def assemble_gene(sgraph, locus_id_str, config):
genome_id_str = ('%s:%d-%d[%s]' %
(sgraph.chrom, sgraph.start, sgraph.end,
Strand.to_gtf(sgraph.strand)))
logging.debug('%s locus: %s nodes: %d' %
(genome_id_str, locus_id_str, len(sgraph.G)))
# output splice graph node data
for f in sgraph.get_node_gtf():
print >>config.splice_graph_gtf_fh, str(f)
if config.change_point:
# detect change points
changepts = sgraph.detect_change_points(
pval=config.change_point_pvalue,
fc_cutoff=config.change_point_fold_change)
logging.debug('%s locus %s change points: %d' %
(genome_id_str, locus_id_str, len(changepts)))
for cp in changepts:
sgraph.apply_change_point(cp, config.change_point_trim)
# output splice graph change points
for f in sgraph.get_change_point_gtf(cp):
print >>config.splice_graph_gtf_fh, str(f)
# must recreate splice graph after finding change points
if len(changepts) > 0:
sgraph.recreate()
# run isoform path finding algorithm, filter and group into genes
for gene_isoforms in assemble_isoforms(sgraph, config):
# assign gene_id and tss_id
assign_ids(gene_isoforms, sgraph.strand, config.gene_id_iter,
config.tss_id_iter)
# write output
for isoform in gene_isoforms:
# assign transcript id
t_id = config.t_id_iter.next()
# get strings for each id
t_id_str = "TU%d" % t_id
tss_id_str = "TSS%d" % (isoform.tss_id)
gene_id_str = "G%d" % (isoform.gene_id)
# write to GTF
for f in get_gtf_features(chrom=sgraph.chrom,
strand=sgraph.strand,
exons=isoform.path,
locus_id=locus_id_str,
gene_id=gene_id_str,
tss_id=tss_id_str,
transcript_id=t_id_str,
expr=isoform.expr,
rel_frac=isoform.rel_frac,
abs_frac=isoform.abs_frac):
print >>config.assembly_gtf_fh, str(f)
# write to BED
name = "%s|%s(%.1f)" % (gene_id_str, t_id_str, isoform.expr)
fields = write_bed(sgraph.chrom, name, sgraph.strand,
int(round(1000.0 * isoform.rel_frac)),
isoform.path)
print >>config.assembly_bed_fh, '\t'.join(fields)
def assemble_gene(sgraph, locus_id_str, config):
genome_id_str = (
'%s:%d-%d[%s]' %
(sgraph.chrom, sgraph.start, sgraph.end, Strand.to_gtf(sgraph.strand)))
logging.debug('%s locus: %s nodes: %d' %
(genome_id_str, locus_id_str, len(sgraph.G)))
# output splice graph node data
for f in sgraph.get_node_gtf():
print >> config.splice_graph_gtf_fh, str(f)
if config.change_point:
# detect change points
changepts = sgraph.detect_change_points(
pval=config.change_point_pvalue,
fc_cutoff=config.change_point_fold_change)
logging.debug('%s locus %s change points: %d' %
(genome_id_str, locus_id_str, len(changepts)))
for cp in changepts:
sgraph.apply_change_point(cp, config.change_point_trim)
# output splice graph change points
for f in sgraph.get_change_point_gtf(cp):
print >> config.splice_graph_gtf_fh, str(f)
# must recreate splice graph after finding change points
if len(changepts) > 0:
sgraph.recreate()
# run isoform path finding algorithm, filter and group into genes
for gene_isoforms in assemble_isoforms(sgraph, config):
# assign gene_id and tss_id
assign_ids(gene_isoforms, sgraph.strand, config.gene_id_iter,
config.tss_id_iter)
# write output
for isoform in gene_isoforms:
# assign transcript id
t_id = config.t_id_iter.next()
# get strings for each id
t_id_str = "TU%d" % t_id
tss_id_str = "TSS%d" % (isoform.tss_id)
gene_id_str = "G%d" % (isoform.gene_id)
# write to GTF
for f in get_gtf_features(chrom=sgraph.chrom,
strand=sgraph.strand,
exons=isoform.path,
locus_id=locus_id_str,
gene_id=gene_id_str,
tss_id=tss_id_str,
transcript_id=t_id_str,
expr=isoform.expr,
rel_frac=isoform.rel_frac,
abs_frac=isoform.abs_frac):
print >> config.assembly_gtf_fh, str(f)
# write to BED
name = "%s|%s(%.1f)" % (gene_id_str, t_id_str, isoform.expr)
fields = write_bed(sgraph.chrom, name, sgraph.strand,
int(round(1000.0 * isoform.rel_frac)),
isoform.path)
print >> config.assembly_bed_fh, '\t'.join(fields)
def test_multi_strand2():
t_dict, locus = read_single_locus("multi_strand2.gtf")
transfrags_pos = locus.get_transfrags(Strand.POS)
sgpos = SpliceGraph.create(transfrags_pos)
sgdict = {}
for sg in sgpos.split():
k = "%s:%d-%d[%s]" % (sg.chrom, sg.start, sg.end, Strand.to_gtf(sg.strand))
sgdict[k] = sg
assert "chr1:100-300[+]" in sgdict
assert "chr1:400-600[+]" in sgdict
def test_multi_strand2():
t_dict, locus = read_single_locus('multi_strand2.gtf')
transfrags_pos = locus.get_transfrags(Strand.POS)
sgpos = SpliceGraph.create(transfrags_pos)
sgdict = {}
for sg in sgpos.split():
k = ('%s:%d-%d[%s]' % (sg.chrom, sg.start, sg.end,
Strand.to_gtf(sg.strand)))
sgdict[k] = sg
assert 'chr1:100-300[+]' in sgdict
assert 'chr1:400-600[+]' in sgdict
def get_gtf_features(chrom, strand, exons, locus_id, gene_id, tss_id,
transcript_id, expr, rel_frac, abs_frac):
tx_start = exons[0].start
tx_end = exons[-1].end
strand_str = Strand.to_gtf(strand)
attr_dict = {
'locus_id': locus_id,
'gene_id': gene_id,
'tss_id': tss_id,
'transcript_id': transcript_id
}
f = GTF.Feature()
f.seqid = chrom
f.source = 'taco'
f.feature = 'transcript'
f.start = tx_start
f.end = tx_end
f.score = int(round(1000.0 * rel_frac))
f.strand = strand_str
f.phase = '.'
f.attrs = {
'expr': '%.3f' % expr,
'rel_frac': '%.5f' % rel_frac,
'abs_frac': '%.5f' % abs_frac
}
f.attrs.update(attr_dict)
yield f
for e in exons:
f = GTF.Feature()
f.seqid = chrom
f.source = 'taco'
f.feature = 'exon'
f.start = e.start
f.end = e.end
f.score = int(round(1000.0 * rel_frac))
f.strand = strand_str
f.phase = '.'
f.attrs = {}
f.attrs.update(attr_dict)
yield f
def assemble_isoforms(sgraph, config):
# read in transfrag paths
pgf = PathGraphFactory(sgraph)
K, k = pgf.create_optimal(kmax=config.path_graph_kmax,
loss_threshold=config.path_graph_loss_threshold,
stats_fh=config.path_graph_stats_fh)
if K is None or len(K) == 0:
return []
# smooth kmer graph
K.apply_smoothing()
genome_id_str = (
'%s:%d-%d[%s]' %
(sgraph.chrom, sgraph.start, sgraph.end, Strand.to_gtf(sgraph.strand)))
logging.debug('%s finding isoforms in k=%d graph (%d kmers) '
'source_expr=%f' %
(genome_id_str, k, len(K), K.exprs[K.SOURCE_ID]))
paths = []
for kmer_path, expr in find_paths(K, config.path_frac, config.max_paths):
path = reconstruct_path(kmer_path, K, sgraph)
paths.append((path, expr))
logging.debug('%s isoforms: %d' % (genome_id_str, len(paths)))
# build gene clusters
clusters, filtered = Cluster.build(paths, min_frac=config.isoform_frac)
logging.debug('%s gene clusters: %d filtered transfrags: %d' %
(genome_id_str, len(clusters), len(filtered)))
gene_isoforms = []
for cluster in clusters:
isoforms = []
for path, expr, rel_frac, abs_frac in cluster.iterpaths():
isoforms.append(
Isoform(path=path,
expr=expr,
rel_frac=rel_frac,
abs_frac=abs_frac))
# apply max isoforms limit (per cluster)
if config.max_isoforms > 0:
isoforms = isoforms[:config.max_isoforms]
gene_isoforms.append(isoforms)
return gene_isoforms
def assemble_isoforms(sgraph, config):
# read in transfrag paths
pgf = PathGraphFactory(sgraph)
K, k = pgf.create_optimal(kmax=config.path_graph_kmax,
loss_threshold=config.path_graph_loss_threshold,
stats_fh=config.path_graph_stats_fh)
if K is None or len(K) == 0:
return []
# smooth kmer graph
K.apply_smoothing()
genome_id_str = ('%s:%d-%d[%s]' %
(sgraph.chrom, sgraph.start, sgraph.end,
Strand.to_gtf(sgraph.strand)))
logging.debug('%s finding isoforms in k=%d graph (%d kmers) '
'source_expr=%f' %
(genome_id_str, k, len(K), K.exprs[K.SOURCE_ID]))
paths = []
for kmer_path, expr in find_paths(K, config.path_frac, config.max_paths):
path = reconstruct_path(kmer_path, K, sgraph)
paths.append((path, expr))
logging.debug('%s isoforms: %d' % (genome_id_str, len(paths)))
# build gene clusters
clusters, filtered = Cluster.build(paths, min_frac=config.isoform_frac)
logging.debug('%s gene clusters: %d filtered transfrags: %d' %
(genome_id_str, len(clusters), len(filtered)))
gene_isoforms = []
for cluster in clusters:
isoforms = []
for path, expr, rel_frac, abs_frac in cluster.iterpaths():
isoforms.append(Isoform(path=path, expr=expr, rel_frac=rel_frac,
abs_frac=abs_frac))
# apply max isoforms limit (per cluster)
if config.max_isoforms > 0:
isoforms = isoforms[:config.max_isoforms]
gene_isoforms.append(isoforms)
return gene_isoforms
def write_bed(chrom, name, strand, score, exons):
assert all(exons[0].start < x.start for x in exons[1:])
assert all(exons[-1].end > x.end for x in exons[:-1])
tx_start = exons[0].start
tx_end = exons[-1].end
block_sizes = []
block_starts = []
for e in exons:
block_starts.append(e.start - tx_start)
block_sizes.append(e.end - e.start)
# make bed fields
fields = [
chrom,
str(tx_start),
str(tx_end),
str(name),
str(score),
Strand.to_gtf(strand),
str(tx_start),
str(tx_start), '0',
str(len(exons)), ','.join(map(str, block_sizes)) + ',',
','.join(map(str, block_starts)) + ','
]
return fields
def write_bed(chrom, name, strand, score, exons):
assert all(exons[0].start < x.start for x in exons[1:])
assert all(exons[-1].end > x.end for x in exons[:-1])
tx_start = exons[0].start
tx_end = exons[-1].end
block_sizes = []
block_starts = []
for e in exons:
block_starts.append(e.start - tx_start)
block_sizes.append(e.end - e.start)
# make bed fields
fields = [chrom,
str(tx_start),
str(tx_end),
str(name),
str(score),
Strand.to_gtf(strand),
str(tx_start),
str(tx_start),
'0',
str(len(exons)),
','.join(map(str, block_sizes)) + ',',
','.join(map(str, block_starts)) + ',']
return fields
def get_gtf_features(chrom, strand, exons, locus_id, gene_id, tss_id,
transcript_id, expr, rel_frac, abs_frac):
tx_start = exons[0].start
tx_end = exons[-1].end
strand_str = Strand.to_gtf(strand)
attr_dict = {'locus_id': locus_id,
'gene_id': gene_id,
'tss_id': tss_id,
'transcript_id': transcript_id}
f = GTF.Feature()
f.seqid = chrom
f.source = 'taco'
f.feature = 'transcript'
f.start = tx_start
f.end = tx_end
f.score = int(round(1000.0 * rel_frac))
f.strand = strand_str
f.phase = '.'
f.attrs = {'expr': '%.3f' % expr,
'rel_frac': '%.5f' % rel_frac,
'abs_frac': '%.5f' % abs_frac}
f.attrs.update(attr_dict)
yield f
for e in exons:
f = GTF.Feature()
f.seqid = chrom
f.source = 'taco'
f.feature = 'exon'
f.start = e.start
f.end = e.end
f.score = int(round(1000.0 * rel_frac))
f.strand = strand_str
f.phase = '.'
f.attrs = {}
f.attrs.update(attr_dict)
yield f