def main():
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description="""Convert featureCount output to BED12 with exon-union coordinates at
meta-feature level.""")
parser.add_argument('tsv', help="The featureCount tsv file")
parser.add_argument('out', help="The (output) BED12 file, compressed by default")
parser.add_argument('-p', '--num-cpus', help="The number of CPUs to use",
type=int, default=12)
logging_utils.add_logging_options(parser)
args = parser.parse_args()
logging_utils.update_logging(args)
msg = "Reading featureCount tsv file"
logger.info(msg)
tsv = pd.read_csv(args.tsv,
usecols=['Geneid', 'Chr', 'Start', 'End', 'Strand', 'Length'],
sep='\t',
comment='#')
msg = "Merging..."
logger.info(msg)
merged = parallel.apply_parallel(tsv, args.num_cpus, merge_gene_group)
merged = pd.DataFrame(merged)
msg = "Sorting..."
logger.info(msg)
# We will break ties among transcripts by the order they appear
# in the GTF file. This is the same way star breaks ties.
merged = bed_utils.sort(merged)
msg = "Writing BED12 to disk"
logger.info(msg)
fields = bed_utils.bed12_field_names
fields.append('length')
bed_utils.write_bed(merged[fields], args.out)
def main():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description='''Label the ORFs based on their transcript
exon structure wrt the annotated transcripts.''')
parser.add_argument('annotated_transcripts',
help='''The annotated transcripts for the genome
in BED12+ format.''')
parser.add_argument('extracted_orfs',
help='''The ORFs extracted from the transcripts
in BED12+ format.''')
parser.add_argument('out', help='''The output (BED12+.gz) file.''')
parser.add_argument('-e',
'--annotated-exons',
help='''The annotated transcript
exons can be passed with this option. If they are not given, they will be
split from the annotated transcripts.''',
default=None)
parser.add_argument('-o',
'--orf-exons',
help='''The exon blocks for the ORFs, in BED6+ format,
obtained from "split-bed12-blocks". If they are not given, they will be split from the
extracted ORFs.''',
default=None)
parser.add_argument('-n',
'--nonoverlapping-label',
help='''If this option is given,
then the ORFs which do not overlap the annotated transcripts at all will be given this label.
By default, remaining oof overlapping ORFs are assigned the "overlap" label.
If not given, the ORFs outside of annotated regions are labeled as "suspect".''',
default=None)
parser.add_argument('-l',
'--label-prefix',
help='''This string is prepended to all labels
assigned to ORFs, e.g. to indicate ORFs from a de novo assembly (Rp-Bp assigns the label
"novel" to these, however the string is not prepended to "canonical ORFs").''',
default='')
parser.add_argument('-f',
'--filter',
help='''If this flag is given, then ORFs
which are completely covered by an annotated transcript are discarded. Use to filter
uninteresting ORFs from a de novo assembly.''',
action='store_true')
parser.add_argument('-p',
'--num-cpus',
help='''The number of CPUs to use to perform
BED operations.''',
type=int,
default=default_num_cpus)
logging_utils.add_logging_options(parser)
args = parser.parse_args()
logging_utils.update_logging(args)
msg = "Reading annotated transcripts"
logger.info(msg)
annotated_transcripts = bed_utils.read_bed(args.annotated_transcripts)
# get the annotated transcript exons
if args.annotated_exons is None:
msg = "Splitting the annotated transcripts into exon blocks"
logger.info(msg)
annotated_exons = bed_utils.split_bed12(annotated_transcripts,
num_cpus=args.num_cpus,
progress_bar=True)
else:
msg = "Reading the annotated transcript exons"
logger.info(msg)
annotated_exons = bed_utils.read_bed(args.annotated_exons)
msg = "Reading extracted ORFs"
logger.info(msg)
extracted_orfs = bed_utils.read_bed(args.extracted_orfs)
if args.orf_exons is None:
msg = "Splitting the extracted ORFs into exon blocks"
logger.info(msg)
extracted_orf_exons = bed_utils.split_bed12(extracted_orfs,
num_cpus=args.num_cpus,
progress_bar=True)
else:
msg = "Reading the extracted ORFs exons"
logger.info(msg)
extracted_orf_exons = bed_utils.read_bed(args.orf_exons)
msg = "Found {} extracted ORFs with {} exons".format(
len(extracted_orfs), len(extracted_orf_exons))
logger.debug(msg)
# filter out the ORFs that are entirely within annotated transcripts
if args.filter:
msg = "Removing ORFs which are completely covered by the annotated transcripts"
logger.info(msg)
nonoverlapping_ids = bed_utils.subtract_bed(extracted_orf_exons,
annotated_exons,
min_a_overlap=1)
m_unfiltered = extracted_orfs['id'].isin(nonoverlapping_ids)
extracted_orfs = extracted_orfs[m_unfiltered]
# discard the unnecessary exons
m_unfiltered = extracted_orf_exons['id'].isin(nonoverlapping_ids)
extracted_orf_exons = extracted_orf_exons[m_unfiltered]
msg = "After filtering, {} extracted ORFs remain".format(
len(extracted_orfs))
logger.info(msg)
# annotate and remove the ORFs which do not at all overlap the annotations
if args.nonoverlapping_label is not None:
nonoverlapping_ids = bed_utils.subtract_bed(
extracted_orfs,
annotated_transcripts,
exons_a=extracted_orf_exons,
exons_b=annotated_exons)
m_nonoverlapping = extracted_orf_exons['id'].isin(nonoverlapping_ids)
extracted_orf_exons = extracted_orf_exons[~m_nonoverlapping]
m_nonoverlapping = extracted_orfs['id'].isin(nonoverlapping_ids)
extracted_orfs.loc[m_nonoverlapping,
'orf_type'] = args.nonoverlapping_label
msg = ("Found {} ORFs completely non-overlapping annotated transcripts"
.format(len(nonoverlapping_ids)))
logger.info(msg)
msg = "Removing the annotated UTRs from the transcripts"
logger.info(msg)
canonical_orfs = bed_utils.retain_all_thick_only(annotated_transcripts,
num_cpus=args.num_cpus)
msg = "Splitting the canonical ORFs into exons"
logger.info(msg)
canonical_orf_exons = bed_utils.split_bed12(canonical_orfs,
num_cpus=args.num_cpus,
progress_bar=True)
msg = "Extracting annotated 5' leader regions"
logger.info(msg)
five_prime_regions = bed_utils.retain_all_five_prime_of_thick(
annotated_transcripts, num_cpus=args.num_cpus)
if len(five_prime_regions) == 0:
msg = "No annotated 5' leader regions were found"
logger.warning(msg)
msg = "Splitting the 5' leaders into exons"
logger.info(msg)
five_prime_exons = bed_utils.split_bed12(five_prime_regions,
num_cpus=args.num_cpus,
progress_bar=True)
msg = "Extracting annotated 3' trailer regions"
logger.info(msg)
three_prime_regions = bed_utils.retain_all_three_prime_of_thick(
annotated_transcripts, num_cpus=args.num_cpus)
if len(three_prime_regions) == 0:
msg = "No annotated 3' trailer regions were found"
logger.warning(msg)
msg = "Splitting the 3' trailers into exons"
logger.info(msg)
three_prime_exons = bed_utils.split_bed12(three_prime_regions,
num_cpus=args.num_cpus,
progress_bar=True)
msg = "Splitting non-coding transcripts into exons"
logger.info(msg)
m_no_thick_start = annotated_transcripts['thick_start'] == -1
m_no_thick_end = annotated_transcripts['thick_end'] == -1
m_no_thick = m_no_thick_start & m_no_thick_end
noncoding_transcripts = annotated_transcripts[m_no_thick]
noncoding_exons = bed_utils.split_bed12(noncoding_transcripts,
num_cpus=args.num_cpus,
progress_bar=True)
# First, remove all in-frame (canonical, canonical variants), and also within and oof ORFs
msg = "Marking canonical and extracted ORFs with the same stop codon"
logger.info(msg)
# first, add the "true" ORF end
m_reverse_canonical = canonical_orfs['strand'] == '-'
canonical_orfs['orf_end'] = canonical_orfs['end']
canonical_orfs.loc[m_reverse_canonical,
'orf_end'] = canonical_orfs.loc[m_reverse_canonical,
'start']
m_reverse_extracted = extracted_orfs['strand'] == '-'
extracted_orfs['orf_end'] = extracted_orfs['end']
extracted_orfs.loc[m_reverse_extracted,
'orf_end'] = extracted_orfs.loc[m_reverse_extracted,
'start']
# then, find extracted ORFs with the same "orf_end" (and seqname, strand) as canonical ORFs
merge_fields = ['seqname', 'strand', 'orf_end']
canonical_extracted_orf_ends = canonical_orfs.merge(
extracted_orfs, on=merge_fields, suffixes=['_canonical', '_extracted'])
# finally, pull this into a set
zip_it = zip(canonical_extracted_orf_ends['id_canonical'],
canonical_extracted_orf_ends['id_extracted'])
canonical_extracted_matching_ends = {(c, a) for c, a in zip_it}
msg = "Finding ORFs which exactly overlap the canonical ORFs"
logger.info(msg)
exact_matches = bed_utils.get_bed_overlaps(canonical_orf_exons,
extracted_orf_exons,
min_a_overlap=1,
min_b_overlap=1)
exact_match_orf_ids = {m.b_info for m in exact_matches}
m_exact_orf_matches = extracted_orf_exons['id'].isin(exact_match_orf_ids)
extracted_orf_exons = extracted_orf_exons[~m_exact_orf_matches]
m_canonical = extracted_orfs['id'].isin(exact_match_orf_ids)
label = 'canonical'
extracted_orfs.loc[m_canonical, 'orf_type'] = label
msg = "Found {} canonical ORFs".format(len(exact_match_orf_ids))
logger.info(msg)
msg = "Finding truncated canonical ORFs"
logger.info(msg)
truncated_matches = bed_utils.get_bed_overlaps(canonical_orf_exons,
extracted_orf_exons,
min_b_overlap=1)
truncated_match_ids = {
m.b_info
for m in truncated_matches
if (m.a_info, m.b_info) in canonical_extracted_matching_ends
}
m_truncated_matches = extracted_orf_exons['id'].isin(truncated_match_ids)
extracted_orf_exons = extracted_orf_exons[~m_truncated_matches]
m_canonical_truncated = extracted_orfs['id'].isin(truncated_match_ids)
msg = "Finding extended canonical ORFs"
logger.info(msg)
extended_matches = bed_utils.get_bed_overlaps(canonical_orf_exons,
extracted_orf_exons,
min_a_overlap=1)
# For standard assembly, we also need to make sure that
# all extended matches are fully contained within the
# transcript structure (i.e start upstream but otherwise
# have the same structure).
if args.nonoverlapping_label is None:
transcript_matches = bed_utils.get_bed_overlaps(annotated_exons,
extracted_orf_exons,
min_b_overlap=1)
transcript_match_pairs = {(m.a_info, m.b_info)
for m in transcript_matches}
extended_match_ids = {
m.b_info
for m in extended_matches
if (m.a_info, m.b_info) in transcript_match_pairs and (
m.a_info, m.b_info) in canonical_extracted_matching_ends
}
else:
extended_match_ids = {
m.b_info
for m in extended_matches
if (m.a_info, m.b_info) in canonical_extracted_matching_ends
}
m_extended_matches = extracted_orf_exons['id'].isin(extended_match_ids)
extracted_orf_exons = extracted_orf_exons[~m_extended_matches]
m_canonical_extended = extracted_orfs['id'].isin(extended_match_ids)
m_canonical_variants = m_canonical_truncated | m_canonical_extended
label = "{}canonical_variant".format(args.label_prefix)
extracted_orfs.loc[m_canonical_variants, 'orf_type'] = label
msg = "Found {} canonical_variant ORFs".\
format(len(extended_match_ids | truncated_match_ids))
logger.info(msg)
msg = ("Finding within canonical ORFs that do not share an "
"annotated stop codon with a canonical ORF (e.g. in "
"frame stop, out-of-frame)")
logger.info(msg)
within_ids = {
m.b_info
for m in truncated_matches if m.b_info not in truncated_match_ids
}
m_within_matches = extracted_orf_exons['id'].isin(within_ids)
extracted_orf_exons = extracted_orf_exons[~m_within_matches]
m_within = extracted_orfs['id'].isin(within_ids)
label = "{}within".format(args.label_prefix)
extracted_orfs.loc[m_within, 'orf_type'] = label
msg = "Found {} within ORFs".format(len(within_ids))
logger.info(msg)
# find all overlapping ORFs
msg = "Finding all UTR overlap matches"
logger.info(msg)
out_of_frame_matches = bed_utils.get_bed_overlaps(canonical_orf_exons,
extracted_orf_exons)
leader_matches = bed_utils.get_bed_overlaps(five_prime_exons,
extracted_orf_exons)
trailer_matches = bed_utils.get_bed_overlaps(three_prime_exons,
extracted_orf_exons)
msg = ("Labeling ORFs which have (out-of-frame) overlaps with both a "
"canonical ORF and annotated leaders or trailers")
logger.info(msg)
# We need to choose how to ensure that up-/downstream overlaps are unique.
# Where an ORF overlaps both the 5'UTR and the 3'UTR of different same
# sense overlapping transcripts, it is assigned by default to the downstream overlap.
# For de novo, everything is labeled as overlap.
leader_match_pairs = {(m.a_info, m.b_info) for m in leader_matches}
trailer_match_pairs = {(m.a_info, m.b_info) for m in trailer_matches}
if args.nonoverlapping_label is None:
# For standard assembly, we also need to make sure that
# all overlap matches are fully contained within the
# transcript structure.
transcript_matches = bed_utils.get_bed_overlaps(annotated_exons,
extracted_orf_exons,
min_b_overlap=1)
transcript_match_pairs = {(m.a_info, m.b_info)
for m in transcript_matches}
leader_overlap_pairs = {
(m.a_info, m.b_info)
for m in out_of_frame_matches
if (m.a_info, m.b_info) in leader_match_pairs and (
m.a_info, m.b_info) not in trailer_match_pairs and (
m.a_info, m.b_info) in transcript_match_pairs
}
trailer_overlap_pairs = {
(m.a_info, m.b_info)
for m in out_of_frame_matches
if (m.a_info, m.b_info) in trailer_match_pairs and (
m.a_info, m.b_info) not in leader_match_pairs and (
m.a_info, m.b_info) in transcript_match_pairs
}
# We do not assign preference where the ORF overlaps both sides
# of the coding sequence on the same transcript, any ORF
# satisfying both will be labeled simply as overlap.
overlap_ids = {
m.b_info
for m in out_of_frame_matches
if (m.a_info, m.b_info) in leader_match_pairs and (
m.a_info, m.b_info) in trailer_match_pairs and (
m.a_info, m.b_info) in transcript_match_pairs
}
trailer_overlap_ids = {
pair[1]
for pair in trailer_overlap_pairs if pair[1] not in overlap_ids
}
leader_overlap_ids = {
pair[1]
for pair in leader_overlap_pairs if
pair[1] not in trailer_overlap_ids and pair[1] not in overlap_ids
}
m_overlap_matches = extracted_orf_exons['id'].isin(overlap_ids)
extracted_orf_exons = extracted_orf_exons[~m_overlap_matches]
m_leader_overlap_matches = extracted_orf_exons['id'].isin(
leader_overlap_ids)
extracted_orf_exons = extracted_orf_exons[~m_leader_overlap_matches]
m_five_prime_overlap = extracted_orfs['id'].isin(leader_overlap_ids)
label = "{}five_prime_overlap".format(args.label_prefix)
extracted_orfs.loc[m_five_prime_overlap, 'orf_type'] = label
m_trailer_overlap_matches = extracted_orf_exons['id'].isin(
trailer_overlap_ids)
extracted_orf_exons = extracted_orf_exons[~m_trailer_overlap_matches]
m_three_prime_overlap = extracted_orfs['id'].isin(trailer_overlap_ids)
label = "{}three_prime_overlap".format(args.label_prefix)
extracted_orfs.loc[m_three_prime_overlap, 'orf_type'] = label
msg = "Found {} five_prime_overlap ORFs".format(
len(leader_overlap_ids))
logger.info(msg)
msg = "Found {} three_prime_overlap ORFs".format(
len(trailer_overlap_ids))
logger.info(msg)
else:
overlap_ids = {m.b_info for m in out_of_frame_matches}
overlap_ids |= {m.b_info for m in leader_matches}
overlap_ids |= {m.b_info for m in trailer_matches}
m_overlap_matches = extracted_orf_exons['id'].isin(overlap_ids)
extracted_orf_exons = extracted_orf_exons[~m_overlap_matches]
m_overlap = extracted_orfs['id'].isin(overlap_ids)
label = "{}overlap".format(args.label_prefix)
extracted_orfs.loc[m_overlap, 'orf_type'] = label
msg = "Found {} overlap ORFs".format(len(overlap_ids))
logger.info(msg)
msg = "Finding ORFs completely within 5' or 3' leaders"
logger.info(msg)
leader_matches = bed_utils.get_bed_overlaps(five_prime_exons,
extracted_orf_exons,
min_b_overlap=1)
leader_ids = {m.b_info for m in leader_matches}
m_leader_matches = extracted_orf_exons['id'].isin(leader_ids)
extracted_orf_exons = extracted_orf_exons[~m_leader_matches]
m_five_prime = extracted_orfs['id'].isin(leader_ids)
label = "{}five_prime".format(args.label_prefix)
extracted_orfs.loc[m_five_prime, 'orf_type'] = label
msg = "Found {} five_prime ORFs".format(len(leader_ids))
logger.info(msg)
trailer_matches = bed_utils.get_bed_overlaps(three_prime_exons,
extracted_orf_exons,
min_b_overlap=1)
trailer_ids = {m.b_info for m in trailer_matches}
m_trailer_matches = extracted_orf_exons['id'].isin(trailer_ids)
extracted_orf_exons = extracted_orf_exons[~m_trailer_matches]
m_three_prime = extracted_orfs['id'].isin(trailer_ids)
label = "{}three_prime".format(args.label_prefix)
extracted_orfs.loc[m_three_prime, 'orf_type'] = label
msg = "Found {} three_prime ORFs".format(len(trailer_ids))
logger.info(msg)
msg = "Finding ORFs completely within annotated, non-coding transcripts"
logger.info(msg)
noncoding_matches = bed_utils.get_bed_overlaps(noncoding_exons,
extracted_orf_exons,
min_b_overlap=1)
noncoding_ids = {m.b_info for m in noncoding_matches}
m_noncoding_matches = extracted_orf_exons['id'].isin(noncoding_ids)
extracted_orf_exons = extracted_orf_exons[~m_noncoding_matches]
m_noncoding = extracted_orfs['id'].isin(noncoding_ids)
label = "{}noncoding".format(args.label_prefix)
extracted_orfs.loc[m_noncoding, 'orf_type'] = label
msg = "Found {} noncoding ORFs".format(len(noncoding_ids))
logger.info(msg)
# all of the remaining ORFs fall into the "suspect" category
suspect_ids = {orf_id for orf_id in extracted_orf_exons['id']}
m_suspect = extracted_orfs['id'].isin(suspect_ids)
label = "{}suspect".format(args.label_prefix)
extracted_orfs.loc[m_suspect, 'orf_type'] = label
n_suspect_ids = len(suspect_ids)
msg = "Remaining {} ORFs labeled as suspect".format(n_suspect_ids)
logger.info(msg)
m_no_orf_type = extracted_orfs['orf_type'].isnull()
msg = "Found {} unlabeled ORFs".format(sum(m_no_orf_type))
logger.info(msg)
msg = "Writing ORFs with labels to disk"
logger.info(msg)
extracted_orfs = bed_utils.sort(extracted_orfs)
msg = ("The ORF labels will be written to {} in the next major release.".
format(args.out))
logger.warning(msg)
additional_columns = ['orf_num', 'orf_len', 'orf_type']
fields = bed_utils.bed12_field_names + additional_columns
orfs_genomic = extracted_orfs[fields]
bed_utils.write_bed(orfs_genomic, args.extracted_orfs)
label_columns = ['id', 'duplicates', 'orf_type']
extracted_orfs = extracted_orfs[label_columns]
bed_utils.write_bed(extracted_orfs, args.out)
def main():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description='''Prepare a reference genome and matching
annotations, including labelled ORFs, for use with the Rp-Bp periodicity estimation
and ORF translation prediction pipeline.''')
parser.add_argument('config', help='''The (yaml) configuration file''')
parser.add_argument('--overwrite',
help='''If this flag is present, existing files
will be overwritten.''',
action='store_true')
slurm.add_sbatch_options(parser,
num_cpus=default_num_cpus,
mem=default_mem)
logging_utils.add_logging_options(parser)
pgrm_utils.add_star_options(parser, star_executable)
args = parser.parse_args()
logging_utils.update_logging(args)
config = yaml.load(open(args.config), Loader=yaml.FullLoader)
# check required callable programs, config keys and files
programs = [
'extract-orf-coordinates', 'label-orfs', 'bowtie2-build-s',
'split-bed12-blocks', 'gtf-to-bed12', args.star_executable
]
shell_utils.check_programs_exist(programs)
required_keys = [
'genome_base_path', 'genome_name', 'gtf', 'fasta', 'ribosomal_fasta',
'ribosomal_index', 'star_index'
]
utils.check_keys_exist(config, required_keys)
files = [config['gtf'], config['fasta'], config['ribosomal_fasta']]
if 'de_novo_gtf' in config:
files += [config['de_novo_gtf']]
utils.check_files_exist(files, source='prepare-rpbp-genome')
# check if we want to use slurm
if args.use_slurm:
cmd = ' '.join(sys.argv)
slurm.check_sbatch(cmd, args=args)
return
call = not args.do_not_call
# the rRNA index
cmd = "bowtie2-build-s {} {}".format(config['ribosomal_fasta'],
config['ribosomal_index'])
in_files = [config['ribosomal_fasta']]
out_files = pgrm_utils.get_bowtie2_index_files(config['ribosomal_index'])
shell_utils.call_if_not_exists(cmd,
out_files,
in_files=in_files,
overwrite=args.overwrite,
call=call)
# the STAR index
mem = utils.human2bytes(args.mem)
cmd = ("{} --runMode genomeGenerate --genomeDir {} --genomeFastaFiles {} "
"--runThreadN {} --limitGenomeGenerateRAM {}".format(
args.star_executable, config['star_index'], config['fasta'],
args.num_cpus, mem))
in_files = [config['fasta']]
out_files = pgrm_utils.get_star_index_files(config['star_index'])
shell_utils.call_if_not_exists(cmd,
out_files,
in_files=in_files,
overwrite=args.overwrite,
call=call)
# get the ORFs
get_orfs(config['gtf'], args, config, is_annotated=True, is_de_novo=False)
# we will use these files later in the pipeline
annotated_orfs = filenames.get_orfs(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=True,
is_de_novo=False)
orfs_genomic = filenames.get_orfs(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'))
annotated_exons_file = filenames.get_exons(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=True,
is_de_novo=False)
exons_file = filenames.get_exons(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'))
annotated_labeled_orfs = filenames.get_labels(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=True,
is_de_novo=False)
labeled_orfs = filenames.get_labels(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'))
use_gff3_specs = config['gtf'].endswith('gff')
gtf_file = filenames.get_gtf(config['genome_base_path'],
config['genome_name'],
is_gff3=use_gff3_specs,
is_star_input=True)
# now, check if we have a de novo assembly
if 'de_novo_gtf' in config:
get_orfs(config['de_novo_gtf'],
args,
config,
is_annotated=False,
is_de_novo=True)
# we need to concat the ORF and exon files
de_novo_orfs = filenames.get_orfs(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=False,
is_de_novo=True)
orfs_files = [annotated_orfs, de_novo_orfs]
orfs_files_str = ' '.join(orfs_files)
msg = ("Concatenating files. Output file: {}; Input files: {}".format(
orfs_genomic, orfs_files_str))
logger.info(msg)
if call:
concatenated_bed = bed_utils.concatenate(orfs_files, sort_bed=True)
concatenated_bed['orf_num'] = range(len(concatenated_bed))
additional_columns = ['orf_num', 'orf_len', 'orf_type']
fields = bed_utils.bed12_field_names + additional_columns
bed_utils.write_bed(concatenated_bed[fields], orfs_genomic)
else:
msg = "Skipping concatenation due to --call value"
logger.info(msg)
de_novo_exons_file = filenames.get_exons(config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=False,
is_de_novo=True)
exons_files = [annotated_exons_file, de_novo_exons_file]
exons_files_str = ' '.join(exons_files)
msg = ("Concatenating files. Output file: {}; Input files: {}".format(
exons_file, exons_files_str))
logger.info(msg)
if call:
concatenated_bed = bed_utils.concatenate(exons_files,
sort_bed=True)
fields = bed_utils.bed6_field_names + [
'exon_index', 'transcript_start'
]
bed_utils.write_bed(concatenated_bed[fields], exons_file)
else:
msg = "Skipping concatenation due to --call value"
logger.info(msg)
de_novo_labeled_orfs = filenames.get_labels(
config['genome_base_path'],
config['genome_name'],
note=config.get('orf_note'),
is_annotated=False,
is_de_novo=True)
label_files = [annotated_labeled_orfs, de_novo_labeled_orfs]
label_files_str = ' '.join(label_files)
msg = ("Concatenating files. Output file: {}; Input files: {}".format(
labeled_orfs, label_files_str))
logger.info(msg)
if call:
# not sorted, as is
concatenated_bed = bed_utils.concatenate(label_files,
sort_bed=False)
bed_utils.write_bed(concatenated_bed, labeled_orfs)
else:
msg = "Skipping concatenation due to --call value"
logger.info(msg)
# we also need to concat the annotations to inform STAR
# there is no particular reason to merge and sort the files, so
# we just concatenate them...
if (config['de_novo_gtf'].endswith('gff') == use_gff3_specs):
cmd = ("awk '!/^#/' {} {} > {}".format(config['gtf'],
config['de_novo_gtf'],
gtf_file))
in_files = [config['gtf'], config['de_novo_gtf']]
out_files = [gtf_file]
shell_utils.call_if_not_exists(cmd,
out_files,
in_files=in_files,
overwrite=args.overwrite,
call=call)
else:
msg = (
"Skipping concatenation due to mismatch in format specifications (GTF2/GFF3)"
"for reference and do novo annotations. Symlink to reference annotations created."
)
logger.warning(msg)
if os.path.exists(config['gtf']):
shell_utils.create_symlink(config['gtf'], gtf_file, call)
else:
# if we do not have a de novo assembly, symlink the files
if os.path.exists(annotated_orfs):
shell_utils.create_symlink(annotated_orfs, orfs_genomic, call)
if os.path.exists(annotated_exons_file):
shell_utils.create_symlink(annotated_exons_file, exons_file, call)
if os.path.exists(annotated_labeled_orfs):
shell_utils.create_symlink(annotated_labeled_orfs, labeled_orfs,
call)
if os.path.exists(config['gtf']):
shell_utils.create_symlink(config['gtf'], gtf_file, call)
def main():
parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description="""Given a list of ORFs with associated Bayes
factors and a fasta sequence file, this script extracts the sequences of the ORFs whose
Bayes factor exceeds the given threshold. Finally, biopython is used to translate the
selected ORFs into protein sequences. The min-length and minimum-profile-sum filters
are applied in the obvious way. For both BF and chi-square predictions, only ORFs
which have more reads in the first reading frame than either of the other two will
be selected as translated. (This is called the 'frame filter' below.) The selection
based on Bayes factors follows this logic: if max_bf_var is given, then it and
min_bf_mean are taken as a hard threshold on the estimated Bayes factor mean.
If min_bf_likelihood is given, then this min_bf_mean is taken as the boundary value;
that is, an ORF is 'translated' if: [P(bf > min_bf_mean)] > min_bf_likelihood.
If both max_bf_var and min_bf_likelihood are None, then min_bf_mean is taken as a
hard threshold on the mean for selecting translated ORFs. If both max_bf_var and
min_bf_likelihood are given, then both filters will be applied and the result will
be the intersection. If the --use-chi-square option is given, the significance value is
Bonferroni-corrected based on the number of ORFs which meet the length, profile
and frame filters.""")
parser.add_argument('bayes_factors', help="""The file containing the ORFs and Bayes'
factors (BED12+).""")
parser.add_argument('fasta', help="The *genome* fasta file")
parser.add_argument('predicted_orfs', help="""The (output) BED12+ file containing
the predicted ORFs.""")
parser.add_argument('predicted_dna_sequences', help="""The (output) fasta file
containing the predicted ORF sequences, as DNA sequences.""")
parser.add_argument('predicted_protein_sequences', help="""The (output) fasta file
containing the predicted ORF sequences, as protein sequences.""")
parser.add_argument('--select-longest-by-stop', help="""If this flag is given, then
the selected ORFs will be merged based on stop codons. In particular, only the
longest translated ORF at each stop codon will be selected.""", action='store_true')
parser.add_argument('--select-best-overlapping', help="""If this flag is given, then
only the ORF with the highest estimated Bayes factor will be kept among each
set of overlapping ORFs. N.B. This filter is applied *AFTER* selecting the
longest ORF at each stop codon, if the --select-longest-by-stop flag is given.""",
action='store_true')
parser.add_argument('--min-length', help="The minimum length to predict an ORF as translated",
type=int, default=translation_options['orf_min_length'])
parser.add_argument('--min-bf-mean', help="""The minimum Bayes' factor mean to predict
an ORF as translated (use --help for more details)""",
type=float, default=translation_options['min_bf_mean'])
parser.add_argument('--max-bf-var', help="""The maximum Bayes' factor variance to predict
an ORF as translated (use --help for more details).""",
type=float, default=translation_options['max_bf_var'])
parser.add_argument('--min-bf-likelihood', help="""If given, then this is taken a threshold
on the likelihood of translation (use --help for more details).""",
type=float, default=translation_options['min_bf_likelihood'])
parser.add_argument('--min-profile', help="""The minimum sum across all reading frames to consider
an ORF as translated""", type=float, default=translation_options['orf_min_profile_count'])
parser.add_argument('--chi-square-only', help="""If this flag is present, then the
chi square value will be used to predict ORFs rather than the Bayes' factor.""",
action='store_true')
parser.add_argument('--chisq-significance-level', help="""If using chi square, then this
value is Bonferroni corrected and used as the significance cutoff, else it is ignored.""",
type=float, default=translation_options['chisq_alpha'])
parser.add_argument('--filtered-orf-types', help=""""A list of ORF types which will be
removed before selecting the final prediction set.""", nargs='*', default=[])
logging_utils.add_logging_options(parser)
args = parser.parse_args()
logging_utils.update_logging(args)
# first, extract all of the predictions which exceed the threshold
msg = "Reading Bayes factor information"
logger.info(msg)
bayes_factors = bed_utils.read_bed(args.bayes_factors)
if len(args.filtered_orf_types) > 0:
filtered_orf_types_str = ','.join(args.filtered_orf_types)
msg = "Filtering these ORF types: {}".format(filtered_orf_types_str)
logger.info(msg)
m_orf_types = bayes_factors['orf_type'].isin(args.filtered_orf_types)
bayes_factors = bayes_factors[~m_orf_types]
msg = "Identifying ORFs which meet the prediction thresholds"
logger.info(msg)
all_orfs, predicted_orfs = ribo_utils.get_predicted_orfs(
bayes_factors,
min_signal=args.min_profile,
min_length=args.min_length,
min_bf_mean=args.min_bf_mean,
max_bf_var=args.max_bf_var,
min_bf_likelihood=args.min_bf_likelihood,
chisq_alpha=args.chisq_significance_level,
select_longest_by_stop=args.select_longest_by_stop,
use_chi_square=args.chi_square_only
)
msg = "Number of selected ORFs: {}".format(len(predicted_orfs))
logger.info(msg)
if args.select_best_overlapping:
msg = "Finding overlapping ORFs"
logger.info(msg)
merged_intervals = bed_utils.merge_all_intervals(predicted_orfs)
msg = "Selecting best among overlapping ORFs"
logger.info(msg)
predicted_orfs = parallel.apply_iter_simple(
merged_intervals['merged_ids'],
get_best_overlapping_orf,
predicted_orfs,
progress_bar=True
)
predicted_orfs = pd.DataFrame(predicted_orfs)
msg = "Sorting selected ORFs"
logger.info(msg)
predicted_orfs = bed_utils.sort(predicted_orfs)
msg = "Writing selected ORFs to disk"
logger.info(msg)
bed_utils.write_bed(predicted_orfs, args.predicted_orfs)
# now get the sequences
msg = "Extracting predicted ORFs DNA sequence"
logger.info(msg)
split_exons = True
transcript_sequences = bed_utils.get_all_bed_sequences(
predicted_orfs,
args.fasta,
split_exons
)
fastx_utils.write_fasta(transcript_sequences,
args.predicted_dna_sequences,
compress=False)
# translate the remaining ORFs into protein sequences
msg = "Converting predicted ORF sequences to amino acids"
logger.info(msg)
records = fastx_utils.get_read_iterator(args.predicted_dna_sequences)
protein_records = {
r[0]: Bio.Seq.translate(r[1]) for r in records
}
fastx_utils.write_fasta(
protein_records.items(),
args.predicted_protein_sequences,
compress=False
)
def main():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description='''Extract the ORFs from the given transcripts and
write as a BED12+ file. Additional fields, 'orf_len' and 'orf_num', give the length
of each ORF and it's index (used to write the ORF profiles). A third additional field
records duplicated ORFs from transcript variants.''')
parser.add_argument('transcripts_bed',
help='''The BED12 file containing the
transcript information.''')
parser.add_argument('transcripts_fasta',
help='''The fasta file containing the
spliced transcript sequences.''')
parser.add_argument('out', help='''The output (BED12+ gz) file.''')
parser.add_argument('--start-codons',
help='''A list of codons which will be treated
as start codons when extracting the ORFs.''',
nargs='+',
default=default_start_codons)
parser.add_argument('--stop-codons',
help='''A list of codons which will be treated
as stop codons when extracting the ORFs.''',
nargs='+',
default=default_stop_codons)
slurm.add_sbatch_options(parser)
logging_utils.add_logging_options(parser)
args = parser.parse_args()
logging_utils.update_logging(args)
# check if we wanted to use slurm
if args.use_slurm:
cmd = ' '.join(sys.argv)
slurm.check_sbatch(cmd, args=args)
return
msg = "Compiling start and stop codon regular expressions"
logger.info(msg)
start_codons_re = '|'.join(args.start_codons)
stop_codons_re = '|'.join(args.stop_codons)
start_codons_re = re.compile(start_codons_re)
stop_codons_re = re.compile(stop_codons_re)
msg = "Reading transcripts bed file"
logger.info(msg)
transcripts_bed = bed_utils.read_bed(args.transcripts_bed)
msg = "Creating the sequence iterator"
logger.info(msg)
transcripts_fasta = fastx_utils.get_read_iterator(args.transcripts_fasta)
transcripts_iter = ((get_transcript(transcript_header,
transcripts_bed), transcript_sequence)
for (transcript_header,
transcript_sequence) in transcripts_fasta)
msg = "Finding all ORFs"
logger.info(msg)
orfs = parallel.apply_parallel_iter(transcripts_iter,
args.num_cpus,
get_orfs,
start_codons_re,
stop_codons_re,
total=len(transcripts_bed),
progress_bar=True)
msg = "Joining ORFs in a large data frame"
logger.info(msg)
orfs = pd.concat(orfs)
orfs.reset_index(drop=True, inplace=True)
# This is done arbitrarily, however we keep all matching
# transcripts for reference
msg = "Marking and removing duplicate ORFs"
logger.info(msg)
groupby_duplicates = orfs.groupby(DUPLICATE_FIELDS,
as_index=False).agg({'id': ','.join})
orfs = orfs.merge(groupby_duplicates, how='left', on=DUPLICATE_FIELDS)
orfs.drop_duplicates(subset=DUPLICATE_FIELDS, inplace=True, keep='first')
orfs.rename(columns={'id_x': 'id', 'id_y': 'duplicates'}, inplace=True)
msg = "Numbering remaining ORFs"
logger.info(msg)
orfs['orf_num'] = np.arange(len(orfs))
msg = "Writing ORFs to disk"
logger.info(msg)
bed_utils.write_bed(orfs, args.out)
def main():
global profiles_data, profiles_indices, profiles_indptr, profiles_shape
global translated_models, untranslated_models
global args
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
description="""This script uses Hamiltonian MCMC with Stan
to estimate translation parameters for a set of regions (presumably ORFs). Roughly, it takes
as input: (1) a set of regions (ORFs) and their corresponding profiles
(2) a "translated" model which gives the probability that a region is translated
(3) an "untranslated" model which gives the probability that a region is not translated.
The script first smoothes the profiles using LOWESS. It then calculates both the Bayes' factor
(using the smoothed profile) and chi2 value (using the raw counts) for each ORF."""
)
parser.add_argument('profiles', help="The ORF profiles (counts) (mtx)")
parser.add_argument(
'regions',
help="The regions (ORFs) for which predictions will be made (BED12+)")
parser.add_argument('out',
help="The output file for the Bayes' factors (BED12+)")
parser.add_argument('--chi-square-only',
help="""If this flag is present, then only the chi
square test will be performed for each ORF. This can also be a way to get the counts within
each of the ORFs.""",
action='store_true')
parser.add_argument('--translated-models',
help="The models to use as H_t (pkl)",
nargs='+')
parser.add_argument('--untranslated-models',
help="The models to use as H_u (pkl)",
nargs='+')
# filtering options
parser.add_argument(
'--orf-types',
help=
"If values are given, then only orfs with those types are processed.",
nargs='*',
default=translation_options['orf_types'])
parser.add_argument('--orf-type-field', default=default_orf_type_field)
parser.add_argument(
'--min-length',
help="ORFs with length less than this value will not be processed",
type=int,
default=translation_options['orf_min_length_pre'])
parser.add_argument(
'--max-length',
help="ORFs with length greater than this value will not be processed",
type=int,
default=translation_options['orf_max_length_pre'])
parser.add_argument(
'--min-profile',
help="""ORFs with profile sum (i.e., number of reads) less than this
value will not be processed.""",
type=float,
default=translation_options['orf_min_profile_count_pre'])
# smoothing options
parser.add_argument('--fraction',
help="The fraction of signal to use in LOWESS",
type=float,
default=translation_options['smoothing_fraction'])
parser.add_argument(
'--reweighting-iterations',
help="The number of reweighting "
"iterations to use in LOWESS. "
"Please see the statsmodels documentation for a "
"detailed description of this parameter.",
type=int,
default=translation_options['smoothing_reweighting_iterations'])
# MCMC options
parser.add_argument('-s',
'--seed',
help="The random seeds to use for inference",
type=int,
default=translation_options['seed'])
parser.add_argument('-c',
'--chains',
help="The number of MCMC chains to use",
type=int,
default=translation_options['chains'])
parser.add_argument(
'-i',
'--iterations',
help="The number of MCMC iterations to use for each chain",
type=int,
default=translation_options['translation_iterations'])
# behavior options
parser.add_argument(
'--num-orfs',
help="If n>0, then only this many ORFs will be processed",
type=int,
default=0)
parser.add_argument('--orf-num-field', default=default_orf_num_field)
parser.add_argument('--do-not-compress',
help="Unless otherwise specified, the output will "
"be written in GZip format",
action='store_true')
parser.add_argument('-g',
'--num-groups',
help="The number of groups into which to split "
"the ORFs. More groups means the progress bar is "
"updated more frequently but incurs more overhead "
"because of the parallel calls.",
type=int,
default=default_num_groups)
slurm.add_sbatch_options(parser)
logging_utils.add_logging_options(parser)
args = parser.parse_args()
logging_utils.update_logging(args)
if args.use_slurm:
cmd = ' '.join(sys.argv)
slurm.check_sbatch(cmd, args=args)
return
# read in the regions and apply the filters
msg = "Reading and filtering ORFs"
logger.info(msg)
regions = bed_utils.read_bed(args.regions)
# by default, keep everything
m_filters = np.array([True] * len(regions))
if len(args.orf_types) > 0:
m_orf_type = regions[args.orf_type_field].isin(args.orf_types)
m_filters = m_orf_type & m_filters
# min length
if args.min_length > 0:
m_min_length = regions['orf_len'] >= args.min_length
m_filters = m_min_length & m_filters
# max length
if args.max_length > 0:
m_max_length = regions['orf_len'] <= args.max_length
m_filters = m_max_length & m_filters
# min profile
profiles = scipy.io.mmread(args.profiles).tocsr()
profiles_sums = profiles.sum(axis=1)
good_orf_nums = np.where(profiles_sums >= args.min_profile)
good_orf_nums = set(good_orf_nums[0])
m_profile = regions['orf_num'].isin(good_orf_nums)
m_filters = m_profile & m_filters
regions = regions[m_filters]
if args.num_orfs > 0:
regions = regions.head(args.num_orfs)
regions = regions.reset_index(drop=True)
msg = "Number of regions after filtering: {}".format(len(regions))
logger.info(msg)
logger.debug("Reading models")
translated_models = [
pickle.load(open(tm, 'rb')) for tm in args.translated_models
]
untranslated_models = [
pickle.load(open(bm, 'rb')) for bm in args.untranslated_models
]
profiles_data = multiprocessing.RawArray(ctypes.c_double,
profiles.data.flat)
profiles_indices = multiprocessing.RawArray(ctypes.c_int, profiles.indices)
profiles_indptr = multiprocessing.RawArray(ctypes.c_int, profiles.indptr)
profiles_shape = multiprocessing.RawArray(ctypes.c_int, profiles.shape)
with suppress_stdout_stderr():
bfs_l = parallel.apply_parallel_split(regions,
args.num_cpus,
get_all_bayes_factors_args,
num_groups=args.num_groups,
progress_bar=True,
backend='multiprocessing')
bfs = pd.concat(bfs_l)
# write the results as a bed12+ file
bed_utils.write_bed(bfs, args.out)