def clique_with_peel(seed_i, cursor):
THRESHOLD = CLIQUE_MIN_SIZE * QUASI_GAMMA - 1
G = seed_graph(seed_i, cursor)
# first time we're lenient
G.delete_nodes_from( filter(lambda n: G.degree(n)<=THRESHOLD, G.nodes_iter()) )
print >> sys.stderr, "initial seed graph gets {0} nodes, {1} edges".format(G.number_of_nodes(),G.number_of_edges())
G_nodes = G.nodes()
S,H = p.convert_graph_connectivity_to_sparse(G, G_nodes)
tQ = p.grasp(S, H, QUASI_GAMMA, QUASI_MAXITR, G_nodes.index(seed_i))
Q = map(lambda x: G_nodes[x], tQ)
print >> sys.stderr, "before peel clique size {0}, graph {1} nodes, {2} edges".format(len(Q),\
G.number_of_nodes(), G.number_of_edges())
if seed_i not in Q:
print >> sys.stderr, "seed node got kicked out PART 1. NOOOOO!!!"
return ([],[]) #cHANGE BACK LATER
peel(G, seed_i, Q, QUASI_GAMMA, cursor)
print >> sys.stderr, "after peel seed graph gets {0} nodes, {1} edges".format(G.number_of_nodes(),G.number_of_edges())
# return Q, G # TODO: delete later
G_nodes = G.nodes()
S,H = p.convert_graph_connectivity_to_sparse(G, G_nodes)
print >> sys.stderr, "index of seed i", G_nodes.index(seed_i)
tQ = p.grasp(S, H, QUASI_GAMMA, QUASI_MAXITR, G_nodes.index(seed_i))
# TESTING (TODO: DELETE LATER) for quasi=0.8 followed by quasi=0.6
print >> sys.stderr, "after peel clique size {0}".format(len(tQ))
Q0 = map(lambda x: G_nodes[x], tQ)
p.local_extra(H, tQ, 0.6)
print >> sys.stderr, "after local quasi=0.6 clique size {0}".format(len(tQ))
Q = map(lambda x: G_nodes[x], tQ)
if seed_i not in Q:
print >> sys.stderr, "seed node got kicked out PART 2. NOOOOO!!!"
return ([],[])
# sanity check, delete later (TODO)
if len(Q) <= CLIQUE_MIN_SIZE:
return (Q0,Q) # this line is just for avoiding error at the sanity check, delete later
threshold = 0.6 * len(Q) # CHANGE THIS BACK LATER #threshold = QUASI_GAMMA * len(Q)
for q in Q:
assert sum( map(lambda x: G.has_edge(q,x), Q) ) >= threshold
return (Q0, Q) # CHANGE BACK TO return Q later
def init_cluster_by_clique(fasta_filename, qver_get_func, bestn=100, ece_penalty=1, ece_min_len=20, nproc=8, maxScore=-1000):
"""
fasta_filename --- initial fasta filename, probably called aloha.fa_split00.fa
qver_get_func --- function that returns QVs on reads
bestn --- parameter in BLASR, higher helps in finding perfect cliques but bigger output
nproc, maxScore --- parameter in BLASR, set maxScore appropriate to input transcript length
ece_penalty, ece_min_len --- parameter in isoform hit calling
Self-blasr input then iteratively find all mutually exclusive cliques (in decreasing size)
Returns dict of cluster_index --> list of seqids
which is the 'uc' dict that can be used by ICE
"""
out_filename = fasta_filename + '.self.blasr'
if os.path.exists(out_filename):
print >> sys.stderr, "{0} already exists. No need to run BLASR.".format(out_filename)
else:
cmd = "blasr {i} {i} -m 5 -maxLCPLength 15 -nproc {cpu} -maxScore {score} -bestn {n} -nCandidates {n} -out {o}".format(\
i=fasta_filename, n=bestn, o=out_filename, cpu=nproc, score=maxScore)
print >> sys.stderr, cmd
subprocess.check_call(cmd, shell=True)
G = nx.Graph()
for r in iCEC.blasr_against_ref(out_filename, is_FL=True, sID_starts_with_c=False, qver_get_func=qver_get_func, ece_penalty=ece_penalty, ece_min_len=ece_min_len):
if r[0] == r[1]: continue # self hit, ignore
if r[-1] is not None:
print >> sys.stderr, "adding edge {0},{1}".format(r[0], r[1])
G.add_edge(r[0], r[1])
uc = {}
used = []
ind = 0
deg = G.degree().items()
deg.sort(key=lambda x:x[1], reverse=True)
for d in deg:
if d[0] not in G: continue
# just get the immediate neighbors since we're looking for perfect cliques
G_prime = G.subgraph([d[0]] + G.neighbors(d[0]))
G_prime_nodes = G_prime.nodes()
S,H = pClique.convert_graph_connectivity_to_sparse(G_prime, G_prime_nodes)
seed_i = G_prime_nodes.index(d[0])
tQ = pClique.grasp(S, H, 1., 5, seed_i)
if len(tQ) > 0:
c = [G_prime_nodes[i] for i in tQ]
uc[ind] = c
ind += 1
used += c
G.remove_nodes_from(c)
# ------ below, old way of using find_cliques, inefficient on large graphs
# cliques = list(nx.find_cliques(G))
# cliques.sort(key=lambda x: len(x), reverse=True)
# for c in cliques:
# if all(map(lambda x: x in G, c)):
# uc[ind] = c
# ind += 1
# used += c
# G.remove_nodes_from(c)
for r in SeqIO.parse(open(fasta_filename), 'fasta'):
if r.id not in used:
uc[ind] = [r.id]
ind += 1
return uc
def new_cluster_pipe(rfam_fam, shuffle_ratio):
assert type(shuffle_ratio) is int
output_prefix = "Rfam_{fam}_shuffle{X}X".format(fam=rfam_fam, X=shuffle_ratio)
fasta_filename = output_prefix+'.fna'
blast_output = "{input}.M8N7Q16R2W3E2.WUblast".format(input=fasta_filename)
report_f = open(output_prefix+'.report', 'w')
if not os.path.exists(blast_output):
dummy_id = 0
nodes_to_index = {}
with open(fasta_filename, 'w') as f:
with get_conn_ncRNA() as cursor:
cursor.execute("select id,seq from Rfam_fasta where rfam_fam='{fam}' order by id".format(fam=rfam_fam))
for _id,seq in cursor.fetchall():
id = "TP{0}_{1}".format(dummy_id, _id)
f.write(">{id}\n{seq}\n".format(id=id, seq=seq))
nodes_to_index[id] = dummy_id
dummy_id += 1
ushuffle.shuffle(seq, len(seq), 2)
for x in xrange(shuffle_ratio):
id = "FP{0}_{1}".format(dummy_id, _id)
f.write(">{id}\n{seq}\n".format(id=id, seq=ushuffle.shuffle2()))
nodes_to_index[id] = dummy_id
dummy_id += 1
start_t = time.time()
# now blast it
os.system("xdformat -n -o {input} {input}".format(input=fasta_filename))
os.system("blastn -d {input} -i {input} -M 8 -N -7 -Q 16 -R 2 -E 2 \
-W 3 -mformat 2 -cpus 4 -o {output}".format(input=fasta_filename, output=blast_output))
report_f.write("(1) BLAST TIME: {0} sec\n".format(time.time()-start_t))
# now parse the blast
nodes_to_index = c1.NodesToIndex(nodes_to_index, -1)
G = Graph()
c1.step1_process_blast(blast_output=blast_output,\
score_cutoff=35, nodes_to_index=nodes_to_index, G=G, program='WU')
print >> sys.stderr, "Homology graph has {0} nodes, {1} edges....".format(\
G.number_of_nodes(), G.number_of_edges())
c1.export_to_db(G, nodes_to_index, 0, blast_output)
# convert nodes_to_index into dict nodes_ind --> acc id
nodes_to_index = dict( map(lambda (x,y):(y,x), nodes_to_index.d.items()) )
with open(blast_output+'.nodes_to_index', 'w') as handle:
for ind,id in nodes_to_index.iteritems():
handle.write("{0}\t{1}\n".format(ind,id))
# read back the .parsed and .sets_for_nodes files
G = Graph()
sets_for_nodes = {}
nodes_to_index = {}
with open(blast_output+'.parsed') as handle:
for line in handle:
raw = map(int, line.strip().split('\t'))
G.add_edge(raw[0],raw[1])
with open(blast_output+'.sets_for_nodes') as handle:
for line in handle:
raw = map(int, line.strip().split('\t'))
sets_for_nodes[raw[0]] = {'nodes_ind':raw[1],'start':raw[2],'end':raw[3]}
with open(blast_output+'.nodes_to_index') as handle:
for line in handle:
raw = line.strip().split()
nodes_to_index[int(raw[0])] = raw[1]
tmp = len(filter(lambda x: nodes_to_index[sets_for_nodes[x]['nodes_ind']].startswith('FP'), G.nodes_iter()))
report_f.write("(2) AFTER parsing BLAST, graph has {0} negative control nodes, {1} TP nodes\n".format(tmp, G.number_of_nodes()-tmp))
# remove low deg (< 3) nodes
x = filter(lambda n: G.degree(n)<NEW_MIN_CLIQUE_SIZE, G.nodes_iter())
while len(x) > 0:
G.delete_nodes_from(x)
x = filter(lambda n: G.degree(n)<NEW_MIN_CLIQUE_SIZE, G.nodes_iter())
tmp = len(filter(lambda x: nodes_to_index[sets_for_nodes[x]['nodes_ind']].startswith('FP'), G.nodes_iter()))
report_f.write("(3) AFTER recursively removing nodes of degree < 3, graph has {0} negative control nodes, {1} TP nodes\n".format(tmp, G.number_of_nodes()-tmp))
report_f.write("----------------------------------------------------------------------------\n")
report_f.write("OUT\tDIR\tCLIQUE_SIZE\tSCANNED_TP\tSCANNED_FP\tCM_time\n")
report_f.write("----------------------------------------------------------------------------\n")
# for now just brute force....go through node by node as seeds
dummy_round = 0
while G.number_of_nodes()>=NEW_MIN_CLIQUE_SIZE and G.number_of_edges()>=NEW_MIN_CLIQUE_SIZE:
# find perfect max cliques with a random starting node
G_nodes = G.nodes()
S,H = p.convert_graph_connectivity_to_sparse(G, G_nodes)
tQ = p.grasp(S, H, gamma=1.0, maxitr=20, given_starting_node=None)
Q = map(lambda x: G_nodes[x], tQ)
if len(Q) < NEW_MIN_CLIQUE_SIZE: # delete these nodes
G.delete_nodes_from(Q)
continue
# PERFECT CLIQUE SANITY TESTING, DELETE LATER
for x in Q:
for y in Q:
if x!=y:
print >> sys.stderr, "testing....", x,y
try:
assert G.has_edge(x,y)
except:
return Q,G
Q.sort()
print >> sys.stderr, "clique is...", Q
prefix = output_prefix + str(dummy_round) + '_size' + str(len(Q)) + '_'
dummy_round += 1
start_t = time.time()
scan_dir,scan_result = run_cmfinder(Q, nodes_to_index, sets_for_nodes, prefix, os.path.abspath(fasta_filename))
cm_time = time.time()-start_t
if scan_result is not None:
outf = open(os.path.basename(scan_dir)+'.gv','w')
outf.write("""graph test{
edge [ dir=none ];
node [ style=filled, fontsize=2.0, height=0.1, width=0.1, fixedsize=true ];
""")
# draw this graph
scanned = {'TP':0, 'FP':0}
for n in G.nodes_iter():
id = nodes_to_index[sets_for_nodes[n]['nodes_ind']] # id is something like TP1_NC_XXXX.... or FP10_NC_XXXX....
_id = id[:id.find('_')]
shape = 'circle' if id.startswith('TP') else 'box'
if n in Q:
outf.write("{0} [color=dodgerblue1, shape={1}];\n".format(n, shape))
elif id in scan_result:
outf.write("{0} [color=darkorange, shape={1}];\n".format(n, shape))
scanned[_id[:2]] += 1
else:
outf.write("{0} [color=grey, shape={1}];\n".format(n, shape))
for (n1,n2) in G.edges_iter(data=False):
id1 = nodes_to_index[sets_for_nodes[n1]['nodes_ind']]
id1 = id1[:id1.find('_')]
id2 = nodes_to_index[sets_for_nodes[n2]['nodes_ind']]
id2 = id2[:id2.find('_')]
outf.write("{0} -- {1};\n".format(n1,n2))
outf.write("}")
outf.close()
report_f.write("{0}\t{1}\t{2}\t{3}\t{4}\t{5}\n".format(outf.name,scan_dir,len(Q),scanned['TP'],scanned['FP'],cm_time))
report_f.flush()
# delete the edges from the graph
G.delete_edges_from(itertools.combinations(Q, 2))
# again, remove low-degree nodes
x = filter(lambda n: G.degree(n)<NEW_MIN_CLIQUE_SIZE, G.nodes_iter())
while len(x) > 0:
G.delete_nodes_from(x)
x = filter(lambda n: G.degree(n)<NEW_MIN_CLIQUE_SIZE, G.nodes_iter())
report_f.close()