def get_guild_based_synergy():
# Get network
network = wrappers.get_network(network_file=CONFIG.get("network_file"),
only_lcc=True)
nodes = set(network.nodes())
#create_edge_file(network)
# Get drug info
drug_to_values = get_drug_info(nodes=nodes)
#print drug_to_values.keys()
# Get gexp info
gexp_norm, gene_to_idx, cell_line_to_idx = None, None, None
#gexp_norm, gene_to_idx, cell_line_to_idx = wrappers.get_expression_info(gexp_file = CONFIG.get("gexp_file"), process=set(["z", "abs"]), dump_file = CONFIG.get("gexp_dump"))
# Check individual drugs
guild_drugs(drug_to_values, nodes, gexp_norm, gene_to_idx,
cell_line_to_idx)
# Check pairwise drug combinations
guild_combinations(drug_to_values, nodes, gexp_norm, gene_to_idx,
cell_line_to_idx)
return
# Now using create_feature_file instead of below
# Get synergy info
combination_to_values = get_synergy_info()
# Check synergy between known pairs
combination_to_guild_values = get_guild_based_synergy_scores(
drug_to_values.keys(), None, gexp_norm, gene_to_idx, cell_line_to_idx)
out_file = CONFIG.get("guild_file")
f = open(out_file, 'w')
f.write("comb.id cell.line max.a max.b med mean sd max min syn\n")
for comb_id, cell_line_to_values in combination_to_values.iteritems():
#drug1, drug2 = comb_id.split(".")
for cell_line, vals in cell_line_to_values.iteritems():
values = combination_to_guild_values[comb_id]
max_a, max_b, synergy = vals
f.write("%s %s %f %f %f %f %f %f %f %s\n" %
(comb_id, cell_line, max_a, max_b, numpy.median(values),
numpy.mean(values), numpy.std(values), numpy.max(values),
numpy.min(values), synergy))
f.close()
return
def create_feature_file():
# Get drug info
drug_to_values = get_drug_info(nodes=None)
#drug_to_values = get_drug_info_with_predicted_targets() # worsens the performance
#for drug, values in drug_to_values.iteritems():
# if values[1] != "":
# print drug, values[1]
#print drug_to_values.items()[:3]
# Get cell line info
cell_line_to_value = get_cell_line_info()
#print cell_line_to_value.items()[:3]
# Get synergy info
combination_to_values = get_synergy_info()
cell_line_to_synergy, combination_to_synergy = get_synergy_values_per_cell_line_and_combination(
)
#print combination_to_values.items()[:3]
# Get gexp info
gexp_norm, gene_to_idx, cell_line_to_idx = wrappers.get_expression_info(
gexp_file=CONFIG.get("gexp_file"),
process=set(["z"]),
dump_file=CONFIG.get("gexp_dump")) # process=set(["z", "abs"])
#values = [gene_to_idx["TSPAN6"], gene_to_idx["TNMD"]]
#print gexp_norm[values, cell_line_to_idx["647-V"]]
#print "TNMD @ 647-V", gexp_norm[gene_to_idx["TNMD"], cell_line_to_idx["647-V"]]
# Get methylation info
meth, meth_gene_to_idx, meth_cell_line_to_idx = wrappers.get_expression_info(
gexp_file=CONFIG.get("methylation_file"),
process=set(["z"]),
dump_file=CONFIG.get("methylation_dump"))
#print "A1BG @ 647-V", meth[meth_gene_to_idx["A1BG"], meth_cell_line_to_idx["647-V"]]
# Get mutation info
gene_to_cell_line_to_mutation = get_mutation_info()
#print gene_to_cell_line_to_mutation.items()[:3]
# Get CNV info
#gene_to_cell_line_to_cnv = {}
gene_to_cell_line_to_cnv = get_cnv_info(CONFIG.get("cnv_file"))
#print gene_to_cell_line_to_cnv.items()[:2]
# Get cancer gene & pathway info
pathway_to_genes = get_pathway_info(nodes=None)
#print pathway_to_genes.keys()
genes_pathway = set()
for genes in (pathway_to_genes["kegg"], pathway_to_genes["census"]):
genes_pathway |= genes
genes_pathway = list(genes_pathway)
# Get network
network = wrappers.get_network(network_file=CONFIG.get("network_file"),
only_lcc=True)
network_nodes = set(network.nodes())
# Get GUILD info
#combination_to_guild_values = {}
combination_to_guild_values = get_guild_based_synergy_scores(
drug_to_values.keys(), genes_pathway, gexp_norm, gene_to_idx,
cell_line_to_idx)
#print combination_to_guild_values.items()[:2]
# Get drug similarity info
combination_to_similarity = get_drug_similarity(drug_to_values)
#print combination_to_similarity.items()[:5]
#values = combination_to_similarity.items()
#values.sort(key=lambda x: x[1])
#print values[-20:]
task = CONFIG.get("task")
if task.endswith("-train"):
out_file = CONFIG.get("feature_file_train")
elif task == "ch1-test":
out_file = CONFIG.get("feature_file_test_ch1")
elif task == "ch2-test":
out_file = CONFIG.get("feature_file_test_ch2")
else:
raise ValueError("Uknown task: " + task)
f = open(out_file, 'w')
features = [
"gexpA.med", "gexpA.amed", "gexpB.med", "gexpB.amed", "mutA", "mutB",
"cnvA", "cnvB", "metA.med", "metA.amed", "metB.med", "metB.amed",
"cell.med", "comb.med", "sim.target", "sim.chemical", "kA", "kB",
"dAB", "guild.common", "guild.med", "guild.max", "kegg.inA",
"kegg.inB", "kegg.gexp.med", "kegg.gexp.max", "kegg.mut.med",
"kegg.mut.max", "kegg.cnv.med", "kegg.cnv.max", "kegg.cnvA",
"kegg.cnvB", "cosmic.inA", "cosmic.inB", "cosmic.gexp.med",
"cosmic.gexp.max", "cosmic.mut.med", "cosmic.mut.max",
"cosmic.cnv.med", "cosmic.cnv.max", "cosmic.cnvA", "cosmic.cnvB"
]
drugs = drug_to_values.keys()
seen_combinations = set()
# Get all targets
targets_all = set()
for i, drug1 in enumerate(drugs):
for j, drug2 in enumerate(drugs):
if i >= j:
continue
targets1 = drug_to_values[drug1][0]
targets2 = drug_to_values[drug2][0]
targets_all |= targets1 | targets2
targets_all = list(targets_all)
target_to_idx = dict((target, i) for i, target in enumerate(targets_all))
# Get all pathway genes
#targets_pathway = set()
#for genes in (pathway_to_genes["kegg"], pathway_to_genes["census"]):
# targets_pathway |= genes
#targets_pathway = list(targets_pathway & set(targets_all))
#pathway_target_to_idx = dict((gene, i) for i, gene in enumerate(targets_pathway))
targets_pathway = []
for pathway, genes in pathway_to_genes.iteritems():
targets_pathway.append(pathway)
# Create header
features = map(lambda x: ".g" + x,
targets_all) + map(lambda x: ".m" + x, targets_all) + map(
lambda x: ".c" + x, targets_all) + map(
lambda x: ".z" + x, targets_all) + map(
lambda x: ".e" + x, targets_all) + map(
lambda x: ".p" + x.replace("_", "."),
targets_pathway) + features
f.write("comb.id cell.line %s\n" % " ".join(features))
for i, drug1 in enumerate(drugs):
for j, drug2 in enumerate(drugs):
if i >= j:
continue
#comb_id = ".".join(sorted([drug1, drug2]))
if drug1.lower() < drug2.lower():
comb_id = "%s.%s" % (drug1, drug2)
else:
comb_id = "%s.%s" % (drug2, drug1)
if comb_id not in combination_to_values:
continue
if task != "ch2-test": # task.startswith("ch1-") or task.endswith("-train"):
cell_line_to_mono_values = combination_to_values[comb_id]
targets1 = drug_to_values[drug1][0]
targets2 = drug_to_values[drug2][0]
for cell_line in cell_line_to_value:
if task != "ch2-test":
if cell_line not in cell_line_to_mono_values:
continue
#print comb_id, cell_line, drug1, drug2
seen_combinations.add((comb_id, cell_line))
feature_values = []
# GEXP categorized
values = [0] * len(targets_all)
if cell_line in cell_line_to_idx:
for targets in (targets1, targets2):
for target in targets:
if target in gene_to_idx:
values[target_to_idx[target]] += gexp_norm[
gene_to_idx[target],
cell_line_to_idx[cell_line]]
feature_values.extend(values)
#print len(feature_values)
# MUT categorized
values = [0] * len(targets_all)
for targets in (targets1, targets2):
for target in targets:
if target in gene_to_cell_line_to_mutation:
d = gene_to_cell_line_to_mutation[target]
if cell_line in d:
values[target_to_idx[target]] += d[cell_line]
feature_values.extend(values)
#print len(feature_values)
# CNV & ZYG categorized
values = [0] * len(targets_all)
values2 = [0] * len(targets_all)
for targets in (targets1, targets2):
for target in targets:
if target in gene_to_cell_line_to_cnv:
d = gene_to_cell_line_to_cnv[target]
if cell_line in d:
# d contains cnv value and zygosity
values[
target_to_idx[target]] += d[cell_line][0]
values2[
target_to_idx[target]] += d[cell_line][1]
feature_values.extend(values)
feature_values.extend(values2)
# METH categorized
values = [0] * len(targets_all)
if cell_line in meth_cell_line_to_idx:
for targets in (targets1, targets2):
for target in targets:
if target in meth_gene_to_idx:
values[target_to_idx[target]] += meth[
meth_gene_to_idx[target],
meth_cell_line_to_idx[cell_line]]
feature_values.extend(values)
#print len(feature_values)
# KEGG / COSMIC INVOLVEMENT categorized
values = [0] * len(targets_pathway)
#for targets in (targets1, targets2):
#for target in targets:
#if target in pathway_target_to_idx:
# values[pathway_target_to_idx[target]] += 1
for k, pathway in enumerate(targets_pathway):
values[k] += len(targets1 & pathway_to_genes[pathway])
values[k] += len(targets2 & pathway_to_genes[pathway])
feature_values.extend(values)
#print len(feature_values)
# GEXP
for targets in (targets1, targets2):
indices = []
for target in targets:
if target in gene_to_idx:
indices.append(gene_to_idx[target])
if len(indices) == 0 or cell_line not in cell_line_to_idx:
vals = ["NA"] * 2
else:
values = gexp_norm[indices,
cell_line_to_idx[cell_line]]
vals = [
numpy.median(values),
numpy.median(numpy.abs(values))
] #values.flat[numpy.abs(values).argmax()]]
feature_values.extend(vals)
#print len(feature_values)
# MUT
for targets in (targets1, targets2):
values = []
for target in targets:
if target in gene_to_cell_line_to_mutation:
d = gene_to_cell_line_to_mutation[target]
if cell_line in d:
values.append(d[cell_line])
if len(values) == 0:
val = "NA"
else:
val = numpy.max(values)
feature_values.append(val)
#print len(feature_values)
# CNV
for targets in (targets1, targets2):
values = []
for target in targets:
if target in gene_to_cell_line_to_cnv:
d = gene_to_cell_line_to_cnv[target]
if cell_line in d:
values.append(d[cell_line][0])
if len(values) == 0:
val = "NA"
else:
val = numpy.max(values)
feature_values.append(val)
#print len(feature_values)
# METH
for targets in (targets1, targets2):
indices = []
for target in targets:
if target in meth_gene_to_idx:
indices.append(meth_gene_to_idx[target])
if len(indices
) == 0 or cell_line not in meth_cell_line_to_idx:
vals = ["NA"] * 2
else:
#val = numpy.median(numpy.abs(meth[indices, meth_cell_line_to_idx[cell_line]]))
values = meth[indices,
meth_cell_line_to_idx[cell_line]]
vals = [
numpy.median(values),
numpy.median(numpy.abs(values))
]
feature_values.extend(vals)
#print len(feature_values)
# MEDIAN SYNERGY per cell line / combination
vals = ["NA"] * 2
if cell_line in cell_line_to_synergy:
vals[0] = cell_line_to_synergy[cell_line]
if comb_id in combination_to_synergy:
vals[1] = combination_to_synergy[comb_id]
feature_values.extend(vals)
#print len(feature_values)
# SIMILARITY
vals = ["NA"] * 2
if comb_id in combination_to_similarity:
vals = combination_to_similarity[comb_id]
feature_values.extend(vals)
#print len(feature_values)
# Interaction network based features (degree A/B)
for targets in (targets1, targets2):
values = []
for target in targets:
if target in network_nodes:
d = network.degree(target)
values.append(d)
if len(values) == 0:
val = "NA"
else:
val = numpy.max(values)
feature_values.append(val)
# Interaction network based distance between A-B
values = []
for target1 in targets1:
if target1 not in network_nodes:
continue
for target2 in targets2:
if target2 not in network_nodes:
continue
d = network_utilities.get_shortest_path_length_between(
network, target1, target2)
values.append(d)
if len(values) == 0:
val = "NA"
else:
val = numpy.min(values)
feature_values.append(val)
# GUILD
vals = ["NA"] * 3
if comb_id in combination_to_guild_values:
values_guild = combination_to_guild_values[comb_id]
if cell_line in values_guild:
vals = values_guild[cell_line]
feature_values.extend(vals)
#print len(feature_values)
# KEGG / COSMIC
for genes in (pathway_to_genes["kegg"],
pathway_to_genes["census"]):
# INVOLVEMENT
val = 0
for targets in (targets1, targets2):
val = len(targets & genes)
feature_values.append(val)
#print len(feature_values)
# GEXP
for target in genes:
if target in gene_to_idx:
indices.append(gene_to_idx[target])
if len(indices) == 0 or cell_line not in cell_line_to_idx:
vals = ["NA"] * 2
else:
values = gexp_norm[indices,
cell_line_to_idx[cell_line]]
vals = [numpy.median(values), numpy.max(values)]
feature_values.extend(vals)
#print len(feature_values)
# MUT
values = []
for target in genes:
if target in gene_to_cell_line_to_mutation:
d = gene_to_cell_line_to_mutation[target]
if cell_line in d:
values.append(d[cell_line])
if len(values) == 0:
vals = ["NA"] * 2
else:
vals = [numpy.median(values), numpy.max(values)]
feature_values.extend(vals)
#print len(feature_values)
# CNV
values = []
for target in genes:
if target in gene_to_cell_line_to_cnv:
d = gene_to_cell_line_to_cnv[target]
if cell_line in d:
values.append(d[cell_line][0])
if len(values) == 0:
vals = ["NA"] * 2
else:
vals = [numpy.median(values), numpy.max(values)]
feature_values.extend(vals)
#print len(feature_values)
# CNV target
for targets in (targets1, targets2):
values = []
for target in targets & genes:
if target in gene_to_cell_line_to_cnv:
d = gene_to_cell_line_to_cnv[target]
if cell_line in d:
values.append(d[cell_line][0])
if len(values) == 0:
val = "NA"
else:
val = numpy.mean(values)
feature_values.append(val)
#print len(feature_values)
f.write(
"%s %s %s\n" %
(comb_id, cell_line, " ".join(map(str, feature_values))))
#if comb_id == "BCL2L1.Vinorelbine" and cell_line == "NCI-H1437":
# return
#f.close()
print "Not in seen combinations:"
for comb_id, cell_line_to_mono_values in combination_to_values.iteritems():
for cell_line in cell_line_to_mono_values:
if (comb_id, cell_line) not in seen_combinations:
print comb_id, cell_line
f.write("%s %s %s\n" % (comb_id, cell_line, " ".join(
map(str, [0] * len(feature_values))))
) # to amend problematic ch2 test file
f.close()
return
# ==================================================
# -- load data
# PPI graph
if args.ppi == 'STRING':
G = nx.read_edgelist(
"resources/BioNEV/data/STRING_PPI/STRING_PPI.edgelist")
nodes = pd.read_csv(
'resources/BioNEV/data/STRING_PPI/node_list.txt', sep='\t', index_col=0)
drug_target_dict = load_DTI()
covid_protein_list = make_SARSCOV2_PPI()
# ppid2id
nodes['id'] = nodes.index
ppid2id = nodes.set_index('STRING_id').to_dict()['id'] # len 15131
else:
network_file = "2016data/network/network.sif"
G = wrappers.get_network(network_file, only_lcc=True)
# provide drugs objects
drugs = load_drugs_from("2016data/target/drug_to_geneids.pcl.all")
# provide disease objects
diseases = load_diseases_from("2016data/disease/disease_genes.tsv")
# ==================================================
# ==================================================
# -- graph embedding - graphrep
# try loading pretrained embedding
file_name = f"{args.ppi}_PPI_{args.embs}_embs.txt"
emb_folder = 'saved/embs'
if args.embs == 'struc2vec':
if exists(join(emb_folder, file_name)):