.rpartition(';')[0]
rname_index = reference_index.l_rname_to_string[rname]
coverages, unique_coverages = defaultdict(int), defaultdict(int)
for (pos, sample_indexes_and_diffs) in itertools.groupby(
xpartition, lambda val: val[0]):
input_line_count += 1
pos = int(pos)
for sample_index, diffs in itertools.groupby(sample_indexes_and_diffs,
lambda val: val[1]):
for _, _, uniqueness, diff in diffs:
coverages[sample_index] += int(diff)
if uniqueness == '1':
unique_coverages[sample_index] += int(diff)
parser.add_argument(\
'--verbose', action='store_const', const=True, default=False,
help='Print out extra debugging statements')
bowtie.add_args(parser)
group_reads.add_args(parser)
args = parser.parse_args()
start_time = time.time()
input_line_count = 0
counter = Counter('cojunction_fasta')
register_cleanup(counter.flush)
reference_index = bowtie_index.BowtieIndexReference(
os.path.expandvars(args.bowtie_idx))
group_reads_object = group_reads.IndexGroup(args.index_count)
for (rname, poses, end_poses), xpartition in xstream(sys.stdin,
3,
skip_duplicates=True):
counter.add('partitions')
reverse_strand_string = rname[-1]
rname = rname[:-1]
read_seqs = dlist()
poses = [int(pos) for pos in poses.split(',')]
end_poses = [int(end_pos) for end_pos in end_poses.split(',')]
max_left_extend_size, max_right_extend_size = None, None
for left_extend_size, right_extend_size, read_seq in xpartition:
counter.add('inputs')
input_line_count += 1
max_left_extend_size = max(max_left_extend_size, int(left_extend_size))
max_right_extend_size \
= max(max_right_extend_size, int(right_extend_size))
read_seqs.append(read_seq)
# To convert sample-rname index to sample index-rname index tuple
sample_and_rname_indexes = SampleAndRnameIndexes(
manifest_object,
args.output_by_chromosome
)
import time
start_time = time.time()
from alignment_handlers import AlignmentPrinter
alignment_printer = AlignmentPrinter(manifest_object, reference_index,
tie_margin=args.tie_margin)
input_line_count = 0
if args.suppress_bam:
# Just grab stats
if args.output_by_chromosome:
for (index, _), xpartition in xstream(sys.stdin, 2):
sample_index, rname_index = (
sample_and_rname_indexes.sample_and_rname_indexes(index)
)
unique_count, total_count = 0, 0
for record in xpartition:
if not (int(record[2]) & 256):
total_count += 1
try:
# seq is at position 8
if alignment_printer.unique(record, seq_index=8):
unique_count += 1
except IndexError:
# Unmapped read; it's unique
unique_count += 1
input_line_count += 1
def go(input_stream=sys.stdin,
output_stream=sys.stdout,
bowtie_exe='bowtie',
bowtie_index_base='genome',
bowtie_args='',
gzip_level=3,
verbose=False,
report_multiplier=1.2,
scratch=None):
""" Runs Rail-RNA-align_readlets.
Aligns input readlet sequences and writes a single output line per
readlet belonging to a distinct read sequence.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns:
1. Readlet sequence or its reversed complement, whichever is first in
alphabetical order
2. read sequence ID + ('-' if readlet
sequence is reverse-complemented; else '+') + '\x1e' + displacement
of readlet's 5' end from read's 5' end + '\x1e' + displacement of
readlet's 3' end from read's 3' end (+, for EXACTLY one readlet of
a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence)) + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective sample in list
A) + '\x1e' + (an '\x1f'-separated list of the number of instances
of the read sequence's reversed complement for each respective
sample in list B). Here, a read sequence ID takes the form X:Y,
where X is the "mapred_task_partition" environment variable -- a
unique index for a task within a job -- and Y is the index of the
read sequence relative to the beginning of the input stream.
Input is partitioned by field 1, the readlet sequence or its reversed
complement.
Hadoop output (written to stdout)
----------------------------
Tab-delimited output tuple columns, where each line corresponds to a
readlet from a distinct read rather than a unique readlet sequence:
1. Read sequence ID
2. Displacement of readlet's 5' end from read's 5' end + '\x1e' +
displacement of readlet's 3' end from read's 3' end (+, for EXACTLY
one readlet of a read sequence, '\x1e' + read sequence + '\x1e' +
number of instances of read sequence + '\x1e' + number of instances
of read sequence's reversed complement + '\x1e' (+, for EXACTLY one
readlet of a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence))] + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective
sample in list A) + '\x1e' + (an '\x1f'-separated list of the
number of instances of the read sequence's reversed complement for
each respective sample in list B)
3. '\x1f'-separated list of alignment RNAMEs or '\x1c' if no alignments
4. '\x1f'-separated list of alignment FLAGs or '\x1c' if no alignments
5. '\x1f-separated list of alignment POSes or '\x1c' if no alignments
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input reads.
output_stream: where to emit exonic chunks and introns.
bowtie_exe: filename of Bowtie executable; include path if not in
$PATH.
bowtie_index_base: the basename of the Bowtie index files associated
with the reference.
bowtie_args: string containing precisely extra command-line arguments
to pass to first-pass Bowtie, e.g., "--tryhard --best"; or None.
gzip_level: level of gzip compression to use for qname file
verbose: True iff more informative messages should be written to
stderr.
report_multiplier: if verbose is True, the line number of an alignment
written to stderr increases exponentially with base
report_multiplier.
scratch: scratch directory for storing temporary files or None if
securely created temporary directory
No return value.
"""
global _input_line_count
# For storing long qnames
temp_dir = make_temp_dir(scratch)
register_cleanup(tempdel.remove_temporary_directories, [temp_dir])
qnames_file = os.path.join(temp_dir, 'qnames.temp.gz')
readlet_file = os.path.join(temp_dir, 'readlets.temp.gz')
with xopen(True, qnames_file, 'w', gzip_level) as qname_stream:
with xopen(True, readlet_file, 'w', gzip_level) as readlet_stream:
for (seq_count, ((seq,), xpartition)) \
in enumerate(xstream(input_stream, 1)):
print >>readlet_stream, \
'\t'.join([str(seq_count), seq, 'I'*len(seq)])
print >> qname_stream, next(iter(xpartition))[0]
for (qname, ) in xpartition:
_input_line_count += 1
print >> qname_stream, qname
# Separate qnames with single + character
print >> qname_stream, '+'
input_command = 'gzip -cd %s' % readlet_file
bowtie_command = ' '.join([
bowtie_exe, bowtie_args, '-S -t --sam-nohead --mm', bowtie_index_base,
'--12 -'
])
delegate_command = ''.join([
sys.executable, ' ',
os.path.realpath(__file__)[:-3],
'_delegate.py --report-multiplier %08f --qnames-file %s %s' %
(report_multiplier, qnames_file, '--verbose' if verbose else '')
])
full_command = ' | '.join(
[input_command, bowtie_command, delegate_command])
print >> sys.stderr, 'Starting Bowtie with command: ' + full_command
bowtie_process = subprocess.Popen(' '.join(
['set -exo pipefail;', full_command]),
bufsize=-1,
stdout=sys.stdout,
stderr=sys.stderr,
shell=True,
executable='/bin/bash')
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError('Error occurred while reading Bowtie output; '
'exitlevel was %d.' % return_code)
def go(manifest_object, input_stream=sys.stdin, output_stream=sys.stdout,
sample_fraction=0.05, coverage_threshold=5, verbose=False):
""" Runs Rail-RNA-bed_pre
Writes indels and junctions for outputting BEDs by sample and
TSVs across samples.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns:
1. 'I', 'D', or 'N' for insertion, deletion, or junction line
2. Number string representing RNAME
3. Start position (Last base before insertion, first base of deletion,
or first base of intron)
4. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
5. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
6. Sample index
----Next fields are for junctions only; they are '\x1c' for indels----
7. Number of nucleotides between 5' end of intron and 5' end of read
from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND. That is, if the sense strand is the reverse strand,
this is the distance between the 3' end of the read and the 3' end
of the intron.
8. Number of nucleotides between 3' end of intron and 3' end of read
from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND.
--------------------------------------------------------------------
9. Number of instances of junction, insertion, or deletion in sample;
this is always +1 before bed_pre combiner/reducer
Input is partitioned by fields 1-5 and sorted by field 6.
Hadoop output (written to stdout)
----------------------------
Tab-delimited output tuple columns (bed):
1. 'I', 'D', or 'N' for insertion, deletion, or junction line
2. Sample index
3. Number string representing RNAME (+ '+ or -' if junction; same as
field 6)
4. Start position (Last base before insertion, first base of deletion,
or first base of intron)
5. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
6. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
----Next fields are for junctions only; they are '\x1c' for indels----
7. MAX number of nucleotides between 5' end of intron and 5' end of
read from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND. That is, if the sense strand is the reverse strand,
this is the distance between the 3' end of the read and the 3' end
of the intron.
8. MAX number of nucleotides between 3' end of intron and 3' end of
read from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND.
9. MAXIMIN (number of nucleotides between 5' end of intron and 5' end
of read, number of nucleotides between 3' end of intron and
3' end of read);
min is between the args above; max is across reads.
Tab-delimited output tuple columns (collect)
1. '0' if insertion, '1' if deletion, or '2' if junction line
2. Number string representing RNAME (+ '+ or -' if junction; same as
field 6)
3. Start position (Last base before insertion, first base of deletion,
or first base of intron)
4. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
5. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
6. Coverage of feature for sample with index N
...
N + 6. Coverage of feature in sample with index N
--------------------------------------------------------------------
10. SUMMED number of instances of junction, insertion, or deletion in
sample
OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input indels/junctions
output_stream: where to write output
manifest_object: object of class LabelsAndIndices that maps indices
to labels and back; used to count number of samples.
sample_fraction: fraction of samples in which an indel must appear
to pass filter if coverage_threshold criterion is not satisfied
coverage_threshold: number of reads that must overlap indel in at
least one sample to pass filter of sample_fraction criterion is not
satisfied
verbose: output extra debugging statements
Return value: tuple (input line count, output line count)
"""
input_line_count, output_line_count = 0, 0
'''Compute minimum number of samples in which indel should appear to be
output if coverage threshold not met.'''
total_sample_count = len(manifest_object.label_to_index)
min_sample_count = int(round(
total_sample_count * sample_fraction
))
for (line_type, rname, pos, end_pos, strand_or_seq), xpartition in xstream(
input_stream, 5
):
collect_specs = [rname, pos, end_pos if line_type != 'N'
else str(int(end_pos) - 1),
strand_or_seq]
coverages = []
i = 0
if line_type == 'N':
for sample_index, data in itertools.groupby(
xpartition,
key=lambda val: val[0]
):
sample_index = int(sample_index)
while i != sample_index:
# Write 0 coverage for sample indexes reporting 0 junctions
coverages.append(0)
i += 1
coverage_sum = 0
max_left_displacement, max_right_displacement = None, None
maximin_displacement = None
for _, left_displacement, right_displacement, coverage in data:
input_line_count += 1
left_displacement = int(left_displacement)
right_displacement = int(right_displacement)
max_left_displacement = max(left_displacement,
max_left_displacement)
max_right_displacement = max(right_displacement,
max_right_displacement)
maximin_displacement = max(
min(left_displacement, right_displacement),
maximin_displacement
)
coverage_sum += int(coverage)
assert max_left_displacement is not None
assert max_right_displacement is not None
assert maximin_displacement is not None
print >>output_stream, \
'bed\tN\t%d\t%s\t%s\t%s\t%s\t%d\t%d\t%d\t%d' % (
sample_index, rname, pos, end_pos, strand_or_seq,
max_left_displacement, max_right_displacement,
maximin_displacement, coverage_sum
)
coverages.append(coverage_sum)
i += 1
output_line_count += 1
output_stream.write('collect\t2\t')
print >>output_stream, '\t'.join(
collect_specs
+ [str(coverage_value) for coverage_value in coverages]
+ ['0']*(total_sample_count - len(coverages))
)
output_line_count += 1
else:
assert line_type in 'ID'
sample_count = 0
for sample_index, data in itertools.groupby(
xpartition,
key=lambda val: val[0]
):
sample_index = int(sample_index)
while i != sample_index:
# Write 0 coverage for sample indexes reporting 0 indels
coverages.append(0)
i += 1
coverage_sum = 0
for _, _, _, coverage in data:
input_line_count += 1
coverage_sum += int(coverage)
print >>output_stream, \
'bed\t%s\t%s\t%s\t%s\t%s\t%s\t\x1c\t\x1c\t\x1c\t%d' % (
line_type, sample_index, rname, pos, end_pos,
strand_or_seq, coverage_sum
)
coverages.append(coverage_sum)
sample_count += 1
i += 1
output_line_count += 1
max_coverage = max(coverages)
if (sample_count >= min_sample_count
or (max_coverage >= coverage_threshold
and coverage_threshold != -1)):
if line_type == 'I':
output_stream.write('collect\t0\t')
else:
output_stream.write('collect\t1\t')
print >>output_stream, \
'\t'.join(
collect_specs
+ [str(coverage_value) for coverage_value in coverages]
+ ['0']*(total_sample_count - len(coverages))
)
output_line_count += 1
elif verbose:
print >>sys.stderr, (
'Indel (%s, %s, %s, %s) filtered out; it appeared in '
'%d sample(s), and its coverage in any one sample did '
'not exceed %d.'
) % (rname, strand_or_seq, pos, end_pos, sample_count,
max_coverage)
return input_line_count, output_line_count
print >> sizes_stream, '%s %d' % (rname,
reference_index.rname_lengths[rname])
input_line_count, output_line_count = 0, 0
output_url = Url(args.out)
if output_url.is_local:
# Set up destination directory
try:
os.makedirs(output_url.to_url())
except:
pass
mover = filemover.FileMover(args=args)
track_line = ('track type=bedGraph name="{name}" '
'description="{description}" visibility=full '
'color=227,29,118 altColor=0,179,220 priority=400')
for (sample_index, ), xpartition in xstream(sys.stdin, 1):
try:
sample_label = manifest_object.index_to_label[sample_index]
except KeyError:
# It's a mean or median
if 'mean' in sample_index or 'median' in sample_index:
sample_label = sample_index
else:
raise RuntimeError('Sample label index "%s" was not recorded.' %
sample_label)
'''Dictionary for which each key is a coverage (i.e., number of ECs
covering a given base). Its corresponding value is the number of bases with
that coverage.'''
coverage_histogram, unique_coverage_histogram = (defaultdict(int),
defaultdict(int))
with open(bed_filename, 'w') as bed_stream, \
def go(qname_stream, output_stream=sys.stdout, input_stream=sys.stdin,
verbose=False, report_multiplier=1.2):
""" Emits readlet alignments.
qname_stream contains long QNAMEs in the order in which readlets passed
to Bowtie appeared. These names would have been truncated. Each QNAME
takes the form
'\x1d'-separated list of [read sequence ID + ('-' if readlet
sequence is reverse-complemented; else '+') + '\x1e' + displacement
of readlet's 5' end from read's 5' end + '\x1e' + displacement of
readlet's 3' end from read's 3' end (+, for EXACTLY one readlet of
a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence)) + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective sample in list
A) + '\x1e' + (an '\x1f'-separated list of the number of instances
of the read sequence's reversed complement for each respective
sample in list B)]
A line is written per readlet per associated read sequence. So if a
given readlet can be found on 3 reads, 3 lines are written, each
containing the readlet's alignments.
qname_stream: where to retrieve extended qnames
input_stream: where to retrieve Bowtie output
output_stream: where to emit exon and intron tuples; typically, this is
sys.stdout.
verbose: True if alignments should occasionally be written to stderr.
report_multiplier: if verbose is True, the line number of an
alignment written to stderr increases exponentially with base
report_multiplier.
"""
output_line_count, next_report_line, i = 0, 0, 0
for (qname,), xpartition in xstream(input_stream, 1):
'''While labeled multireadlet, this list may end up simply a
unireadlet.'''
multireadlet = []
for tokens in xpartition:
(flag, rname, pos, mapq, cigar,
rnext, pnext, tlen, seq, qual) = tokens[:10]
flag = int(flag)
multireadlet.append((rname, flag, pos))
if verbose and next_report_line == i:
print >>sys.stderr, \
'SAM output record %d: rdname="%s", flag=%d' % (i,
qname,
flag)
next_report_line = int((next_report_line + 1)
* report_multiplier + 1) - 1
i += 1
'''If the next qname doesn't match the last qname or there are no
more lines, all of a multireadlet's alignments have been
collected.'''
if not flag & 4:
'''Last readlet has at least one alignment; print all
alignments for each read from which readlet sequence is
derived.'''
rnames, flags, poses = zip(*multireadlet)
reverse_flags = [a_flag ^ 16 for a_flag in flags]
flags = '\x1f'.join([str(a_flag) for a_flag in flags])
reverse_flags = '\x1f'.join(
[str(a_flag) for a_flag
in reverse_flags]
)
rnames = '\x1f'.join(rnames)
poses = '\x1f'.join(poses)
read = qname_stream.readline().strip()
while read != '+':
read_id, _, read_rest = read.partition('\x1e')
if read_id[-1] == '-':
current_flags = reverse_flags
else:
current_flags = flags
print >>output_stream, '%s\t%s\t%s\t%s\t%s' % \
(read_id[:-1], read_rest, rnames,
current_flags, poses)
output_line_count += 1
read = qname_stream.readline().strip()
else:
'''Readlet had no reported alignments; print ONLY when readlet
contains general info about read.'''
read = qname_stream.readline().strip()
while read != '+':
read_id, _, read_rest = read.partition('\x1e')
if len(read_rest.split('\x1e')) > 2:
print >>output_stream, \
'%s\t%s\t\x1c\t\x1c\t\x1c' % (read_id[:-1],
read_rest)
output_line_count += 1
read = qname_stream.readline().strip()
output_stream.flush()
print >>sys.stderr, ('align_readlets_delegate.py reports %d output lines.'
% output_line_count)
def go(input_stream=sys.stdin, output_stream=sys.stdout, fudge=5,
stranded=False, verbose=False, max_refs=300, report_multiplier=1.2):
""" Emits junction combinations associated with reads.
Soft-clipped Bowtie 2 alignments of read sequences to the transcript
fragment index are used infer which cojunctions could possibly be
overlapped by reads. Then maximal cliques of the graph described in
the maximal_cliques() function are enumerated to obtain which
junction combinations could possibly be overlapped by reads.
input_stream: where to retrieve Bowtie 2 output
output_stream: where to emit exon and junction tuples; typically, this
is sys.stdout.
fudge: by how many bases to extend left and right extend sizes
to accommodate potential indels
stranded: True iff input reads are strand-specific; this affects
whether an output partition has a terminal '+' or '-' indicating
the sense strand. Further, if stranded is True, an alignment is
returned only if its strand agrees with the junction's strand.
verbose: True if alignments should occasionally be written to stderr.
max_refs: maximum number of reference sequences to enumerate per read;
if more are present, prioritize those sequences that overlap
the fewest junctions
report_multiplier: if verbose is True, the line number of an
alignment written to stderr increases exponentially with base
report_multiplier.
"""
output_line_count, next_report_line, i = 0, 0, 0
for (qname,), xpartition in xstream(input_stream, 1):
'''While labeled multireadlet, this list may end up simply a
unireadlet.'''
multiread = []
for tokens in xpartition:
flag = int(tokens[0])
if verbose and next_report_line == i:
print >>sys.stderr, \
'SAM output record %d: rdname="%s", flag=%d' % (i,
qname,
flag)
next_report_line = int((next_report_line + 1)
* report_multiplier + 1) - 1
i += 1
multiread.append((qname,) + tokens)
if flag & 4: continue
corrected_multiread = multiread_with_junctions(multiread,
stranded)
cojunctions, all_junctions = defaultdict(set), {}
for alignment in multiread_with_junctions(multiread, stranded):
cigar = alignment[5]
md = [field for field in alignment
if field[:5] == 'MD:Z:'][0][5:]
pos = int(alignment[3])
seq = alignment[9]
reversed_complement_seq = seq[::-1].translate(
_reversed_complement_translation_table
)
if seq < reversed_complement_seq:
seq_to_print = seq
else:
seq_to_print = reversed_complement_seq
seq_size = len(seq)
rname = alignment[2]
sense = [field for field in alignment
if field[:5] == 'XS:A:'][0][5:]
if (rname, sense) not in all_junctions:
all_junctions[(rname, sense)] = defaultdict(list)
_, _, junctions, _, _ = indels_junctions_exons_mismatches(
cigar, md, pos, seq,
junctions_only=True
)
cojunctions[(rname, sense)].add(
tuple([(junction[0], junction[1])
for junction in junctions])
)
for junction in junctions:
if (junction[0], junction[1]) \
not in all_junctions[(rname, sense)]:
all_junctions[(rname, sense)][(junction[0], junction[1])] \
= [junction[2], junction[3]]
else:
all_junctions[(rname, sense)][
(junction[0], junction[1])
][0] = max(all_junctions[(rname, sense)][
(junction[0], junction[1])
][0], junction[2])
all_junctions[(rname, sense)][
(junction[0], junction[1])
][1] = max(all_junctions[(rname, sense)][
(junction[0], junction[1])
][1], junction[3])
for rname, sense in all_junctions:
to_write = set()
for cojunction in selected_cojunctions(paths_from_cojunctions(
list(cojunctions[(rname, sense)]), span=(seq_size + fudge)
), max_refs=max_refs, seq=seq, rname=rname, sense=sense):
left_extend_size = all_junctions[(rname, sense)][
cojunction[0]
][0]
right_extend_size = all_junctions[(rname, sense)][
cojunction[-1]
][1]
to_write.add(('{rname}{sense}\t{starts}'
'\t{ends}\t{left_size}'
'\t{right_size}\t{seq}').format(
rname=rname,
sense=sense,
starts=','.join(
[str(junction[0])
for junction in cojunction]
),
ends=','.join(
[str(junction[1])
for junction in cojunction]
),
left_size=(left_extend_size
+ fudge),
right_size=(right_extend_size
+ fudge),
seq=seq_to_print
))
for line_to_write in to_write:
print line_to_write
output_line_count += 1
output_stream.flush()
print >>sys.stderr, ('cojunction_enum_delegate.py reports %d output lines.'
% output_line_count)
site.addsitedir(base_path)
import bowtie
import bowtie_index
from dooplicity.tools import xstream
# Print file's docstring if -h is invoked
parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
bowtie.add_args(parser)
args = parser.parse_args()
input_line_count, output_line_count = 0, 0
start_time = time.time()
reference_index = bowtie_index.BowtieIndexReference(os.path.expandvars(args.bowtie_idx))
for (_, rname_string, intron_pos, intron_end_pos, sense, sample_index), xpartition in xstream(sys.stdin, 6):
coverage = 0
for value in xpartition:
input_line_count += 1
try:
# Assume intron line
_, _, instance_count = value
except ValueError:
# Alignment line
print(
"\t".join(
(value[2], value[3], reference_index.string_to_rname[rname_string], str(int(value[1]))) + value[4:]
)
+ ("\tXC:i:%d" % coverage)
)
output_line_count += 1
else:
output_path = os.path.join(temp_dir_path, output_filename)
input_line_count = 0
# Get RNAMEs in order of descending length
sorted_rnames = [reference_index.string_to_rname['%012d' % i]
for i in xrange(
len(reference_index.string_to_rname)
)]
sample_indexes_seen = set()
with xopen(True, output_path, 'w', args.gzip_level) as output_stream:
print >>output_stream, '\t'.join(
[''] + sorted_rnames
+ ['total mapped reads', 'total reads']
)
for (_, sample_index), xpartition in xstream(sys.stdin, 2):
counter.add('partitions')
sample_label = manifest_object.index_to_label[sample_index]
total_counts, unique_counts = defaultdict(int), defaultdict(int)
for rname_index, total_count, unique_count in xpartition:
counter.add('inputs')
rname = reference_index.string_to_rname[rname_index]
total_counts[rname] = int(total_count)
unique_counts[rname] = int(unique_count)
total_reads = sum(total_counts.values())
total_mapped_reads = total_reads - total_counts['*']
total_uniques = sum(unique_counts.values())
total_mapped_uniques = total_uniques - unique_counts['*']
counter.add('total_reads', total_reads)
counter.add('total_mapped_reads', total_mapped_reads)
counter.add('total_uniques', total_uniques)
def input_files_from_input_stream(input_stream,
output_stream,
temp_dir_path=None,
verbose=False,
gzip_level=3):
""" Generates FASTA reference to index and file with reads.
Each line of the read file is in the following format:
read number <TAB> SEQ <TAB> QUAL
input_stream: where to find Hadoop input
output_stream: where to write unmapped reads
temp_dir_path: where to store files
verbose: output extra debugging messages
gzip_level: gzip compression level (0-9)
Yield value: tuple (path to FASTA reference file, path to read file)
"""
global _input_line_count
if temp_dir_path is None: temp_dir_path = tempfile.mkdtemp()
prefasta_filename = os.path.join(temp_dir_path, 'temp.prefa')
deduped_fasta_filename = os.path.join(temp_dir_path, 'temp.deduped.prefa')
final_fasta_filename = os.path.join(temp_dir_path, 'temp.fa')
reads_filename = os.path.join(temp_dir_path, 'reads.temp.gz')
for (group_counter, ((index_group, ),
xpartition)) in enumerate(xstream(input_stream, 1)):
counter.add('partitions')
if verbose:
print >> sys.stderr, (
'Group %d: Writing prefasta and input reads...' %
group_counter)
with open(prefasta_filename, 'w') as fasta_stream:
with xopen(True, reads_filename, 'w') as read_stream:
for read_seq, values in itertools.groupby(
xpartition, key=lambda val: val[0]):
fasta_printed = False
counter.add('inputs')
for value in values:
_input_line_count += 1
if value[1][0] == '0':
# Print FASTA line
print >> fasta_stream, '\t'.join(
[value[1][1:-2], value[2]])
fasta_printed = True
elif fasta_printed:
'''Add to temporary seq stream only if an
associated FASTA line was found.'''
if value[1] == '1':
print >> read_stream, '\t'.join(
[value[2], read_seq, value[3]])
else:
print >> read_stream, '\t'.join([
value[2], read_seq[::-1].translate(
_reversed_complement_translation_table
), value[3][::-1]
])
else:
# Print unmapped read
if value[1] == '1':
seq_to_write = read_seq
qual_to_write = value[3]
else:
seq_to_write = read_seq[::-1].translate(
_reversed_complement_translation_table)
qual_to_write = value[3][::-1]
'''Write only essentials; handle "formal" writing
in next step.'''
output_stream.write(
'%s\t4\t\x1c\t\x1c\t\x1c\t\x1c'
'\t\x1c\t\x1c\t\x1c\t%s\t%s\n' %
(value[2], seq_to_write, qual_to_write))
if verbose:
print >> sys.stderr, (
'Group %d: Done! Sorting and deduplicating prefasta...' %
group_counter)
# Sort prefasta and eliminate duplicate lines
dedup_process_return = subprocess.call(
r'''sort %s | uniq >%s''' %
(prefasta_filename, deduped_fasta_filename),
shell=True,
executable='/bin/bash')
if dedup_process_return != 0:
raise RuntimeError(
'Problem encountered deduplicating FASTA reference')
if verbose:
print >> sys.stderr, ('Group %d Done! Writing final FASTA.' %
group_counter)
with open(final_fasta_filename, 'w') as final_fasta_stream:
with open(deduped_fasta_filename) as fasta_stream:
for line in fasta_stream:
rname, seq = line.strip().split('\t')
print >> final_fasta_stream, rname
final_fasta_stream.write('\n'.join(
[seq[i:i + 80] for i in xrange(0, len(seq), 80)]))
final_fasta_stream.write('\n')
os.remove(deduped_fasta_filename)
os.remove(prefasta_filename)
output_stream.flush()
yield final_fasta_filename, reads_filename
if output_url.is_local:
# Set up destination directory
try: os.makedirs(output_url.to_url())
except: pass
else:
mover = filemover.FileMover(args=args)
# Set up temporary destination
import tempfile
temp_dir_path = make_temp_dir(tempdel.silentexpandvars(args.scratch))
register_cleanup(tempdel.remove_temporary_directories, [temp_dir_path])
input_line_count = 0
counter = Counter('tsv')
register_cleanup(counter.flush)
for (line_type,), xpartition in xstream(sys.stdin, 1):
type_string = ('insertions' if line_type == '0' else
('deletions' if line_type == '1' else
('junctions' if line_type == '2' else
('coverages' + line_type[1:]
if line_type.startswith('3') else
'normalization'))))
counter.add(type_string + '_partitions')
output_filename = ((args.tsv_basename + '.'
if args.tsv_basename != '' else '')
+ type_string + '.tsv.gz')
if output_url.is_local:
output_path = os.path.join(args.out, output_filename)
else:
output_path = os.path.join(temp_dir_path, output_filename)
with xopen(True, output_path, 'w', args.gzip_level) as output_stream:
with open(sizes_filename, 'w') as sizes_stream:
for rname in reference_index.rname_lengths:
print >>sizes_stream, '%s %d' % (rname,
reference_index.rname_lengths[rname])
input_line_count, output_line_count = 0, 0
output_url = Url(args.out)
if output_url.is_local:
# Set up destination directory
try: os.makedirs(output_url.to_url())
except: pass
mover = filemover.FileMover(args=args)
track_line = ('track type=bedGraph name="{name}" '
'description="{description}" visibility=full '
'color=227,29,118 altColor=0,179,220 priority=400')
for (sample_index,), xpartition in xstream(sys.stdin, 1):
counter.add('partitions')
real_sample = True
try:
sample_label = manifest_object.index_to_label[sample_index]
except KeyError:
# It's a nonref track, a mean, or a median
real_sample = False
if search('\.[ATCGN]', sample_index):
try:
sample_label = (
manifest_object.index_to_label[sample_index[:-2]]
+ sample_index[-2:]
)
except KeyError:
raise RuntimeError('Sample label index "%s" was not recorded.'
def go(manifest_object, input_stream=sys.stdin, output_stream=sys.stdout,
sample_fraction=0.05, coverage_threshold=5, collect_junctions=False,
verbose=False):
""" Runs Rail-RNA-junction_filter.
Filters out every junction from input_stream that is not either:
(1) in round(sample_fraction * (total number of samples)) samples OR
(2) found in at least coverage_threshold reads in at least one
sample.
Input (read from stdin)
----------------------------
Tab-delimited columns:
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Intron start position (inclusive)
3. Intron end position (exclusive)
4. '\x1f'-separated list of sample indexes in which junction was found
5. '\x1f'-separated list of numbers of reads in which junction was
found in respective sample specified by field 4
Input is partitioned by fields 1-3.
Hadoop output (written to stdout)
----------------------------
Tab-delimited tuple columns (filter):
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Sample index
3. Intron start position (inclusive)
4. Intron end position (exclusive)
If --collect-junctions is True:
Tab-delimited tuple columns (collect):
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Intron start position (inclusive)
3. Intron end position (exclusive)
4. '\x1f'-separated list of sample indexes in which junction was found
5. '\x1f'-separated list of numbers of reads in which junction was
found in respective sample specified by field 4
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input junctions
output_stream: where to write output
manifest_object: object of class LabelsAndIndices that maps indices
to labels and back; used to count number of samples.
sample_fraction: fraction of samples in which a junction must appear
to pass filter if coverage_threshold criterion is not satisfied
coverage_threshold: number of reads that must overlap junction in at
least one sample to pass filter of sample_fraction criterion is not
satisfied
collect_junctions: collects and outputs junctions across samples;
ignores sample_fraction and coverage_threshold
verbose: output extra debugging statements
Return value: tuple (input line count, output line count)
"""
input_line_count, output_line_count = 0, 0
min_sample_count = int(round(
len(manifest_object.label_to_index) * sample_fraction
))
for (rname_and_strand, pos, end_pos), xpartition in xstream(
input_stream, 3
):
counter.add('partitions')
sample_indexes = defaultdict(int)
for current_sample_indexes, current_sample_counts in xpartition:
input_line_count += 1
counter.add('inputs')
current_sample_counts = current_sample_counts.split('\x1f')
for i, sample_index in enumerate(
current_sample_indexes.split('\x1f')
):
sample_indexes[sample_index] += int(current_sample_counts[i])
pos, end_pos = int(pos), int(end_pos)
if collect_junctions:
samples_to_dump = sorted(sample_indexes.items(),
key=lambda sample: int(sample[0]))
counter.add('collect_junction_lines')
print >>output_stream, 'collect\t%s\t%012d\t%012d\t%s\t%s' % (
rname_and_strand, pos, end_pos,
','.join([sample[0] for sample in samples_to_dump]),
','.join([str(sample[1]) for sample in samples_to_dump])
)
output_line_count += 1
sample_count = len(sample_indexes)
max_coverage = max(sample_indexes.values())
if end_pos > pos and (sample_count >= min_sample_count
or (max_coverage >= coverage_threshold
and coverage_threshold != -1)):
for sample_index in sample_indexes:
counter.add('junctions_passing_filter')
print >>output_stream, 'filter\t%s\t%s\t%012d\t%012d' % (
rname_and_strand, sample_index,
pos, end_pos
)
output_line_count += 1
else:
counter.add('junctions_failing_filter')
if verbose:
print >>sys.stderr, (
'Junction (%s, %d, %d) filtered out; it appeared in %d '
'sample(s), and its coverage in any one sample did '
'not exceed %d.'
) % (rname_and_strand, pos, end_pos,
sample_count, max_coverage)
return input_line_count, output_line_count
alignment_count_to_report, seed, non_deterministic \
= bowtie.parsed_bowtie_args(bowtie_args)
alignment_printer = AlignmentPrinter(manifest_object,
reference_index,
output_stream=sys.stdout,
bin_size=args.partition_length,
exon_ivals=args.exon_intervals,
exon_diffs=args.exon_differentials,
drop_deletions=args.drop_deletions,
output_bam_by_chr=args.output_bam_by_chr,
tie_margin=args.tie_margin)
input_line_count, output_line_count = 0, 0
start_time = time.time()
for (qname, ), xpartition in xstream(sys.stdin, 1):
alignments = [(qname, ) + alignment for alignment in xpartition]
input_line_count += len(alignments)
intron_counts = [alignment[5].count('N') for alignment in alignments]
min_intron_count = min(intron_counts)
if not min_intron_count:
'''There is at least one alignment that overlaps no introns; report
an alignment with the highest score at random. Separate into alignments
that overlap the fewest introns and alignments that don't.'''
clipped_alignments = [
alignments[i] for i in xrange(len(intron_counts))
if intron_counts[i] == 0
]
alignments_and_scores = [(alignment, [
int(tokens[5:]) for tokens in alignment if tokens[:5] == 'AS:i:'
][0]) for alignment in clipped_alignments]
def edges_from_input_stream(input_stream,
readlet_size=20,
min_overlap_exon_size=1):
""" Generates edges of directed acyclic graph (DAG) of introns.
A DAG is constructed for each strand. Each node of the DAG represents
a unique intron and is labeled by the tuple (intron_start, intron_end),
where intron_start is the (1-based) coordinate of the first base of the
intron, and intron_end is the coordinate of the first base after the
intron. An edge occurs between two introns A and B iff they do not
overlap, and no intron C occurs between A and B such that A, B, and C
do not overlap. The intron with larger coordinates is the child of
the intron with smaller coordinates. Weight each edge by the number of
exonic bases between the introns it connects.
The DAG has sources and sinks. Pad the DAG with new sources and sinks
as follows. Extend an edge to each original source
(source_start, source_end) from a new source labeled by the tuple
(None, max(1, source_start - readlet_size + 1)). Extend an edge
from each original sink (sink_start, sink_end) to a new sink labeled by
the tuple (sink_end + source_start + readlet_size - 1, None).
So each original source is assigned exactly one new source, and each
original sink is assigned exactly one new sink.
The paths through this DAG span all possible combinations of
nonoverlapping introns on the strand. Finding all subpaths (sequences
of introns), each of whose weights is <= readlet_size, redundantly
enumerates all possible combinations of introns a readlet
can overlap. Unfortunately, obtaining such combinations in
"hot spots," where there are many alternative splicings and short
exons, can become computationally intractable for large readlet sizes.
To (optionally) control these combinatorial blow-ups, impose an
effective minimum exon size min_overlap_exon_size by redefining overlap
between two introns: two introns overlap if there are fewer than
min_overlap_exon_size exonic bases between them.
The algorithm for generating the edges of the DAG operates on an input
stream whose lines are composed of the following tab-separated fields:
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Sample index
3. Intron start position (inclusive)
4. Intron end position (exclusive)
The input is partitioned by strand/sample index (fields 1-2) and sorted
by the remaining fields. INPUT COORDINATES ARE ASSUMED TO BE 1-INDEXED.
Introns are sorted by start position. To begin, the first set of
mutually overlapping introns are connected to their new sources. Two
data structures encode the graph as it is constructed [1]: the set
unlinked_nodes, containing introns that do not yet have any children,
and linked_nodes, a dictionary that, where possible, maps each intron A
to its corresponding successive nonoverlapping intron B with the
smallest end position read so far. With each new intron N read, the
nodes in unlinked_nodes and linked_nodes are checked for whether N is
their child, and these edges are yielded. (Note that by construction,
nodes are streamed in topological order.) Nodes from unlinked_nodes may
thus be promoted to linked_nodes. Each value node in linked_nodes is
also replaced with N if N has a smaller end position. Then every node A
in linked_nodes is checked for a path A -> B -> N. If such a path
exists, an edge can NEVER connect A with any successive introns, and A
is removed from linked_nodes. The algorithm continues until the end of
the strand is reached, when the edges connecting remaining
unlinked_nodes and their new sinks are yielded.
[1] Technically, there are three data structures. unlinked_nodes and
linked_nodes contain only indices of introns, and "introns" is
a dictionary that maps indices to tuples (intron_start, intron_end).
input_stream: where to find sorted introns of the form specified above.
fudge: by how much a readlet_size should be extended.
min_exon_size: minimum number of exonic bases between two introns
for them to be considered nonoverlapping.
Yield value: An edge tuple (strand,
sample_index,
(intron A start, intron A end),
(intron B start, intron B end)) or None
at the beginning of a new partition.
"""
global _input_line_count
for key, xpartition in xstream(input_stream, 2, skip_duplicates=True):
unlinked_nodes = set()
for q, value in enumerate(xpartition):
assert len(value) == 2
_input_line_count += 1
if not q:
# Denote start of new partition
yield None
# Handle first intron from partition
intron_start, intron_end = int(value[0]), int(value[1])
# Create fake source before first intron
fake_source = (None, max(intron_start - (readlet_size - 1), 1))
introns = {0: fake_source, 1: (intron_start, intron_end)}
linked_nodes = {0: 1}
unlinked_nodes = set([1])
index = 2
# Yield first edge for strand (before first intron)
yield key + (fake_source, (intron_start, intron_end))
first_intron = False
else:
# Handle next introns from partition
intron_start, intron_end = int(value[0]), int(value[1])
introns[index] = (intron_start, intron_end)
nodes_to_trash = []
for node in unlinked_nodes:
if intron_start >= introns[node][1] + \
min_overlap_exon_size:
nodes_to_trash.append(node)
for node in nodes_to_trash:
linked_nodes[node] = index
unlinked_nodes.remove(node)
unlinked_nodes.add(index)
nodes_to_trash = []
for node in linked_nodes:
intermediate_node = linked_nodes[node]
if intermediate_node in linked_nodes:
nodes_to_trash.append(node)
else:
yield key + (introns[node], (intron_start, intron_end))
if introns[intermediate_node][1] > intron_end:
linked_nodes[node] = index
for node in nodes_to_trash:
del linked_nodes[node]
del introns[node]
index += 1
# Yield final edges for strand
for node in unlinked_nodes:
current_intron = introns[node]
yield key + (current_intron,
(current_intron[1] + readlet_size - 1, None))
def input_files_from_input_stream(input_stream,
output_stream,
temp_dir_path=None,
verbose=False,
gzip_level=3):
""" Generates FASTA reference to index and file with reads.
Each line of the read file is in the following format:
read number <TAB> SEQ <TAB> QUAL
input_stream: where to find Hadoop input
output_stream: where to write unmapped reads
temp_dir_path: where to store files
verbose: output extra debugging messages
gzip_level: gzip compression level (0-9)
Yield value: tuple (path to FASTA reference file, path to read file)
"""
global _input_line_count
if temp_dir_path is None: temp_dir_path = tempfile.mkdtemp()
prefasta_filename = os.path.join(temp_dir_path, 'temp.prefa')
deduped_fasta_filename = os.path.join(temp_dir_path, 'temp.deduped.prefa')
final_fasta_filename = os.path.join(temp_dir_path, 'temp.fa')
reads_filename = os.path.join(temp_dir_path, 'reads.temp.gz')
for (counter, ((index_group,), xpartition)) in enumerate(
xstream(input_stream, 1)
):
if verbose:
print >>sys.stderr, (
'Group %d: Writing prefasta and input reads...'
% counter
)
with open(prefasta_filename, 'w') as fasta_stream:
with xopen(True, reads_filename, 'w') as read_stream:
for read_seq, values in itertools.groupby(xpartition,
key=lambda val: val[0]):
fasta_printed = False
for value in values:
_input_line_count += 1
if value[1][0] == '0':
# Print FASTA line
print >>fasta_stream, '\t'.join([value[1][1:-2],
value[2]])
fasta_printed = True
elif fasta_printed:
'''Add to temporary seq stream only if an
associated FASTA line was found.'''
if value[1] == '1':
print >>read_stream, '\t'.join([value[2],
read_seq,
value[3]])
else:
print >>read_stream, '\t'.join([
value[2],
read_seq[::-1].translate(
_reversed_complement_translation_table
),
value[3][::-1]])
else:
# Print unmapped read
if value[1] == '1':
seq_to_write = read_seq
qual_to_write = value[3]
else:
seq_to_write = read_seq[::-1].translate(
_reversed_complement_translation_table
)
qual_to_write = value[3][::-1]
'''Write only essentials; handle "formal" writing
in next step.'''
output_stream.write(
'%s\t4\t\x1c\t\x1c\t\x1c\t\x1c'
'\t\x1c\t\x1c\t\x1c\t%s\t%s\n' % (
value[2],
seq_to_write,
qual_to_write
)
)
if verbose:
print >>sys.stderr, (
'Group %d: Done! Sorting and deduplicating prefasta...'
% counter
)
# Sort prefasta and eliminate duplicate lines
dedup_process_return = subprocess.call(
r'''sort %s | uniq >%s'''
% (prefasta_filename, deduped_fasta_filename), shell=True,
executable='/bin/bash'
)
if dedup_process_return != 0:
raise RuntimeError(
'Problem encountered deduplicating FASTA reference'
)
if verbose:
print >>sys.stderr, (
'Group %d Done! Writing final FASTA.' % counter
)
with open(final_fasta_filename, 'w') as final_fasta_stream:
with open(deduped_fasta_filename) as fasta_stream:
for line in fasta_stream:
rname, seq = line.strip().split('\t')
print >>final_fasta_stream, rname
final_fasta_stream.write(
'\n'.join([seq[i:i+80] for i
in xrange(0, len(seq), 80)])
)
final_fasta_stream.write('\n')
os.remove(deduped_fasta_filename)
os.remove(prefasta_filename)
yield final_fasta_filename, reads_filename
alignment_printer = AlignmentPrinter(
manifest_object,
reference_index,
output_stream=sys.stdout,
bin_size=args.partition_length,
exon_ivals=args.exon_intervals,
exon_diffs=args.exon_differentials,
drop_deletions=args.drop_deletions,
output_bam_by_chr=args.output_bam_by_chr,
tie_margin=args.tie_margin
)
input_line_count, output_line_count = 0, 0
start_time = time.time()
for (qname,), xpartition in xstream(sys.stdin, 1):
alignments = [(qname,) + alignment for alignment in xpartition]
input_line_count += len(alignments)
junction_counts = [alignment[5].count('N') for alignment in alignments]
min_junction_count = min(junction_counts)
if not min_junction_count:
'''There is at least one alignment that overlaps no junctions; report
an alignment with the highest score at random. Separate into alignments
that overlap the fewest junctions and alignments that don't.'''
clipped_alignments = [alignments[i] for i in xrange(
len(junction_counts)
) if junction_counts[i] == 0]
alignments_and_scores = [(alignment, [int(tokens[5:])
for tokens in alignment
if tokens[:5] == 'AS:i:'][0])
for alignment in clipped_alignments]
# To convert sample-rname index to sample index-rname index tuple
sample_and_rname_indexes = SampleAndRnameIndexes(
manifest_object,
args.output_by_chromosome
)
import time
start_time = time.time()
from alignment_handlers import AlignmentPrinter
alignment_printer = AlignmentPrinter(manifest_object, reference_index,
tie_margin=args.tie_margin)
input_line_count = 0
if args.suppress_bam:
# Just grab stats
if args.output_by_chromosome:
for (index, _), xpartition in xstream(sys.stdin, 2):
sample_index, rname_index = (
sample_and_rname_indexes.sample_and_rname_indexes(index)
)
unique_count, total_count = 0, 0
for record in xpartition:
if not (int(record[2]) & 256):
total_count += 1
try:
# seq is at position 8
if alignment_printer.unique(record, seq_index=8):
unique_count += 1
except IndexError:
# Unmapped read; it's unique
unique_count += 1
input_line_count += 1
def go(true_bed_stream,
sam_stream=sys.stdin,
generous=False,
base_threshold=0.5,
clip_threshold=1.0,
dump_incorrect=False,
temp_dir=None,
ignore_spliced_reads=False):
""" Finds relevant and retrieved instance counts.
true_bed_stream: file handle for BED output of Flux simulation
sam_stream: where to read in aligner's mappings
generous: True iff aligner cuts off /1 or /2 of a given read
base_threshold: proportion of a read's bases that must align
correctly for a read to be considered a correct mapping
clip_threshold: proportion of a read's bases that must be clipped
for a read to be considered unmapped
dump_incorrect: write incorrect (read) alignments to stderr
ignore_spliced_reads: ignores all spliced reads
"""
from tempdel import remove_temporary_directories
import tempfile
import atexit
if temp_dir is None:
temp_dir_path = tempfile.mkdtemp()
else:
try:
temp_dir_path = tempfile.mkdtemp(dir=temp_dir)
except:
temp_dir_path = tempfile.mkdtemp()
#print >>sys.stderr, temp_dir_path
atexit.register(remove_temporary_directories, [temp_dir_path])
# Store everything in one file, then sort it on read name
combined_file = os.path.join(temp_dir_path, 'combined.temp')
with open(combined_file, 'w') as temp_stream:
if ignore_spliced_reads:
if generous:
for line in true_bed_stream:
tokens = line.strip().split('\t')
if ',' in tokens[-1]: continue # skip intron line
print >> temp_stream, '\t'.join([tokens[3][:-2], '0'] +
tokens[:3] + tokens[4:])
else:
for line in true_bed_stream:
tokens = line.strip().split('\t')
if ',' in tokens[-1]: continue # skip intron line
print >> temp_stream, '\t'.join([tokens[3], '0'] +
tokens[:3] + tokens[4:])
for line in sam_stream:
if line[0] == '@' or not line.strip(): continue
tokens = line.strip().split('\t')
if 'N' in tokens[5]: continue # skip intron line
print >> temp_stream, '\t'.join([tokens[0], '1'] + tokens[1:])
else:
if generous:
for line in true_bed_stream:
tokens = line.strip().split('\t')
print >> temp_stream, '\t'.join([tokens[3][:-2], '0'] +
tokens[:3] + tokens[4:])
else:
for line in true_bed_stream:
tokens = line.strip().split('\t')
print >> temp_stream, '\t'.join([tokens[3], '0'] +
tokens[:3] + tokens[4:])
for line in sam_stream:
if line[0] == '@' or not line.strip(): continue
tokens = line.strip().split('\t')
print >> temp_stream, '\t'.join([tokens[0], '1'] + tokens[1:])
import subprocess
sorted_combined_file = os.path.join(temp_dir_path, 'combined.sorted.temp')
subprocess.check_call(' '.join([
'sort -T %s -k1,1 -k2,2n' % temp_dir_path, combined_file, '>',
sorted_combined_file
]),
bufsize=-1,
shell=True)
basewise_relevant, read_relevant = 0, 0
# Initialize counters for computing accuracy metrics
basewise_retrieved, basewise_intersection = 0, 0
read_retrieved, read_intersection = 0, 0
with open(sorted_combined_file) as sorted_combined_stream:
for (name, ), xpartition in xstream(sorted_combined_stream, 1):
'''Dict mapping read names to alignments
(chrom, 1-based start, 1-based end)'''
true_maps = []
saved = []
for tokens in xpartition:
saved.append(tokens)
if tokens[0] == '0':
if len(tokens) < 12:
continue
chrom = tokens[1]
chrom_start = int(tokens[2])
chrom_end = int(tokens[3])
block_sizes = tokens[10].split(',')
block_starts = tokens[11].split(',')
# Handle trailing commas
try:
int(block_sizes[-1])
except ValueError:
block_sizes = block_sizes[:-1]
try:
int(block_starts[-1])
except ValueError:
block_starts = block_starts[:-1]
block_count = len(block_sizes)
assert block_count == len(block_starts)
exons = [(chrom, chrom_start + int(block_starts[i]),
chrom_start + int(block_starts[i]) +
int(block_sizes[i]))
for i in xrange(block_count)]
true_maps.append(exons)
basewise_relevant += sum(
[int(block_size) for block_size in block_sizes])
read_relevant += 1
elif tokens[0] == '1':
flag = int(tokens[1])
if flag & 256 or flag & 4:
'''Secondary alignment or unmapped and thus not
retrieved; ignore'''
continue
cigar, pos, seq = tokens[5], int(tokens[3]), tokens[9]
(dummy_md, mapped,
unmapped, clip_count, read_length) \
= dummy_md_and_mapped_offsets(
cigar,
clip_threshold=clip_threshold
)
if unmapped:
# Too much clipping
continue
basewise_retrieved += read_length - clip_count
read_retrieved += 1
if not true_maps:
assert ignore_spliced_reads
continue
# Try both /1 and /2; choose the best basewise result
intersected_base_count = 0
for true_map in true_maps:
if tokens[2] != true_map[0][0]:
'''chr is wrong, but this is still counted as a
retrieval above'''
continue
base_counter, base_truths = 0, set()
'''Each tuple in base_truths is
(index of base in read, mapped location)'''
for block in true_map:
base_truths.update([(base_counter + i, j + 1)
for i, j in enumerate(
xrange(block[1], block[2]))
])
base_counter += block[2] - block[1]
base_predictions = set()
if unmapped:
# Too much clipping
continue
_, _, _, exons, _ = indels_junctions_exons_mismatches(
cigar, dummy_md, pos, seq, drop_deletions=True)
mapped_index = 0
for exon in exons:
base_predictions.update([
(mapped[mapped_index + i], j)
for i, j in enumerate(xrange(exon[0], exon[1]))
])
mapped_index += exon[1] - exon[0]
intersected_base_count = max(
intersected_base_count,
len(base_predictions.intersection(base_truths)))
basewise_intersection += intersected_base_count
if intersected_base_count >= read_length * base_threshold:
read_intersection += 1
elif dump_incorrect:
# Incorrect alignment; write to stderr
print >> sys.stderr, '\t'.join(
['.'.join(line) for line in saved])
else:
raise RuntimeError('Invalid intermediate line.')
return (basewise_retrieved, basewise_relevant, basewise_intersection,
read_retrieved, read_relevant, read_intersection)
else Url(os.getcwd())
input_line_count = 0
counter = Counter('bed')
register_cleanup(counter.flush)
if output_url.is_local:
# Set up destination directory
try: os.makedirs(output_url.to_url())
except: pass
else:
mover = filemover.FileMover(args=args)
# Set up temporary destination
import tempfile
temp_dir_path = make_temp_dir(tempdel.silentexpandvars(args.scratch))
register_cleanup(tempdel.remove_temporary_directories, [temp_dir_path])
for (line_type, sample_label), xpartition in xstream(sys.stdin, 2):
assert line_type in 'NID'
sample_label = manifest_object.index_to_label[sample_label]
type_string = ('insertions' if line_type == 'I' else
('deletions' if line_type == 'D' else 'junctions'))
output_filename = ((args.bed_basename + '.'
if args.bed_basename != '' else '')
+ type_string + '.' + sample_label + '.bed')
if output_url.is_local:
output_path = os.path.join(args.out, output_filename)
else:
output_path = os.path.join(temp_dir_path, output_filename)
with open(output_path, 'w') as output_stream:
print >>output_stream, 'track name="%s_%s" description="' \
'Rail-RNA v%s %s for sample %s"' \
% (sample_label,
if output_url.is_local:
output_path = os.path.join(args.out, output_filename)
else:
output_path = os.path.join(temp_dir_path, output_filename)
input_line_count = 0
# Get RNAMEs in order of descending length
sorted_rnames = [
reference_index.string_to_rname['%012d' % i]
for i in xrange(len(reference_index.string_to_rname))
]
sample_indexes_seen = set()
with xopen(True, output_path, 'w', args.gzip_level) as output_stream:
print >> output_stream, '\t'.join([''] + sorted_rnames +
['total mapped reads', 'total reads'])
for (_, sample_index), xpartition in xstream(sys.stdin, 2):
sample_label = manifest_object.index_to_label[sample_index]
total_counts, unique_counts = defaultdict(int), defaultdict(int)
for rname_index, total_count, unique_count in xpartition:
rname = reference_index.string_to_rname[rname_index]
total_counts[rname] = int(total_count)
unique_counts[rname] = int(unique_count)
total_reads = sum(total_counts.values())
total_mapped_reads = total_reads - total_counts['*']
total_uniques = sum(unique_counts.values())
total_mapped_uniques = total_uniques - unique_counts['*']
print >> output_stream, '\t'.join([sample_label] + [
'%d,%d' % (total_counts[rname], unique_counts[rname])
for rname in sorted_rnames
] + [
'%d,%d' % (total_mapped_reads, total_mapped_uniques),
def go(true_bed_stream, sam_stream=sys.stdin, generous=False,
base_threshold=0.5, clip_threshold=1.0, dump_incorrect=False,
temp_dir=None, ignore_spliced_reads=False):
""" Finds relevant and retrieved instance counts.
true_bed_stream: file handle for BED output of Flux simulation
sam_stream: where to read in aligner's mappings
generous: True iff aligner cuts off /1 or /2 of a given read
base_threshold: proportion of a read's bases that must align
correctly for a read to be considered a correct mapping
clip_threshold: proportion of a read's bases that must be clipped
for a read to be considered unmapped
dump_incorrect: write incorrect (read) alignments to stderr
ignore_spliced_reads: ignores all spliced reads
"""
from tempdel import remove_temporary_directories
import tempfile
import atexit
if temp_dir is None:
temp_dir_path = tempfile.mkdtemp()
else:
try:
temp_dir_path = tempfile.mkdtemp(dir=temp_dir)
except:
temp_dir_path = tempfile.mkdtemp()
#print >>sys.stderr, temp_dir_path
atexit.register(remove_temporary_directories, [temp_dir_path])
# Store everything in one file, then sort it on read name
combined_file = os.path.join(temp_dir_path, 'combined.temp')
with open(combined_file, 'w') as temp_stream:
if ignore_spliced_reads:
if generous:
for line in true_bed_stream:
tokens = line.strip().split('\t')
if ',' in tokens[-1]: continue # skip intron line
print >>temp_stream, '\t'.join([tokens[3][:-2], '0']
+ tokens[:3] + tokens[4:])
else:
for line in true_bed_stream:
tokens = line.strip().split('\t')
if ',' in tokens[-1]: continue # skip intron line
print >>temp_stream, '\t'.join(
[tokens[3], '0'] + tokens[:3] + tokens[4:]
)
for line in sam_stream:
if line[0] == '@' or not line.strip(): continue
tokens = line.strip().split('\t')
if 'N' in tokens[5]: continue # skip intron line
print >>temp_stream, '\t'.join([tokens[0], '1'] + tokens[1:])
else:
if generous:
for line in true_bed_stream:
tokens = line.strip().split('\t')
print >>temp_stream, '\t'.join([tokens[3][:-2], '0']
+ tokens[:3] + tokens[4:])
else:
for line in true_bed_stream:
tokens = line.strip().split('\t')
print >>temp_stream, '\t'.join(
[tokens[3], '0'] + tokens[:3] + tokens[4:]
)
for line in sam_stream:
if line[0] == '@' or not line.strip(): continue
tokens = line.strip().split('\t')
print >>temp_stream, '\t'.join([tokens[0], '1'] + tokens[1:])
import subprocess
sorted_combined_file = os.path.join(temp_dir_path, 'combined.sorted.temp')
subprocess.check_call(' '.join(['sort -T %s -k1,1 -k2,2n'
% temp_dir_path, combined_file,
'>', sorted_combined_file]),
bufsize=-1, shell=True)
basewise_relevant, read_relevant = 0, 0
# Initialize counters for computing accuracy metrics
basewise_retrieved, basewise_intersection = 0, 0
read_retrieved, read_intersection = 0, 0
with open(sorted_combined_file) as sorted_combined_stream:
for (name,), xpartition in xstream(sorted_combined_stream, 1):
'''Dict mapping read names to alignments
(chrom, 1-based start, 1-based end)'''
true_maps = []
saved = []
for tokens in xpartition:
saved.append(tokens)
if tokens[0] == '0':
if len(tokens) < 12:
continue
chrom = tokens[1]
chrom_start = int(tokens[2])
chrom_end = int(tokens[3])
block_sizes = tokens[10].split(',')
block_starts = tokens[11].split(',')
# Handle trailing commas
try:
int(block_sizes[-1])
except ValueError:
block_sizes = block_sizes[:-1]
try:
int(block_starts[-1])
except ValueError:
block_starts = block_starts[:-1]
block_count = len(block_sizes)
assert block_count == len(block_starts)
exons = [(chrom,
chrom_start + int(block_starts[i]),
chrom_start + int(block_starts[i])
+ int(block_sizes[i]))
for i in xrange(block_count)]
true_maps.append(exons)
basewise_relevant += sum([int(block_size) for block_size
in block_sizes])
read_relevant += 1
elif tokens[0] == '1':
flag = int(tokens[1])
if flag & 256 or flag & 4:
'''Secondary alignment or unmapped and thus not
retrieved; ignore'''
continue
cigar, pos, seq = tokens[5], int(tokens[3]), tokens[9]
(dummy_md, mapped,
unmapped, clip_count, read_length) \
= dummy_md_and_mapped_offsets(
cigar,
clip_threshold=clip_threshold
)
if unmapped:
# Too much clipping
continue
basewise_retrieved += read_length - clip_count
read_retrieved += 1
if not true_maps:
assert ignore_spliced_reads
continue
# Try both /1 and /2; choose the best basewise result
intersected_base_count = 0
for true_map in true_maps:
if tokens[2] != true_map[0][0]:
'''chr is wrong, but this is still counted as a
retrieval above'''
continue
base_counter, base_truths = 0, set()
'''Each tuple in base_truths is
(index of base in read, mapped location)'''
for block in true_map:
base_truths.update([(base_counter + i, j + 1)
for i, j in enumerate(
xrange(
block[1], block[2]
))])
base_counter += block[2] - block[1]
base_predictions = set()
if unmapped:
# Too much clipping
continue
_, _, _, exons = indels_introns_and_exons(
cigar,
dummy_md, pos, seq,
drop_deletions=True
)
mapped_index = 0
for exon in exons:
base_predictions.update(
[(mapped[mapped_index + i], j)
for i, j in enumerate(
xrange(
exon[0], exon[1]
))])
mapped_index += exon[1] - exon[0]
intersected_base_count = max(intersected_base_count,
len(
base_predictions.intersection(base_truths)
))
basewise_intersection += intersected_base_count
if intersected_base_count >= read_length * base_threshold:
read_intersection += 1
elif dump_incorrect:
# Incorrect alignment; write to stderr
print >>sys.stderr, '\t'.join(
['.'.join(line) for line in saved]
)
else:
raise RuntimeError(
'Invalid intermediate line.'
)
return (basewise_retrieved, basewise_relevant, basewise_intersection,
read_retrieved, read_relevant, read_intersection)
def go(manifest_object,
input_stream=sys.stdin,
output_stream=sys.stdout,
sample_fraction=0.05,
coverage_threshold=5,
verbose=False):
""" Runs Rail-RNA-bed_pre
Writes indels and junctions for outputting BEDs by sample and
TSVs across samples.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns:
1. 'I', 'D', or 'N' for insertion, deletion, or junction line
2. Number string representing RNAME
3. Start position (Last base before insertion, first base of deletion,
or first base of intron)
4. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
5. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
6. Sample index
----Next fields are for junctions only; they are '\x1c' for indels----
7. Number of nucleotides between 5' end of intron and 5' end of read
from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND. That is, if the sense strand is the reverse strand,
this is the distance between the 3' end of the read and the 3' end
of the intron.
8. Number of nucleotides between 3' end of intron and 3' end of read
from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND.
--------------------------------------------------------------------
9. Number of instances of junction, insertion, or deletion in sample;
this is always +1 before bed_pre combiner/reducer
Input is partitioned by fields 1-5 and sorted by field 6.
Hadoop output (written to stdout)
----------------------------
Tab-delimited output tuple columns (bed):
1. 'I', 'D', or 'N' for insertion, deletion, or junction line
2. Sample index
3. Number string representing RNAME (+ '+ or -' if junction; same as
field 6)
4. Start position (Last base before insertion, first base of deletion,
or first base of intron)
5. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
6. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
----Next fields are for junctions only; they are '\x1c' for indels----
7. MAX number of nucleotides between 5' end of intron and 5' end of
read from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND. That is, if the sense strand is the reverse strand,
this is the distance between the 3' end of the read and the 3' end
of the intron.
8. MAX number of nucleotides between 3' end of intron and 3' end of
read from which it was inferred, ASSUMING THE SENSE STRAND IS THE
FORWARD STRAND.
9. MAXIMIN (number of nucleotides between 5' end of intron and 5' end
of read, number of nucleotides between 3' end of intron and
3' end of read);
min is between the args above; max is across reads.
Tab-delimited output tuple columns (collect)
1. '0' if insertion, '1' if deletion, or '2' if junction line
2. Number string representing RNAME (+ '+ or -' if junction; same as
field 6)
3. Start position (Last base before insertion, first base of deletion,
or first base of intron)
4. End position (Last base before insertion, last base of deletion
(exclusive), or last base of intron (exclusive))
5. '+' or '-' indicating which strand is the sense strand for
junctions, inserted sequence for insertions, or deleted sequence
for deletions
6. Coverage of feature for sample with index N
...
N + 6. Coverage of feature in sample with index N
--------------------------------------------------------------------
10. SUMMED number of instances of junction, insertion, or deletion in
sample
OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input indels/junctions
output_stream: where to write output
manifest_object: object of class LabelsAndIndices that maps indices
to labels and back; used to count number of samples.
sample_fraction: fraction of samples in which an indel must appear
to pass filter if coverage_threshold criterion is not satisfied
coverage_threshold: number of reads that must overlap indel in at
least one sample to pass filter of sample_fraction criterion is not
satisfied
verbose: output extra debugging statements
Return value: tuple (input line count, output line count)
"""
input_line_count, output_line_count = 0, 0
'''Compute minimum number of samples in which indel should appear to be
output if coverage threshold not met.'''
total_sample_count = len(manifest_object.label_to_index)
min_sample_count = int(round(total_sample_count * sample_fraction))
for (line_type, rname, pos, end_pos,
strand_or_seq), xpartition in xstream(input_stream, 5):
collect_specs = [
rname, pos, end_pos if line_type != 'N' else str(int(end_pos) - 1),
strand_or_seq
]
sample_indexes, coverages = [], []
if line_type == 'N':
counter.add('junction_line')
for sample_index, data in itertools.groupby(
xpartition, key=lambda val: val[0]):
coverage_sum = 0
max_left_displacement, max_right_displacement = None, None
maximin_displacement = None
for _, left_displacement, right_displacement, coverage in data:
input_line_count += 1
left_displacement = int(left_displacement)
right_displacement = int(right_displacement)
max_left_displacement = max(left_displacement,
max_left_displacement)
max_right_displacement = max(right_displacement,
max_right_displacement)
maximin_displacement = max(
min(left_displacement, right_displacement),
maximin_displacement)
coverage_sum += int(coverage)
assert max_left_displacement is not None
assert max_right_displacement is not None
assert maximin_displacement is not None
counter.add('bed_line')
print >>output_stream, \
'bed\tN\t%s\t%s\t%s\t%s\t%s\t%d\t%d\t%d\t%d' % (
sample_index, rname, pos, end_pos, strand_or_seq,
max_left_displacement, max_right_displacement,
maximin_displacement, coverage_sum
)
sample_indexes.append(sample_index)
coverages.append(coverage_sum)
output_line_count += 1
counter.add('collect_line')
output_stream.write('collect\t2\t')
print >> output_stream, '\t'.join(
collect_specs +
[','.join(sample_indexes), ','.join(map(str, coverages))])
output_line_count += 1
else:
counter.add('insertion_line' if line_type ==
'I' else 'deletion_line')
assert line_type in 'ID'
sample_count = 0
for sample_index, data in itertools.groupby(
xpartition, key=lambda val: val[0]):
coverage_sum = 0
for _, _, _, coverage in data:
input_line_count += 1
coverage_sum += int(coverage)
counter.add('bed_line')
print >>output_stream, \
'bed\t%s\t%s\t%s\t%s\t%s\t%s\t\x1c\t\x1c\t\x1c\t%d' % (
line_type, sample_index, rname, pos, end_pos,
strand_or_seq, coverage_sum
)
sample_indexes.append(sample_index)
coverages.append(coverage_sum)
sample_count += 1
output_line_count += 1
max_coverage = max(coverages)
if (sample_count >= min_sample_count
or (max_coverage >= coverage_threshold
and coverage_threshold != -1)):
counter.add('collect_line')
if line_type == 'I':
output_stream.write('collect\t0\t')
else:
output_stream.write('collect\t1\t')
print >>output_stream, \
'\t'.join(
collect_specs
+ [','.join(sample_indexes),
','.join(map(str, coverages))]
)
output_line_count += 1
else:
counter.add('indel_filtered_out')
if verbose:
print >> sys.stderr, (
'Indel (%s, %s, %s, %s) filtered out; it appeared in '
'%d sample(s), and its coverage in any one sample did '
'not exceed %d.') % (rname, strand_or_seq, pos,
end_pos, sample_count,
max_coverage)
counter.flush()
return input_line_count, output_line_count
from dooplicity.tools import xstream
# Print file's docstring if -h is invoked
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
bowtie.add_args(parser)
args = parser.parse_args()
input_line_count, output_line_count = 0, 0
start_time = time.time()
reference_index = bowtie_index.BowtieIndexReference(
os.path.expandvars(args.bowtie_idx)
)
for (_, rname_string, intron_pos, intron_end_pos,
sense, sample_index), xpartition in xstream(sys.stdin, 6):
coverage = 0
for value in xpartition:
input_line_count += 1
try:
# Assume junction line
_, _, instance_count = value
except ValueError:
# Alignment line
print ('\t'.join((value[2], value[3],
reference_index.string_to_rname[rname_string],
str(int(value[1])))
+ value[4:]) + ('\tXC:i:%d' % coverage))
output_line_count += 1
else:
coverage += int(instance_count)
def go(input_stream=sys.stdin,
output_stream=sys.stdout,
bowtie2_exe='bowtie2',
bowtie_index_base='genome',
bowtie2_index_base='genome2',
manifest_file='manifest',
bowtie2_args=None,
bin_size=10000,
verbose=False,
exon_differentials=True,
exon_intervals=False,
report_multiplier=1.2,
min_exon_size=8,
search_filter=1,
min_readlet_size=15,
max_readlet_size=25,
readlet_interval=12,
capping_multiplier=1.5,
drop_deletions=False,
gzip_level=3,
scratch=None,
index_count=1,
output_bam_by_chr=False,
tie_margin=0,
no_realign=False,
no_polyA=False):
""" Runs Rail-RNA-align_reads.
A single pass of Bowtie is run to find end-to-end alignments. Unmapped
reads are saved for readletizing to determine introns in sucessive
reduce steps as well as for realignment in a later map step.
Input (read from stdin)
----------------------------
Tab-delimited input tuple columns in a mix of any of the following
three formats:
Format 1 (single-end, 3-column):
1. Nucleotide sequence or its reversed complement, whichever is first
in alphabetical order
2. 1 if sequence was reverse-complemented else 0
3. Name
4. Quality sequence or its reverse, whichever corresponds to field 1
Format 2 (paired, 2 lines, 3 columns each)
(so this is the same as single-end)
1. Nucleotide sequence for mate 1 or its reversed complement,
whichever is first in alphabetical order
2. 1 if sequence was reverse-complemented else 0
3. Name for mate 1
4. Quality sequence for mate 1 or its reverse, whichever corresponds
to field 1
(new line)
1. Nucleotide sequence for mate 2 or its reversed complement,
whichever is first in alphabetical order
2. 1 if sequence was reverse complemented else 0
3. Name for mate 2
4. Quality sequence for mate 2 or its reverse, whichever corresponds
to field 1
Input is partitioned and sorted by field 1, the read sequence.
Hadoop output (written to stdout)
----------------------------
A given RNAME sequence is partitioned into intervals ("bins") of some
user-specified length (see partition.py).
Exonic chunks (aka ECs; three formats, any or all of which may be
emitted):
Format 1 (exon_ival); tab-delimited output tuple columns:
1. Reference name (RNAME in SAM format) + ';' + bin number
2. Sample index
3. EC start (inclusive) on forward strand
4. EC end (exclusive) on forward strand
Format 2 (exon_diff); tab-delimited output tuple columns:
1. Reference name (RNAME in SAM format) + ';' + bin number
2. max(EC start, bin start) (inclusive) on forward strand IFF diff is
positive and EC end (exclusive) on forward strand IFF diff is
negative
3. Sample index
4. '1' if alignment from which diff originates is "unique" according to
--tie-margin criterion; else '0'
5. +1 or -1 * count, the number of instances of a read sequence for
which to print exonic chunks
Note that only unique alignments are currently output as ivals and/or
diffs.
Format 3 (sam); tab-delimited output tuple columns:
Standard SAM output except fields are in different order, and the first
field corresponds to sample label. (Fields are reordered to facilitate
partitioning by sample name/RNAME and sorting by POS.) Each line
corresponds to a spliced alignment. The order of the fields is as
follows.
1. Sample index if outputting BAMs by sample OR
sample-rname index if outputting BAMs by chr
2. (Number string representing RNAME; see BowtieIndexReference
class in bowtie_index for conversion information) OR
'0' if outputting BAMs by chr
3. POS
4. QNAME
5. FLAG
6. MAPQ
7. CIGAR
8. RNEXT
9. PNEXT
10. TLEN
11. SEQ
12. QUAL
... + optional fields
Insertions/deletions (indel_bed)
tab-delimited output tuple columns:
1. 'I' or 'D' insertion or deletion line
2. Number string representing RNAME
3. Start position (Last base before insertion or
first base of deletion)
4. End position (Last base before insertion or last base of deletion
(exclusive))
5. Inserted sequence for insertions or deleted sequence for deletions
6. Sample index
----Next fields are for introns only; they are '\x1c' for indels----
7. '\x1c'
8. '\x1c'
--------------------------------------------------------------------
9. Number of instances of insertion or deletion in sample; this is
always +1 * count before bed_pre combiner/reducer
Read whose primary alignment is not end-to-end
Tab-delimited output tuple columns (unmapped):
1. Transcriptome Bowtie 2 index group number
2. SEQ
3. 1 if SEQ is reverse-complemented, else 0
4. QNAME
5. QUAL
Tab-delimited output tuple columns (readletized):
1. Readlet sequence or its reversed complement, whichever is first in
alphabetical order
2. read sequence ID + ('-' if readlet
sequence is reverse-complemented; else '+') + '\x1e' + displacement
of readlet's 5' end from read's 5' end + '\x1e' + displacement of
readlet's 3' end from read's 3' end (+, for EXACTLY one readlet of
a read sequence, '\x1e' + read sequence + '\x1e' +
(an '\x1f'-separated list A of unique sample labels with read
sequences that match the original read sequence) + '\x1e' +
(an '\x1f'-separated list of unique sample labels B with read
sequences that match the reversed complement of the original read
sequence)) + '\x1e' + (an '\x1f'-separated list of the number of
instances of the read sequence for each respective sample in list
A) + '\x1e' + (an '\x1f'-separated list of the number of instances
of the read sequence's reversed complement for each respective
sample in list B). Here, a read sequence ID takes the form X:Y,
where X is the "mapred_task_partition" environment variable -- a
unique index for a task within a job -- and Y is the index of the
read sequence relative to the beginning of the input stream.
Tab-delimited tuple columns (postponed_sam):
Standard 11+ -column raw SAM output
Single column (unique):
1. A unique read sequence
Two columns, exactly one line (dummy); ensures creation of intron
index:
1. character "-"
2. the word "dummy"
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input reads.
output_stream: where to emit exonic chunks and introns.
bowtie2_exe: filename of Bowtie2 executable; include path if not in
$PATH.
bowtie_index_base: the basename of the Bowtie1 index files associated
with the reference.
bowtie2_index_base: the basename of the Bowtie2 index files associated
with the reference.
manifest_file: filename of manifest
bowtie2_args: string containing precisely extra command-line arguments
to pass to first-pass Bowtie2.
bin_size: genome is partitioned in units of bin_size for later load
balancing.
verbose: True iff more informative messages should be written to
stderr.
exon_differentials: True iff EC differentials are to be emitted.
exon_intervals: True iff EC intervals are to be emitted.
report_multiplier: if verbose is True, the line number of an alignment
or read written to stderr increases exponentially with base
report_multiplier.
min_exon_size: minimum exon size searched for in intron_search.py later
in pipeline; used to determine how large a soft clip on one side of
a read is necessary to pass it on to intron search pipeline
search_filter: how large a soft clip on one side of a read is necessary
to pass it on to intron search pipeline
min_readlet_size: "capping" readlets (that is, readlets that terminate
at a given end of the read) are never smaller than this value
max_readlet_size: size of every noncapping readlet
readlet_interval: number of bases separating successive readlets along
the read
capping_multiplier: successive capping readlets on a given end of a
read are increased in size exponentially with base
capping_multiplier
drop_deletions: True iff deletions should be dropped from coverage
vector
gzip_level: compression level to use for temporary files
scratch: scratch directory for storing temporary files or None if
securely created temporary directory
index_count: number of transcriptome Bowtie 2 indexes to which to
assign unmapped reads for later realignment
output_bam_by_chr: True iff final output BAMs will be by chromosome
tie_margin: allowed score difference per 100 bases among ties in
max score. For example, 150 and 144 are tied alignment scores
for a 100-bp read when --tie-margin is 6.
no_realign: True iff job flow does not need more than readlets: this
usually means only a transcript index is being constructed
no_polyA: kill noncapping readlets that are all As and write as
unmapped all reads with polyA prefixes whose suffixes are <
min_exon_size
No return value.
"""
global _input_line_count
# Required length of prefix after poly(A) is trimmed
remaining_seq_size = max(min_exon_size - 1, 1)
polyA_set = frozenset(['A' * i
for i in xrange(1, remaining_seq_size + 1)] +
['T' * i
for i in xrange(1, remaining_seq_size + 1)] + [''])
reference_index = bowtie_index.BowtieIndexReference(bowtie_index_base)
manifest_object = manifest.LabelsAndIndices(manifest_file)
alignment_printer = AlignmentPrinter(manifest_object,
reference_index,
bin_size=bin_size,
output_stream=output_stream,
exon_ivals=exon_intervals,
exon_diffs=exon_differentials,
drop_deletions=drop_deletions,
output_bam_by_chr=output_bam_by_chr,
tie_margin=tie_margin)
# Get task partition to pass to align_reads_delegate.py
try:
task_partition = os.environ['mapred_task_partition']
except KeyError:
# Hadoop 2.x?
try:
task_partition = os.environ['mapreduce_task_partition']
except KeyError:
# A unit test is probably being run
task_partition = '0'
temp_dir = make_temp_dir(scratch)
register_cleanup(tempdel.remove_temporary_directories, [temp_dir])
align_file = os.path.join(temp_dir, 'first_pass_reads.temp.gz')
other_reads_file = os.path.join(temp_dir, 'other_reads.temp.gz')
second_pass_file = os.path.join(temp_dir, 'second_pass_reads.temp.gz')
k_value, _, _ = bowtie.parsed_bowtie_args(bowtie2_args)
nothing_doing = True
with xopen(True, align_file, 'w', gzip_level) as align_stream, \
xopen(True, other_reads_file, 'w', gzip_level) as other_stream:
for seq_number, ((seq, ),
xpartition) in enumerate(xstream(sys.stdin, 1)):
if no_polyA and (seq[:-remaining_seq_size] in polyA_set
or seq[remaining_seq_size:] in polyA_set):
if not no_realign:
'''If a sequence is too short without its poly(A) tail,
make all reads with that sequence unmapped. Technically,
this also kills poly(A)s at 5' ends, but we probably
couldn't align those sequences anyway.'''
reversed_complement_seq = seq[::-1].translate(
_reversed_complement_translation_table)
for is_reversed, name, qual in xpartition:
if is_reversed == '0':
alignment_printer.print_unmapped_read(
name, seq, qual)
else:
alignment_printer.print_unmapped_read(
name, reversed_complement_seq, qual[::-1])
continue
nothing_doing = False
'''Select highest-quality read with alphabetically last qname
for first-pass alignment.'''
best_name, best_mean_qual, best_qual_index, i = None, None, 0, 0
others_to_print = dlist()
for is_reversed, name, qual in xpartition:
_input_line_count += 1
others_to_print.append('\t'.join(
[str(seq_number), is_reversed, name, qual]))
mean_qual = (float(sum([ord(score)
for score in qual])) / len(qual))
if (mean_qual > best_mean_qual
or mean_qual == best_mean_qual and name > best_name):
best_qual_index = i
best_mean_qual = mean_qual
best_name = name
to_align = '\t'.join(
['%s\x1d%s' % (is_reversed, name), seq, qual])
i += 1
assert i >= 1
if i == 1:
print >> other_stream, str(seq_number)
else:
for j, other_to_print in enumerate(others_to_print):
if j != best_qual_index:
print >> other_stream, other_to_print
print >> align_stream, to_align
# Print dummy line
print 'dummy\t-\tdummy'
sys.stdout.flush(
) # this is REALLY important b/c called script will stdout
if nothing_doing:
# No input
sys.exit(0)
input_command = 'gzip -cd %s' % align_file
bowtie_command = ' '.join([
bowtie2_exe, bowtie2_args if bowtie2_args is not None else '',
' --sam-no-qname-trunc --local -t --no-hd --mm -x', bowtie2_index_base,
'--12 -'
])
delegate_command = ''.join([
sys.executable, ' ',
os.path.realpath(__file__)[:-3],
('_delegate.py --task-partition {task_partition} '
'--other-reads {other_reads} --second-pass-reads '
'{second_pass_reads} --min-readlet-size '
'{min_readlet_size} {drop_deletions} '
'--max-readlet-size {max_readlet_size} '
'--readlet-interval {readlet_interval} '
'--capping-multiplier {capping_multiplier:1.12f} '
'{verbose} --report-multiplier {report_multiplier:1.12f} '
'--k-value {k_value} '
'--bowtie-idx {bowtie_index_base} '
'--partition-length {bin_size} '
'--manifest {manifest_file} '
'{exon_differentials} {exon_intervals} '
'--gzip-level {gzip_level} '
'--search-filter {search_filter} '
'--index-count {index_count} '
'--tie-margin {tie_margin} '
'{no_realign} '
'{no_polyA} '
'{output_bam_by_chr}').format(
task_partition=task_partition,
other_reads=other_reads_file,
second_pass_reads=second_pass_file,
min_readlet_size=min_readlet_size,
drop_deletions=('--drop-deletions' if drop_deletions else ''),
max_readlet_size=max_readlet_size,
readlet_interval=readlet_interval,
capping_multiplier=capping_multiplier,
verbose=('--verbose' if verbose else ''),
report_multiplier=report_multiplier,
k_value=k_value,
bowtie_index_base=bowtie_index_base,
bin_size=bin_size,
manifest_file=manifest_file,
exon_differentials=('--exon-differentials'
if exon_differentials else ''),
exon_intervals=('--exon-intervals' if exon_intervals else ''),
gzip_level=gzip_level,
search_filter=search_filter,
index_count=index_count,
tie_margin=tie_margin,
no_realign=('--no-realign' if no_realign else ''),
no_polyA=('--no-polyA' if no_polyA else ''),
output_bam_by_chr=('--output-bam-by-chr'
if output_bam_by_chr else ''))
])
full_command = ' | '.join(
[input_command, bowtie_command, delegate_command])
print >>sys.stderr, \
'Starting first-pass Bowtie 2 with command: ' + full_command
bowtie_process = subprocess.Popen(' '.join(
['set -exo pipefail;', full_command]),
bufsize=-1,
stdout=sys.stdout,
stderr=sys.stderr,
shell=True,
executable='/bin/bash')
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError('Error occurred while reading first-pass Bowtie 2 '
'output; exitlevel was %d.' % return_code)
os.remove(align_file)
os.remove(other_reads_file)
if not no_realign:
input_command = 'gzip -cd %s' % second_pass_file
bowtie_command = ' '.join([
bowtie2_exe, bowtie2_args if bowtie2_args is not None else '',
' --sam-no-qname-trunc --local -t --no-hd --mm -x',
bowtie2_index_base, '--12 -'
])
delegate_command = ''.join([
sys.executable, ' ',
os.path.realpath(__file__)[:-3],
('_delegate.py --task-partition {task_partition} '
'--min-readlet-size {min_readlet_size} '
'{drop_deletions} '
'--max-readlet-size {max_readlet_size} '
'--readlet-interval {readlet_interval} '
'--capping-multiplier {capping_multiplier:012f} '
'{verbose} '
'--report-multiplier {report_multiplier:012f} '
'--k-value {k_value} '
'--bowtie-idx {bowtie_index_base} '
'--partition-length {bin_size} '
'--manifest {manifest_file} '
'{exon_differentials} {exon_intervals} '
'--gzip-level {gzip_level} '
'--search-filter {search_filter} '
'--index-count {index_count} '
'--tie-margin {tie_margin} '
'{output_bam_by_chr}').format(
task_partition=task_partition,
min_readlet_size=min_readlet_size,
drop_deletions=('--drop-deletions' if drop_deletions else ''),
readlet_interval=readlet_interval,
max_readlet_size=max_readlet_size,
capping_multiplier=capping_multiplier,
verbose=('--verbose' if verbose else ''),
report_multiplier=report_multiplier,
k_value=k_value,
bowtie_index_base=bowtie_index_base,
bin_size=bin_size,
manifest_file=manifest_file,
exon_differentials=('--exon-differentials'
if exon_differentials else ''),
exon_intervals=('--exon-intervals' if exon_intervals else ''),
gzip_level=gzip_level,
search_filter=search_filter,
index_count=index_count,
tie_margin=tie_margin,
output_bam_by_chr=('--output-bam-by-chr'
if output_bam_by_chr else ''))
])
full_command = ' | '.join(
[input_command, bowtie_command, delegate_command])
print >>sys.stderr, \
'Starting second-pass Bowtie 2 with command: ' + full_command
bowtie_process = subprocess.Popen(' '.join(
['set -exo pipefail;', full_command]),
bufsize=-1,
stdout=sys.stdout,
stderr=sys.stderr,
shell=True,
executable='/bin/bash')
return_code = bowtie_process.wait()
if return_code:
raise RuntimeError('Error occurred while reading second-pass '
'Bowtie 2 output; exitlevel was %d.' %
return_code)
sys.stdout.flush()
def edges_from_input_stream(input_stream, readlet_size=20,
min_overlap_exon_size=1):
""" Generates edges of directed acyclic graph (DAG) of introns.
A DAG is constructed for each strand. Each node of the DAG represents
a unique intron and is labeled by the tuple (intron_start, intron_end),
where intron_start is the (1-based) coordinate of the first base of the
intron, and intron_end is the coordinate of the first base after the
intron. An edge occurs between two introns A and B iff they do not
overlap, and no intron C occurs between A and B such that A, B, and C
do not overlap. The intron with larger coordinates is the child of
the intron with smaller coordinates. Weight each edge by the number of
exonic bases between the introns it connects.
The DAG has sources and sinks. Pad the DAG with new sources and sinks
as follows. Extend an edge to each original source
(source_start, source_end) from a new source labeled by the tuple
(None, max(1, source_start - readlet_size + 1)). Extend an edge
from each original sink (sink_start, sink_end) to a new sink labeled by
the tuple (sink_end + source_start + readlet_size - 1, None).
So each original source is assigned exactly one new source, and each
original sink is assigned exactly one new sink.
The paths through this DAG span all possible combinations of
nonoverlapping introns on the strand. Finding all subpaths (sequences
of introns), each of whose weights is <= readlet_size, redundantly
enumerates all possible combinations of introns a readlet
can overlap. Unfortunately, obtaining such combinations in
"hot spots," where there are many alternative splicings and short
exons, can become computationally intractable for large readlet sizes.
To (optionally) control these combinatorial blow-ups, impose an
effective minimum exon size min_overlap_exon_size by redefining overlap
between two introns: two introns overlap if there are fewer than
min_overlap_exon_size exonic bases between them.
The algorithm for generating the edges of the DAG operates on an input
stream whose lines are composed of the following tab-separated fields:
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Sample index
3. Intron start position (inclusive)
4. Intron end position (exclusive)
The input is partitioned by strand/sample index (fields 1-2) and sorted
by the remaining fields. INPUT COORDINATES ARE ASSUMED TO BE 1-INDEXED.
Introns are sorted by start position. To begin, the first set of
mutually overlapping introns are connected to their new sources. Two
data structures encode the graph as it is constructed [1]: the set
unlinked_nodes, containing introns that do not yet have any children,
and linked_nodes, a dictionary that, where possible, maps each intron A
to its corresponding successive nonoverlapping intron B with the
smallest end position read so far. With each new intron N read, the
nodes in unlinked_nodes and linked_nodes are checked for whether N is
their child, and these edges are yielded. (Note that by construction,
nodes are streamed in topological order.) Nodes from unlinked_nodes may
thus be promoted to linked_nodes. Each value node in linked_nodes is
also replaced with N if N has a smaller end position. Then every node A
in linked_nodes is checked for a path A -> B -> N. If such a path
exists, an edge can NEVER connect A with any successive introns, and A
is removed from linked_nodes. The algorithm continues until the end of
the strand is reached, when the edges connecting remaining
unlinked_nodes and their new sinks are yielded.
[1] Technically, there are three data structures. unlinked_nodes and
linked_nodes contain only indices of introns, and "introns" is
a dictionary that maps indices to tuples (intron_start, intron_end).
input_stream: where to find sorted introns of the form specified above.
fudge: by how much a readlet_size should be extended.
min_exon_size: minimum number of exonic bases between two introns
for them to be considered nonoverlapping.
Yield value: An edge tuple (strand,
sample_index,
(intron A start, intron A end),
(intron B start, intron B end)) or None
at the beginning of a new partition.
"""
global _input_line_count
for key, xpartition in xstream(input_stream, 2, skip_duplicates=True):
unlinked_nodes = set()
for q, value in enumerate(xpartition):
assert len(value) == 2
_input_line_count += 1
if not q:
# Denote start of new partition
yield None
# Handle first intron from partition
intron_start, intron_end = int(value[0]), int(value[1])
# Create fake source before first intron
fake_source = (
None,
max(intron_start - (readlet_size - 1), 1)
)
introns = {
0 : fake_source,
1 : (intron_start, intron_end)
}
linked_nodes = { 0 : 1 }
unlinked_nodes = set([1])
index = 2
# Yield first edge for strand (before first intron)
yield key + (fake_source,
(intron_start, intron_end))
first_intron = False
else:
# Handle next introns from partition
intron_start, intron_end = int(value[0]), int(value[1])
introns[index] = (intron_start, intron_end)
nodes_to_trash = []
for node in unlinked_nodes:
if intron_start >= introns[node][1] + \
min_overlap_exon_size:
nodes_to_trash.append(node)
for node in nodes_to_trash:
linked_nodes[node] = index
unlinked_nodes.remove(node)
unlinked_nodes.add(index)
nodes_to_trash = []
for node in linked_nodes:
intermediate_node = linked_nodes[node]
if intermediate_node in linked_nodes:
nodes_to_trash.append(node)
else:
yield key + (introns[node],
(intron_start, intron_end))
if introns[intermediate_node][1] > intron_end:
linked_nodes[node] = index
for node in nodes_to_trash:
del linked_nodes[node]
del introns[node]
index += 1
# Yield final edges for strand
for node in unlinked_nodes:
current_intron = introns[node]
yield key + (current_intron, (current_intron[1]
+ readlet_size - 1, None))
parser = argparse.ArgumentParser(description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument(\
'--verbose', action='store_const', const=True, default=False,
help='Print out extra debugging statements')
bowtie.add_args(parser)
group_reads.add_args(parser)
args = parser.parse_args()
start_time = time.time()
input_line_count = 0
reference_index = bowtie_index.BowtieIndexReference(
os.path.expandvars(args.bowtie_idx)
)
group_reads_object = group_reads.IndexGroup(args.index_count)
for (rname, poses, end_poses), xpartition in xstream(sys.stdin, 3,
skip_duplicates=True):
reverse_strand_string = rname[-1]
rname = rname[:-1]
read_seqs = dlist()
poses = [int(pos) for pos in poses.split(',')]
end_poses = [int(end_pos) for end_pos in end_poses.split(',')]
max_left_extend_size, max_right_extend_size = None, None
for left_extend_size, right_extend_size, read_seq in xpartition:
input_line_count += 1
max_left_extend_size = max(max_left_extend_size, int(left_extend_size))
max_right_extend_size \
= max(max_right_extend_size, int(right_extend_size))
read_seqs.append(read_seq)
intron_combo = zip(poses, end_poses)
assert max_left_extend_size is not None
assert max_right_extend_size is not None
def go(
manifest_object,
input_stream=sys.stdin,
output_stream=sys.stdout,
sample_fraction=0.05,
coverage_threshold=5,
collect_junctions=False,
verbose=False,
):
""" Runs Rail-RNA-junction_filter.
Filters out every junction from input_stream that is not either:
(1) in round(sample_fraction * (total number of samples)) samples OR
(2) found in at least coverage_threshold reads in at least one
sample.
Input (read from stdin)
----------------------------
Tab-delimited columns:
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Intron start position (inclusive)
3. Intron end position (exclusive)
4. '\x1f'-separated list of sample indexes in which junction was found
5. '\x1f'-separated list of numbers of reads in which junction was
found in respective sample specified by field 4
Input is partitioned by fields 1-3.
Hadoop output (written to stdout)
----------------------------
Tab-delimited tuple columns (filter):
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Sample index
3. Intron start position (inclusive)
4. Intron end position (exclusive)
If --collect-junctions is True:
Tab-delimited tuple columns (collect):
1. Reference name (RNAME in SAM format) +
'+' or '-' indicating which strand is the sense strand
2. Intron start position (inclusive)
3. Intron end position (exclusive)
4. '\x1f'-separated list of sample indexes in which junction was found
5. '\x1f'-separated list of numbers of reads in which junction was
found in respective sample specified by field 4
ALL OUTPUT COORDINATES ARE 1-INDEXED.
input_stream: where to find input junctions
output_stream: where to write output
manifest_object: object of class LabelsAndIndices that maps indices
to labels and back; used to count number of samples.
sample_fraction: fraction of samples in which a junction must appear
to pass filter if coverage_threshold criterion is not satisfied
coverage_threshold: number of reads that must overlap junction in at
least one sample to pass filter of sample_fraction criterion is not
satisfied
collect_junctions: collects and outputs junctions across samples;
ignores sample_fraction and coverage_threshold
verbose: output extra debugging statements
Return value: tuple (input line count, output line count)
"""
input_line_count, output_line_count = 0, 0
min_sample_count = int(round(len(manifest_object.label_to_index) * sample_fraction))
for (rname_and_strand, pos, end_pos), xpartition in xstream(input_stream, 3):
sample_indexes = defaultdict(int)
for current_sample_indexes, current_sample_counts in xpartition:
input_line_count += 1
current_sample_counts = current_sample_counts.split("\x1f")
for i, sample_index in enumerate(current_sample_indexes.split("\x1f")):
sample_indexes[sample_index] += int(current_sample_counts[i])
pos, end_pos = int(pos), int(end_pos)
if collect_junctions:
samples_to_dump = sorted(sample_indexes.items(), key=lambda sample: int(sample[0]))
print >> output_stream, "collect\t%s\t%012d\t%012d\t%s\t%s" % (
rname_and_strand,
pos,
end_pos,
",".join([sample[0] for sample in samples_to_dump]),
",".join([str(sample[1]) for sample in samples_to_dump]),
)
output_line_count += 1
sample_count = len(sample_indexes)
max_coverage = max(sample_indexes.values())
if end_pos > pos and (
sample_count >= min_sample_count or (max_coverage >= coverage_threshold and coverage_threshold != -1)
):
for sample_index in sample_indexes:
print >> output_stream, "filter\t%s\t%s\t%012d\t%012d" % (rname_and_strand, sample_index, pos, end_pos)
output_line_count += 1
elif verbose:
print >>sys.stderr, (
"Junction (%s, %d, %d) filtered out; it appeared in %d "
"sample(s), and its coverage in any one sample did "
"not exceed %d."
) % (rname_and_strand, pos, end_pos, sample_count, max_coverage)
return input_line_count, output_line_count
combined_stream,
retrieved_intron_counts,
instance=args.read_instance)
import subprocess
sorted_combined_file = os.path.join(temp_dir_path, 'combined.sorted.temp')
subprocess.check_call(' '.join([
'sort -T %s -k1,1' % temp_dir_path, combined_file, '>',
sorted_combined_file
]),
bufsize=-1,
shell=True)
relevant = 0
retrieved = 0
relevant_and_retrieved = 0
with open(sorted_combined_file) as sorted_combined_stream:
for (name, ), xpartition in xstream(sorted_combined_stream, 1):
relevant_and_retrieved_instances = list(xpartition)
ts = [
instance[:-1] for instance in relevant_and_retrieved_instances
if instance[-1] == 't' and (
args.coverage_threshold is None or any([
intron_counts[intron] <= args.coverage_threshold
for intron in instance[:-1]
]))
]
rs = [
instance[:-1] for instance in relevant_and_retrieved_instances
if instance[-1] == 'r' and (
args.coverage_threshold is None or any([
intron_counts[intron] <= args.coverage_threshold
for intron in instance[:-1]