def get_trained_network(n_epochs, oversampling_ratio):
if uf.manage_obj('trained_networks/VNN_%d_%.2f' %
(n_epochs, oversampling_ratio)) is False:
layer_info, matrix_list, data_set = set_inputs(oversampling_ratio)
net, criterion, optimizer, device = set_network(
layer_info, matrix_list)
train_loader, test_loader = uf.get_train_test_valid_set(data_set, 0.3)
train_loader = td.DataLoader(data_set,
batch_size=50,
shuffle=True,
num_workers=10)
train_epoch(net, optimizer, criterion, train_loader, test_loader,
device, n_epochs)
torch.save(
net, '../data/objects/trained_networks/VNN_%d_%.2f.obj' %
(n_epochs, oversampling_ratio))
net.eval()
return net
else:
net = torch.load('../data/objects/trained_networks/VNN_%d_%.2f.obj' %
(n_epochs, oversampling_ratio))
net.eval()
return net
def get_RWVNN_path(single_data, is_comb, n_epochs, oversampling_ratio, *drug_name):
net = vnn.get_trained_network(n_epochs, oversampling_ratio)
if is_comb:
dir_path = "../RW/RW_combi_DEG_prediction_results"
else:
dir_path = "../RW/RW_DEG_prediction_results"
filenames = os.listdir(dir_path)
drug_target_dict = get_target_dict(is_comb)
if not single_data:
# for all drug
for i, filename in enumerate(filenames):
if not filename.endswith("_RW_result.txt"):
drug_name = filename.split('.')[0]
else:
continue
if drug_name not in drug_target_dict.keys():
continue
full_name = os.path.join(dir_path, filename)
degs = list()
for deg in uf.read_file(full_name):
degs += deg
framework(is_comb, drug_name, drug_target_dict[drug_name], degs, net)
# for single drug
elif single_data:
degs = list()
for deg in uf.read_file(os.path.join(dir_path, drug_name + ".txt")):
degs += deg
if drug_name not in drug_target_dict.keys():
print('test fail')
return 0
framework(is_comb, drug_name, drug_target_dict[drug_name], degs, net)
def get_auroc(n_epochs, cross_validation, oversampling_ratio):
if cross_validation is False:
layer_info, matrix_list, data_set = set_inputs(oversampling_ratio)
net, criterion, optimizer, device = set_network(
layer_info, matrix_list)
train_loader, test_loader = uf.get_train_test_valid_set(data_set, 0.3)
########################################
######### TRAINING ##########
########################################
train_epoch(net, optimizer, criterion, train_loader, test_loader,
device, n_epochs)
########################################
net.eval()
result = uf.get_sigmoid_auroc(test_loader, device, net)
result = [n_epochs, result]
return result
else:
layer_info, matrix_list, data_set = set_inputs(oversampling_ratio)
cross_validation_idx = uf.cross_validation_index(data_set)
result = list()
for train_idx, test_idx in cross_validation_idx:
net, criterion, optimizer, device = set_network(
layer_info, matrix_list)
train_sampler = SubsetRandomSampler(train_idx)
test_sampler = SubsetRandomSampler(test_idx)
train_loader = td.DataLoader(data_set,
batch_size=50,
sampler=train_sampler,
num_workers=10)
test_loader = td.DataLoader(data_set,
batch_size=50,
sampler=test_sampler,
num_workers=10)
########################################
######### TRAINING ##########
########################################
train_epoch(net, optimizer, criterion, train_loader, test_loader,
device, n_epochs)
########################################
net.eval()
scores = uf.get_sigmoid_auroc(test_loader, device, net)
net.train()
scores = [n_epochs, scores]
result.append(scores)
return result
def get_disease_comb_drug():
path = "../data/drug/Combination_Drug_Targets_From_DCDB.txt"
combi_file = uf.read_file(path)
result = list()
for combi in combi_file:
if combi[1] == "P":
result.append(combi[0])
return result
def build_network(file_path: str) -> nx.DiGraph:
result = nx.DiGraph()
relation_list = uf.read_file(file_path)
for relation in relation_list:
result.add_edge(relation[0], relation[1], type=relation[2])
return result
def get_vnn_prediction(degs, net):
# get vnn prediction
result = list()
relation_files = ['GE_MF', 'MF_BP', 'BP_PH']
gene_index = uf.set_layer_info(relation_files)[0]
params = rpf.get_sigmoid_network_params(
net, degs, gene_index) # [pred_score, BP, MF, net_paramters]
result.append(params[0]) # prediction score
return result
def framework(is_comb, drug_name, targets, degs, net):
# get RW result
network_path = '../data/network/RW/GP_GP_Four_Relation_Types.txt'
network = build_network(network_path)
rw_paths = get_rw_path(network, targets, degs)
gene_RW_dict = get_rw_score_dcit(is_comb, drug_name)
#get vnn result
relation_files = ['GE_MF', 'MF_BP', 'BP_PH']
layer_info = uf.set_layer_info(relation_files)
matrix_list = uf.set_mask_matrics(relation_files, layer_info)
gene_index = layer_info[0]
params = get_sigmoid_network_params(net, degs, gene_index) # [pred_score, BP, MF, net_paramters]
vnn_paths = get_vnn_path(degs, params, layer_info, matrix_list)
#write result
result = get_total_path(rw_paths, gene_RW_dict, vnn_paths)
save_path = 'path_result/%s_path.tsv' % drug_name
uf.write_tsv(result, save_path)
def get_vnn_score(dirname, output_path, is_comb, net):
if is_comb:
disease_drug = get_disease_comb_drug()
else:
disease_drug = uf.get_diseas_drug()
filenames = os.listdir(dirname)
result = list()
result.append(['drug_name', 'label', 'prediction'])
for i, filename in enumerate(filenames):
if not filename.endswith("_RW_result.txt"):
drug_name = filename.split('.')[0]
else:
continue
full_name = os.path.join(dirname, filename)
degs = list()
for deg in uf.read_file(full_name):
degs += deg
line = list()
line.append(drug_name)
if not is_comb:
drug_name = drug_name.lower()
if drug_name in disease_drug:
line.append(1)
else:
line.append(0)
line += get_vnn_prediction(degs, net)
result.append(line)
uf.write_tsv(result, output_path)
def get_rw_score_dcit(is_comb, drug_name):
result = dict()
if is_comb:
dir_path = "../RW/RW_combi_DEG_prediction_results/"
else:
dir_path = "../RW/RW_DEG_prediction_results/"
raw_file = uf.read_file(''.join([dir_path, drug_name, "_RW_result.txt"]))
for line in raw_file:
result[line[0]] = float(line[1])
return result
def get_target_dict(is_comb) -> dict:
result = dict()
if not is_comb:
path = "../data/drug/Single_Drug_Target.txt"
target_file = uf.read_file(path)
for line in target_file:
result[line[0]] = line[1].split('|')
return result
elif is_comb:
path = "../data/drug/Combination_Drug_Targets_From_DCDB.txt"
target_file = uf.read_file(path)
target_file.remove(target_file[0])
for line in target_file:
result[line[0]] = line[3].split('|')
return result
def set_inputs(oversampling_ratio):
relation_files = ['GE_MF', 'MF_BP', 'BP_PH']
layer_info = uf.set_layer_info(relation_files)
gene_index = layer_info[0]
matrix_list = uf.set_mask_matrics(relation_files, layer_info)
disease_drug = uf.get_diseas_drug()
deg_drug_train = uf.get_DEG_drug(disease_drug)
data_set = uf.set_input_data(deg_drug_train, gene_index, 'train_data')
# oversampling
data_set = uf.oversampling(data_set, oversampling_ratio)
return [layer_info, matrix_list, data_set]