def prune_transcript_graph(G, strand, transcript_map,
min_trim_length=0,
trim_utr_fraction=0.0,
trim_intron_fraction=0.0):
'''
trim_utr_fraction: float specifying the fraction of the average UTR
coverage below which the ends of the UTR will be trimmed
trim_intron_fraction: float specifying the fraction of the average
intron coverage below which intronic nodes will be removed
'''
# trim utrs and intron retentions
trim_nodes = trim_graph(G, strand, min_trim_length,
trim_utr_fraction,
trim_intron_fraction)
G.remove_nodes_from(trim_nodes)
# collapse consecutive nodes in graph
H = collapse_strand_specific_graph(G, transcript_map, introns=True)
# get connected components of graph which represent independent genes
# unconnected components are considered different genes
Gsubs = nx.weakly_connected_component_subgraphs(H)
for Gsub in Gsubs:
# get partial path data supporting graph
transcript_node_map = get_transcript_node_map(Gsub)
path_score_dict = collections.defaultdict(lambda: 0)
for t_id, nodes in transcript_node_map.iteritems():
# reverse path for negative strand transcripts
if strand == NEG_STRAND:
nodes.reverse()
# get transcript scores
t = transcript_map[t_id]
path_score_dict[tuple(nodes)] += t.score
yield Gsub, strand, path_score_dict.items()
def prune_transcript_graph(G,
strand,
transcript_map,
min_trim_length=0,
trim_utr_fraction=0.0,
trim_intron_fraction=0.0):
'''
trim_utr_fraction: float specifying the fraction of the average UTR
coverage below which the ends of the UTR will be trimmed
trim_intron_fraction: float specifying the fraction of the average
intron coverage below which intronic nodes will be removed
'''
# trim utrs and intron retentions
trim_nodes = trim_graph(G, strand, min_trim_length, trim_utr_fraction,
trim_intron_fraction)
G.remove_nodes_from(trim_nodes)
# collapse consecutive nodes in graph
H = collapse_strand_specific_graph(G, transcript_map, introns=True)
# get connected components of graph which represent independent genes
# unconnected components are considered different genes
Gsubs = nx.weakly_connected_component_subgraphs(H)
for Gsub in Gsubs:
# get partial path data supporting graph
transcript_node_map = get_transcript_node_map(Gsub)
path_score_dict = collections.defaultdict(lambda: 0)
for t_id, nodes in transcript_node_map.iteritems():
# reverse path for negative strand transcripts
if strand == NEG_STRAND:
nodes.reverse()
# get transcript scores
t = transcript_map[t_id]
path_score_dict[tuple(nodes)] += t.score
yield Gsub, strand, path_score_dict.items()
def create_transcript_graphs(chrom, transcripts,
min_trim_length=0,
trim_utr_fraction=0.0,
trim_intron_fraction=0.0,
create_bedgraph=False,
bedgraph_filehs=None):
'''
generates (graph, strand, transcript_map) tuples with transcript
graphs
'''
def get_bedgraph_lines(chrom, G):
for n in sorted(G.nodes()):
if n.start < 0:
continue
fields = (chrom, n.start, n.end, G.node[n][NODE_SCORE])
yield fields
# partition transcripts by strand and resolve unstranded transcripts
logging.debug("\tResolving unstranded transcripts")
strand_transcript_lists, strand_ref_transcripts = \
partition_transcripts_by_strand(transcripts)
# create strand-specific graphs using redistributed score
logging.debug("\tCreating transcript graphs")
transcript_graphs = []
for strand, transcript_list in enumerate(strand_transcript_lists):
# create strand specific transcript graph
G = create_directed_graph(strand, transcript_list)
# output bedgraph
if create_bedgraph:
for fields in get_bedgraph_lines(chrom, G):
print >>bedgraph_filehs[strand], '\t'.join(map(str,fields))
# trim utrs and intron retentions
trim_nodes = trim_graph(G, strand,
min_trim_length,
trim_utr_fraction,
trim_intron_fraction)
G.remove_nodes_from(trim_nodes)
# collapse consecutive nodes in graph
H, node_chain_map = collapse_strand_specific_graph(G, introns=True)
# get connected components of graph which represent independent genes
# unconnected components are considered different genes
Gsubs = nx.weakly_connected_component_subgraphs(H)
# add components as separate transcript graphs
strand_graphs = []
node_subgraph_map = {}
for i,Gsub in enumerate(Gsubs):
for n in Gsub:
node_subgraph_map[n] = i
tg = TranscriptGraph(chrom, strand, Gsub)
tg.partial_paths = collections.defaultdict(lambda: 0.0)
strand_graphs.append(tg)
# populate transcript graphs with partial paths
for t in transcript_list:
# get original transcript nodes and subtract trimmed nodes
# convert to collapsed nodes and bin according to subgraph
# TODO: intronic transcripts may be split into multiple pieces,
# should we allow this?
subgraph_node_map = collections.defaultdict(lambda: set())
for n in split_exons(t, G.graph['boundaries']):
n = Exon(*n)
if n in trim_nodes:
continue
cn = node_chain_map[n]
subgraph_id = node_subgraph_map[cn]
subgraph_node_map[subgraph_id].add(cn)
# add transcript node/score pairs to subgraphs
for subgraph_id, subgraph_nodes in subgraph_node_map.iteritems():
subgraph_nodes = sorted(subgraph_nodes,
key=operator.attrgetter('start'),
reverse=(strand == NEG_STRAND))
tg = strand_graphs[subgraph_id]
tg.partial_paths[tuple(subgraph_nodes)] += t.score
transcript_graphs.extend(strand_graphs)
# convert
for tg in transcript_graphs:
tg.partial_paths = tg.partial_paths.items()
return transcript_graphs
def create_transcript_graphs(chrom,
transcripts,
min_trim_length=0,
trim_utr_fraction=0.0,
trim_intron_fraction=0.0,
create_bedgraph=False,
bedgraph_filehs=None):
'''
generates (graph, strand, transcript_map) tuples with transcript
graphs
'''
def get_bedgraph_lines(chrom, G):
for n in sorted(G.nodes()):
if n.start < 0:
continue
fields = (chrom, n.start, n.end, G.node[n][NODE_SCORE])
yield fields
# partition transcripts by strand and resolve unstranded transcripts
logging.debug("\tResolving unstranded transcripts")
strand_transcript_lists, strand_ref_transcripts = \
partition_transcripts_by_strand(transcripts)
# create strand-specific graphs using redistributed score
logging.debug("\tCreating transcript graphs")
transcript_graphs = []
for strand, transcript_list in enumerate(strand_transcript_lists):
# create strand specific transcript graph
G = create_directed_graph(strand, transcript_list)
# output bedgraph
if create_bedgraph:
for fields in get_bedgraph_lines(chrom, G):
print >> bedgraph_filehs[strand], '\t'.join(map(str, fields))
# trim utrs and intron retentions
trim_nodes = trim_graph(G, strand, min_trim_length, trim_utr_fraction,
trim_intron_fraction)
G.remove_nodes_from(trim_nodes)
# collapse consecutive nodes in graph
H, node_chain_map = collapse_strand_specific_graph(G, introns=True)
# get connected components of graph which represent independent genes
# unconnected components are considered different genes
Gsubs = nx.weakly_connected_component_subgraphs(H)
# add components as separate transcript graphs
strand_graphs = []
node_subgraph_map = {}
for i, Gsub in enumerate(Gsubs):
for n in Gsub:
node_subgraph_map[n] = i
tg = TranscriptGraph(chrom, strand, Gsub)
tg.partial_paths = collections.defaultdict(lambda: 0.0)
strand_graphs.append(tg)
# populate transcript graphs with partial paths
for t in transcript_list:
# get original transcript nodes and subtract trimmed nodes
# convert to collapsed nodes and bin according to subgraph
# TODO: intronic transcripts may be split into multiple pieces,
# should we allow this?
subgraph_node_map = collections.defaultdict(lambda: set())
for n in split_exons(t, G.graph['boundaries']):
n = Exon(*n)
if n in trim_nodes:
continue
cn = node_chain_map[n]
subgraph_id = node_subgraph_map[cn]
subgraph_node_map[subgraph_id].add(cn)
# add transcript node/score pairs to subgraphs
for subgraph_id, subgraph_nodes in subgraph_node_map.iteritems():
subgraph_nodes = sorted(subgraph_nodes,
key=operator.attrgetter('start'),
reverse=(strand == NEG_STRAND))
tg = strand_graphs[subgraph_id]
tg.partial_paths[tuple(subgraph_nodes)] += t.score
transcript_graphs.extend(strand_graphs)
# convert
for tg in transcript_graphs:
tg.partial_paths = tg.partial_paths.items()
return transcript_graphs