def test_concat_2_fail():
hs = khmer.HashSet(5, [10, 12])
hs2 = khmer.HashSet(4, [10, 13])
try:
hs += hs2
assert 0, "inplace concat should fail - different ksize"
except ValueError:
pass
def test_bad_construct():
try:
hs = khmer.HashSet()
assert 0, "HashSet constructor should fail w/o argument"
except TypeError:
pass
try:
hs = khmer.HashSet(5, [{}])
assert 0, "HashSet constructor should fail w/o list of k-mers"
except ValueError:
pass
def test_update_bad():
hs = khmer.HashSet(5)
x = [5, 10, 15, 2**35, {}]
try:
hs.update(x)
assert 0, "cannot add dict to a HashSet"
except ValueError:
pass
def test_remove_2():
hs = khmer.HashSet(5, [8, 10])
assert len(hs) == 2
try:
hs.remove(15)
assert 0, "hs.remove should raise an Exception"
except ValueError:
pass
assert len(hs) == 2
assert list(sorted(hs)) == [8, 10]
def test_traverse_linear_path():
contigfile = utils.get_test_data('simple-genome.fa')
contig = list(screed.open(contigfile))[0].sequence
K = 21
nodegraph = khmer.Nodegraph(K, 1e5, 4)
stopgraph = khmer.Nodegraph(K, 1e5, 4)
nodegraph.consume(contig)
degree_nodes = khmer.HashSet(K)
size, conns, visited = nodegraph.traverse_linear_path(
contig[:K], degree_nodes, stopgraph)
assert size == 980
assert len(conns) == 0
assert len(visited) == 980
def main():
p = argparse.ArgumentParser()
p.add_argument('fastq_files', nargs='+')
args = p.parse_args()
cg = khmer.Countgraph(K, 1e8, 4)
kept = 0
hdn = khmer.HashSet(K)
lh = khmer._GraphLabels(cg)
next_label = 1
next_orf = 1
output = set()
for filename in args.fastq_files:
for n, record in enumerate(screed.open(filename)):
if n and n % 10000 == 0:
print('...', n, file=sys.stderr)
if len(record.sequence) < K:
continue
cov, _, _ = cg.get_median_count(record.sequence)
if cov < 20:
kept += 1
cg.consume(record.sequence)
elif cov < 30:
#print('intermediate', next_label, file=sys.stderr)
seq, pos = cg.trim_on_abundance(record.sequence, 3)
if len(seq) < K:
continue
cg.consume(seq)
hdn = cg.find_high_degree_nodes(seq)
lh.label_across_high_degree_nodes(seq, hdn, next_label)
next_label += 1
elif cov == 30:
contigs = lh.assemble_labeled_path(record.sequence[:K])
for contig in contigs:
for t in translate(contig):
for o in extract_orfs(t):
if hash(o) not in output:
output.add(hash(o))
print('>orf%d\n%s' % (next_orf, o))
next_orf += 1
def main():
p = argparse.ArgumentParser()
p.add_argument('contig_files', nargs='+')
args = p.parse_args()
ng = khmer.Nodegraph(K, 1e8, 4)
starts = []
for filename in args.contig_files:
for n, record in enumerate(screed.open(filename)):
if n and n % 10000 == 0:
print('...', n)
ng.consume(record.sequence)
starts.append(record.sequence[:K])
hdn = khmer.HashSet(K)
for filename in args.contig_files:
for n, record in enumerate(screed.open(filename)):
if n and n % 10000 == 0:
print('...', n)
hdn += ng.find_high_degree_nodes(record.sequence)
lh = khmer._GraphLabels(ng)
for filename in args.contig_files:
for n, record in enumerate(screed.open(filename)):
if n and n % 10000 == 0:
print('...', n)
lh.label_across_high_degree_nodes(record.sequence, hdn, n)
counter = 0
for k in starts:
contigs = lh.assemble_labeled_path(k)
if not contigs:
print('nada...')
for c in contigs:
print('>%d\n%s' % (counter, c))
counter += 1
def test_update():
hs = khmer.HashSet(5)
x = [5, 10, 15, 2**35]
hs.update(x)
assert list(sorted(hs)) == [5, 10, 15, 2**35]
def test_add():
hs = khmer.HashSet(5)
hs.add(7)
hs.add(4)
assert list(sorted(hs)) == [4, 7]
def test_iter_single():
hs = khmer.HashSet(5, [6])
k = iter(hs)
k2 = iter(k)
assert k == k2
def run(args):
# @CTB this is kind of a hack - nothing tricky going on, just want to
# specify memory on the command line rather than graph size...
graph_tablesize = int(args.memory * 8.0 / 4.0)
assert args.ksize % 2, "ksize must be odd"
if args.label:
label_list = []
output_dir = args.output
if not output_dir:
if len(args.seqfiles) > 1:
print('** please specify an output directory with -o',
file=sys.stderr)
sys.exit(-1)
output_dir = os.path.basename(args.seqfiles[0])
if output_dir.endswith('.fa'):
output_dir = output_dir[:-3]
elif output_dir.endswith('.fa.gz'):
output_dir = output_dir[:-6]
# set this so we can read it for logging
args.output = output_dir
# gxtfile = os.path.basename(output_dir) + '.gxt'
gxtfile = os.path.join(output_dir, "cdbg.gxt")
contigfile = os.path.join(output_dir, "contigs.fa.gz")
print('')
print('placing output in directory:', output_dir)
print('gxt will be:', gxtfile)
try:
os.mkdir(output_dir)
except FileExistsError:
print('(note: directory already exists)')
print('')
if args.loadgraph:
print('loading nodegraph from:', args.loadgraph)
graph = khmer.Nodegraph.load(args.loadgraph)
print('creating accompanying stopgraph')
ksize = graph.ksize()
hashsizes = graph.hashsizes()
stop_bf = khmer.Nodegraph(ksize, 1, 1, primes=hashsizes)
else:
print('building graphs and loading files')
# Create graph and a stop bloom filter - one for loading, one for
# traversing. Create them all here so that we can error out quickly
# if memory is a problem.
# @CTB note that hardcoding '2' here is not nec a great idea.
graph = khmer.Nodegraph(args.ksize, graph_tablesize, 2)
stop_bf = khmer.Nodegraph(args.ksize, graph_tablesize, 2)
n = 0
# load in all of the input sequences, one file at a time.
for seqfile in args.seqfiles:
fp = screed.open(seqfile)
for record in khmer.utils.clean_input_reads(fp):
if len(record.cleaned_seq) < graph.ksize(): continue
n += 1
if n % 100000 == 0:
print('...', seqfile, n)
graph.consume(record.cleaned_seq)
fp.close()
# complain if too small set of graphs was used.
fp_rate = khmer.calc_expected_collisions(graph,
args.force,
max_false_pos=.05)
ksize = graph.ksize()
# initialize the object that will track information for us.
pathy = Pathfinder(ksize, gxtfile, contigfile, not args.no_assemble)
print('finding high degree nodes')
if args.label:
print('(and labeling them, per request)')
degree_nodes = khmer.HashSet(ksize)
linear_starts = khmer.HashSet(ksize)
n = 0
skipped = 0
for seqfile in args.seqfiles:
fp = screed.open(seqfile)
for record in khmer.utils.clean_input_reads(fp):
if len(record.cleaned_seq) < ksize:
skipped += 1
continue
n += 1
if n % 100000 == 0:
print('...2', seqfile, n)
# walk across sequences, find all high degree nodes,
# name them and cherish them.
these_hdn = graph.find_high_degree_nodes(record.cleaned_seq)
if these_hdn:
degree_nodes += these_hdn
else:
# possible linear node? check first and last k-mer.
# (the logic here is that every purely linear node must
# start or end in *some* record.sequence - so where we have
# record sequences that have only 1 neighbor, those will be
# all possible linear nodes.
first_kmer = record.sequence[:ksize]
last_kmer = record.sequence[-ksize:]
assert len(last_kmer) == ksize
if len(graph.neighbors(first_kmer)) == 1:
linear_starts.add(graph.hash(first_kmer))
if len(graph.neighbors(last_kmer)) == 1:
linear_starts.add(graph.hash(last_kmer))
if args.label:
label_list.append(record.name)
for kmer in these_hdn:
pathy.add_label(kmer, n)
fp.close()
print('read {}, skipped {} for being too short'.format(n, skipped))
# get all of the degree > 2 kmers and give them IDs.
for kmer in degree_nodes:
pathy.new_hdn(kmer)
stop_bf.add(kmer)
print('traversing linear segments from', len(degree_nodes), 'nodes')
# now traverse from each high degree node into all neighboring nodes,
# seeking adjacencies. if neighbor is high degree node, add it to
# adjacencies; if neighbor is not, then traverse the linear path &
# assemble if desired.
for n, k in enumerate(degree_nodes):
if n % 10000 == 0:
print('...', n, 'of', len(degree_nodes))
# retrieve the node ID of the primary segment.
k_id = pathy.kmers_to_nodes[k]
# here is where we would output this k-mer to the contig file if we
# wanted to.
nk_id = pathy.kmers_to_nodes[k]
k_str = khmer.reverse_hash(k, ksize)
pathy.add_assembly(nk_id, k_str)
# find all the neighbors of this high-degree node.
nbh = graph.neighbors(k)
for nk in nbh:
# neighbor is high degree? fine, mark its adjacencies.
if nk in degree_nodes:
nk_id = pathy.kmers_to_nodes[nk.kmer_u]
pathy.add_adjacency(k_id, nk_id)
else:
# linear! walk it.
traverse_and_mark_linear_paths(graph, nk, stop_bf, pathy,
degree_nodes)
# now, clean up at the end -- make sure we've hit all the possible
# linear nodes.
print('traversing from {} potential linear starts'.format(
len(linear_starts)))
for n, k in enumerate(linear_starts):
traverse_and_mark_linear_paths(graph, k, stop_bf, pathy, degree_nodes)
print('{} linear segments and {} high-degree nodes'.\
format(pathy.node_counter, len(pathy.nodes)))
del graph
del stop_bf
# save to GXT.
print('saving gxtfile', gxtfile)
all_labels = set()
label_counts = {}
pathy.adjfp.close()
adj_fp = open(gxtfile + '.adj', 'rt')
# this uniqifies the edges...
for line in adj_fp:
a, b = line.split(',')
a = int(a)
b = int(b)
pathy.adjacencies[a].add(b)
adj_fp.close()
try:
os.unlink(gxtfile + '.adj')
except:
print('cannot remove', gxtfile + '.adj')
# ...and now print them out.
edges = []
for k, v in pathy.adjacencies.items():
for dest in v:
# don't add loops
if (k != dest):
edges.append((k, dest))
with open(gxtfile, 'wt') as fp:
write(fp, pathy.node_counter, edges)
if not args.no_assemble:
pathy.assemblyfp.close()
if args.label:
print('note: used/assigned %d labels total' % (len(set(all_labels)), ))
print('counts:', label_counts)
assert label_list
print('dumping label list now.')
label_file = os.path.basename(output_dir) + '.labels.txt'
label_file = os.path.join(output_dir, label_file)
with open(label_file, "wt") as fp:
for n, label in enumerate(label_list):
fp.write("{} {}\n".format(n + 0, label))
def test_concat_2():
hs = khmer.HashSet(5, [10, 12])
hs2 = khmer.HashSet(5, [10, 13])
hs += hs2
assert list(sorted(hs)) == [10, 12, 13]
def test_contains_1():
hs = khmer.HashSet(5, [8, 10])
assert 8 in hs
assert 10 in hs
assert 2**35 not in hs
def test_remove():
hs = khmer.HashSet(5, [8, 10])
assert len(hs) == 2
hs.remove(8)
assert len(hs) == 1
assert list(hs) == [10]
def test_iter_single():
hs = khmer.HashSet(5, [6])
for k in hs:
assert k == 6
print(k)
def main():
parser = argparse.ArgumentParser()
parser.add_argument('seqfiles', nargs='+')
parser.add_argument('-o', '--output', default=None)
parser.add_argument('-k', '--ksize', default=DEFAULT_KSIZE, type=int)
parser.add_argument('-x',
'--tablesize',
default=NODEGRAPH_SIZE,
type=float)
parser.add_argument('--force', action='store_true')
args = parser.parse_args()
assert args.ksize % 2, "ksize must be odd"
assert args.output, "you probably want an output file"
print('building graphs and loading files')
# Create graph, and two stop bloom filters - one for loading, one for
# traversing. Create them all here so that we can error out quickly
# if memory is a problem.
graph = khmer.Nodegraph(args.ksize, args.tablesize, 2)
print(graph.ksize(), graph.hashsizes())
stop_bf = khmer.Nodegraph(args.ksize, args.tablesize, 2)
stop_bf2 = khmer.Nodegraph(args.ksize, args.tablesize, 2)
n = 0
# load in all of the input sequences, one file at a time.
for seqfile in args.seqfiles:
for record in screed.open(seqfile):
n += 1
if n % 10000 == 0:
print('...', seqfile, n)
graph.consume(record.sequence)
# complain if too small set of graphs was used.
fp_rate = khmer.calc_expected_collisions(graph,
args.force,
max_false_pos=.05)
# initialize the object that will track information for us.
pathy = Pathfinder(args.ksize)
print('finding high degree nodes')
degree_nodes = khmer.HashSet(args.ksize)
n = 0
for seqfile in args.seqfiles:
for record in screed.open(seqfile):
n += 1
if n % 10000 == 0:
print('...2', seqfile, n)
# walk across sequences, find all high degree nodes,
# name them and cherish them. Don't do this on identical sequences.
if min(stop_bf2.get_kmer_counts(record.sequence)) == 0:
stop_bf2.consume(record.sequence)
degree_nodes += graph.find_high_degree_nodes(record.sequence)
del stop_bf2
if not len(degree_nodes):
print('no high degree nodes; exiting.')
sys.exit(0)
# get all of the degree > 2 nodes and give them IDs.
for node in degree_nodes:
pathy.new_segment(node)
print('traversing linear segments from', len(degree_nodes), 'nodes')
# now traverse from each high degree nodes into all neighboring nodes,
# seeking adjacencies. if neighbor is high degree node, add it to
# adjacencies; if neighbor is not, then traverse the linear path. also
# track minhashes while we're at it.
for n, k in enumerate(degree_nodes):
if n % 10000 == 0:
print('...', n, 'of', len(degree_nodes))
# retrieve the segment ID of the primary node.
k_id = pathy.segments_r[k]
# find all the neighbors of this high-degree node.
nbh = graph.neighbors(k)
for nk in nbh:
# neighbor is high degree? fine, mark its adjacencies.
if nk in degree_nodes:
nk_id = pathy.segments_r[nk]
pathy.add_adjacency(k_id, nk_id)
else:
# linear! walk it.
traverse_and_mark_linear_paths(graph, nk, stop_bf, pathy,
degree_nodes)
print(len(pathy.segments), 'segments, containing',
sum(pathy.segments.values()), 'nodes')
# save to GML
if args.output:
print('saving to', args.output)
fp = open(args.output, 'w')
w = GmlWriter(fp, [], [])
for k, v in pathy.segments.items():
w.add_vertex(k, v, [])
for k, v in pathy.adjacencies.items():
for edge in v:
w.add_edge(k, edge, [])
def test_iter_double():
x = [6, 9, 20]
hs = khmer.HashSet(5, x)
for i, k in enumerate(hs):
assert k == x[i], (k, x[i])
def main():
p = build_counting_args(descr='Streaming assembly with tracking info')
p.add_argument('fastq_files', nargs='+')
p.add_argument('-o',
type=argparse.FileType('w'),
default='assembly-stats.csv')
args = p.parse_args()
cg = create_countgraph(args)
kept = 0
hdn = khmer.HashSet(args.ksize)
lh = khmer._GraphLabels(cg)
next_label = 1
next_orf = 1
output = set()
statswriter = csv.DictWriter(args.o,
delimiter=',',
fieldnames=[
'read_n', 'action', 'cov', 'n_hdn',
'contig_n', 'orf_n', 'new'
])
for filename in args.fastq_files:
for n, record in enumerate(screed.open(filename)):
if n and n % 10000 == 0:
print('...', n, file=sys.stderr)
if len(record.sequence) < args.ksize:
continue
cov, _, _ = cg.get_median_count(record.sequence)
if cov < 20:
kept += 1
cg.consume(record.sequence)
statswriter.writerow({
'read_n': n,
'action': 'c',
'cov': cov,
'n_hdn': None,
'contig_n': None,
'orf_n': None,
'new': None
})
elif cov < 30:
#print('intermediate', next_label, file=sys.stderr)
seq, pos = cg.trim_on_abundance(record.sequence, 3)
if len(seq) < args.ksize:
continue
cg.consume(seq)
hdn = cg.find_high_degree_nodes(seq)
lh.label_across_high_degree_nodes(seq, hdn, next_label)
next_label += 1
statswriter.writerow({
'read_n': n,
'action': 'l',
'cov': cov,
'n_hdn': len(hdn),
'contig_n': None,
'orf_n': None,
'new': None
})
elif cov == 30:
contigs = lh.assemble_labeled_path(
record.sequence[:args.ksize])
for contig_n, contig in enumerate(contigs):
statswriter.writerow({
'read_n': n,
'action': 'a',
'cov': cov,
'n_hdn': None,
'contig_n': contig_n,
'orf_n': None,
'new': None
})
for t in translate(contig):
for orf_n, o in enumerate(extract_orfs(t)):
if hash(o) not in output:
new = True
output.add(hash(o))
print('>orf%d\n%s' % (next_orf, o))
next_orf += 1
else:
new = False
statswriter.writerow({
'read_n': n,
'action': 'a',
'cov': cov,
'n_hdn': None,
'contig_n': contig_n,
'orf_n': orf_n,
'new': new
})
def test_contains_2():
hs = khmer.HashSet(5, [8, 10])
assert khmer.reverse_hash(8, 5) in hs
assert khmer.reverse_hash(10, 5) in hs
assert khmer.reverse_hash(2**35, 5) not in hs
