def main():
# natural order: disease <-> target <-> chem
# disease - chem is what is desired
# disease - target is what is desired
# http://www.disgenet.org/static/disgenet_ap1/files/downloads/curated_gene_disease_associations.tsv.gz
results = pd.DataFrame()
h_network = generate_heterogeneous_network(ppi_path, dg_path, dch_path,
chg_file)
for use_dge, dataset in itt.product([False, True],
AD_DGE_DATASETS + NON_AD_DGE_DATASETS):
disease_abv = dataset_to_disease_abv(dataset)
dge_params = dge_params_ad if disease_abv == 'ad' else dge_params_dis
do_id = disease_identifiers[disease_abv]
logger.debug(
f'Running for disease {disease_abv}, with the dataset {dataset}, '
f'using the id {disease_identifiers[disease_abv]}')
try:
gene_list = parse_dge(
dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
)
h_network_ds = deepcopy(h_network)
h_network_ds.set_up_network(genes=gene_list)
num_walks = gat2vec_config.num_walks
walk_length = gat2vec_config.walk_length
dimension = gat2vec_config.dimension
window_size = gat2vec_config.window_size
metrics_df = predict_links_cv(h_network_ds,
use_dge=use_dge,
dataset=dataset,
num_walks=num_walks,
walk_length=walk_length,
dimension=dimension,
window_size=window_size)
results = results.append(metrics_df, ignore_index=True)
except ValueError:
logger.error(
f'Dataset {dataset} ({do_id}) not found in the graph.')
results.to_csv(os.path.join(g2v_path, 'results_df.tsv'), sep='\t')
print("done")
def get_ppi_results(ppi_graph_path: str,
dataset: str,
evaluation: str = 'cv',
assoc_path: str = None,
phewas: str = None) -> pd.DataFrame:
dge_params = dge_params_ad if dataset in AD_DGE_DATASETS else dge_params_dis
gene_list = parse_dge(dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
csv_separator=';')
network = generate_ppi_network(
ppi_graph_path=ppi_graph_path,
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
ppi_edge_min_confidence=ppi_edge_min_confidence,
current_disease_ids_path='',
disease_associations_path=phewas,
)
logger.info(f'Nodes {len(network.graph.vs)}')
targets = parse_gene_list(targets_file(dataset_to_disease_abv(dataset)),
network)
assoc_score = assoc_path and parse_association_scores(assoc_path)
logger.debug(
f'Number of targets being used for the network: {len(targets)}')
write_gat2vec_input_files(
network=network,
targets=targets,
home_dir=g2v_path,
assoc_score=assoc_score,
)
metrics_df, _ = rank_targets(directory=g2v_path,
network=network,
evaluation=evaluation)
df = pd.DataFrame()
df['auc'] = metrics_df['auc']
df['aps'] = metrics_df['aps']
df['eval'] = evaluation
df['dge'] = dataset
return df
def get_bel_results(dataset) -> (pd.DataFrame, pd.DataFrame):
""""""
dge_params = dge_params_ad if dataset in AD_DGE_DATASETS else dge_params_dis
gene_list = parse_dge(dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
csv_separator=';')
network = generate_bel_network(
bel_graph_path=bel_files,
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
)
logger.info(f'Nodes {len(network.graph.nodes)}')
targets = parse_gene_list(
os.path.join(targets_base_path, dataset_to_disease_abv(dataset),
'ot_symbol.txt'), network)
logger.debug(
f'Number of targets being used for the network: {len(targets)}')
write_gat2vec_input_files(
network=network,
targets=targets,
home_dir=g2v_path,
)
metrics_df, _ = rank_targets(directory=g2v_path,
network=network,
num_walks=30,
walk_length=4,
dimension=256,
window_size=10)
df = pd.DataFrame()
df['auc'] = metrics_df['auc']
df['aps'] = metrics_df['aps']
df['ds'] = 'bel'
df['dge'] = dataset
return df
def get_ppi_results(ppi_graph_path: str, dataset: str) -> pd.DataFrame:
dge_params = dge_params_ad if dataset in AD_DGE_DATASETS else dge_params_dis
gene_list = parse_dge(
dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
)
network = generate_ppi_network(
ppi_graph_path=ppi_graph_path,
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
ppi_edge_min_confidence=ppi_edge_min_confidence,
current_disease_ids_path='',
)
logger.info(f'Nodes {len(network.graph.vs)}')
logger.info(f'Edges {len(network.graph.es)}')
targets = parse_gene_list(targets_file(dataset_to_disease_abv(dataset)),
network)
logger.debug(
f'Number of targets being used for the network: {len(targets)}')
write_gat2vec_input_files(
network=network,
targets=targets,
home_dir=g2v_path,
)
metrics_df, _ = rank_targets(directory=g2v_path, network=network)
df = pd.DataFrame()
df['auc'] = metrics_df['auc']
df['aps'] = metrics_df['aps']
df['ds'] = network_alias[os.path.basename(ppi_graph_path)]
df['dge'] = dataset
return df
def main():
for dge in ['BM10', 'BM22', 'BM36', 'BM44']:
dge_path = dge_base_path % dge
gene_list = parse_dge(dge_path=dge_path,
entrez_id_header=entrez_id_name,
log2_fold_change_header=log_fold_change_name,
adj_p_header=adjusted_p_value_name,
entrez_delimiter=split_char,
base_mean_header=base_mean_name,
csv_separator=';')
dim = len(graph_paths)
fig, axs = plt.subplots(ncols=dim, sharey='all', squeeze=False)
fig.set_size_inches(10, 5)
fig.suptitle(f'DGE {dge}')
df = pd.DataFrame()
axs_ind = 0
for ppi_graph_path in graph_paths:
max_log2_fold_change, min_log2_fold_change = lfc_cutoff, lfc_cutoff * -1
network = generate_ppi_network(
ppi_graph_path=os.path.join(ppi_base_path, ppi_graph_path),
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
ppi_edge_min_confidence=ppi_edge_min_confidence,
current_disease_ids_path='',
disease_associations_path=phewas_path,
)
targets = parse_gene_list(targets_file, network.graph)
assoc_score = assoc_file and parse_association_scores(assoc_file)
# File with no weights
write_gat2vec_input_files(network=network,
targets=targets,
home_dir=g2v_path)
auc_df, _ = rank_targets(
directory=g2v_path,
network=network,
)
df['rr'] = auc_df['auc']
auc_df, _ = rank_targets(directory=g2v_path,
network=network,
evaluation='svm',
class_weights='balanced')
df['bsvm'] = auc_df['auc']
# File with weights
write_gat2vec_input_files(network=network,
targets=targets,
home_dir=g2v_path,
assoc_score=assoc_score)
auc_df, _ = rank_targets(
directory=g2v_path,
network=network,
)
df['wrr'] = auc_df['auc']
auc_df, _ = rank_targets(directory=g2v_path,
network=network,
evaluation='svm',
class_weights='balanced')
df['wbsvm'] = auc_df['auc']
df.boxplot(column=['rr', 'wrr', 'bsvm', 'wbsvm'],
ax=axs[0][axs_ind])
axs[0][axs_ind].set_title(f'PPI {ppi_graph_path}"')
axs_ind += 1
fig.savefig(f'comparison_humanbase({dge}).png')
def main():
for disease in DISEASE_ABBREVIATIONS:
targets_file = targets_file_mask % disease
assoc_file = assoc_file_mask % disease
g2v_path = g2v_path_mask % disease
dge_path = dge_path_mask % disease
graph_paths = graphs_for_diseases[disease]
gene_list = parse_dge(dge_path=dge_path,
entrez_id_header=entrez_id_name,
log2_fold_change_header=log_fold_change_name,
adj_p_header=adjusted_p_value_name,
entrez_delimiter=split_char,
base_mean_header=base_mean_name,
csv_separator=';')
dim = len(graph_paths)
fig, axs = plt.subplots(ncols=dim, sharey='all', squeeze=False)
fig.set_size_inches(10, 5)
fig.suptitle(f'DGE {disease}')
df = pd.DataFrame()
axs_ind = 0
for ppi_graph_path, ppi_edge_min_confidence in itertools.product(
graph_paths, confidence_cutoffs):
max_log2_fold_change, min_log2_fold_change = lfc_cutoff, lfc_cutoff * -1
network = generate_ppi_network(
ppi_graph_path=os.path.join(base_ppi_path, ppi_graph_path),
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
ppi_edge_min_confidence=ppi_edge_min_confidence,
current_disease_ids_path='',
disease_associations_path=phewas_path,
)
targets = parse_gene_list(targets_file, network.graph)
assoc_score = assoc_file and parse_association_scores(assoc_file)
# File with no weights
write_gat2vec_input_files(network=network,
targets=targets,
home_dir=g2v_path)
auc_df, _ = rank_targets(
directory=g2v_path,
network=network,
)
df['rr'] = auc_df['auc']
logger.debug(f'types for {disease}')
logger.debug(type(df['rr']))
logger.debug(type(df))
auc_df, _ = rank_targets(directory=g2v_path,
network=network,
evaluation='svm',
class_weights='balanced')
df['bsvm'] = auc_df['auc']
logger.debug(type(df['bsvm']))
logger.debug(type(df))
# File with weights
write_gat2vec_input_files(network=network,
targets=targets,
home_dir=g2v_path,
assoc_score=assoc_score)
auc_df, _ = rank_targets(
directory=g2v_path,
network=network,
)
df['wrr'] = auc_df['auc']
logger.debug(type(df['wrr']))
logger.debug(type(df))
auc_df, _ = rank_targets(directory=g2v_path,
network=network,
evaluation='svm',
class_weights='balanced')
df['wbsvm'] = auc_df['auc']
logger.debug(type(df['wbsvm']))
logger.debug(type(df))
df.boxplot(column=['rr', 'wrr', 'bsvm', 'wbsvm'], ax=axs)
axs[0][axs_ind].set_title(
f'PPI {ppi_graph_path}", cutoff {ppi_edge_min_confidence}')
axs_ind += 1
fig.savefig(f'comparison_humanbase-{disease}.png')
def main():
results_dict = defaultdict(list)
h_network1 = generate_heterogeneous_network(ppi_path, dg_path, dch_path,
chg_file)
for use_dge, dataset in itt.product([False, True], AD_DGE_DATASETS[2] +
NON_AD_DGE_DATASETS):
disease_abv = dataset_to_disease_abv(dataset)
do_id = disease_identifiers[disease_abv]
dge_params = dge_params_ad if disease_abv == 'ad' else dge_params_dis
logger.debug(
f'Running for disease {disease_abv}, with the dataset {dataset}, using the id {do_id}'
)
try:
gene_list = parse_dge(dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
csv_separator=';')
h_network = deepcopy(h_network1)
h_network.set_up_network(genes=gene_list)
h_network.write_gat2vec_input_files(
home_dir=g2v_path,
disease_id=do_id,
use_dge_data=use_dge,
filter_pleiotropic_targets=True)
for i in range(10):
auc, aps = predict_links(
g2v_path,
gat2vec_config.num_walks,
gat2vec_config.walk_length,
gat2vec_config.dimension,
gat2vec_config.window_size,
)
logger.debug(f'tr: {i}\t{round(auc, 3)}\t{round(aps, 3)}')
results_dict['tr'].append(i)
results_dict['auc'].append(auc)
results_dict['aps'].append(aps)
results_dict['dge'].append(dataset)
results_dict['eval'].append(use_dge)
except ValueError:
logger.error(
f'Dataset {dataset} ({do_id}) not found in the graph.')
results = pd.DataFrame(results_dict)
results.to_csv(os.path.join(g2v_path, 'results_df.tsv'), sep='\t')
for key, datasets in DGE_DATASETS.items():
for metric in ['auc', 'aps']:
fig = sns.boxplot(data=results[results['dge'].isin(datasets)],
x='dge',
y='auc',
hue='eval').get_figure()
fig.suptitle('Link Prediction.')
fig.savefig(f'comp7_{key}_{metric}.png')
plt.close()
print("done")
def optimize_g2v_parameters(
ppi_graph_path: str,
dataset: str,
evaluation: str = 'cv',
assoc_path: str = None
) -> pd.DataFrame:
dge_params = dge_params_ad if dataset in AD_DGE_DATASETS else dge_params_dis
gene_list = parse_dge(
dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
csv_separator=';'
)
network = generate_ppi_network(
ppi_graph_path=ppi_graph_path,
dge_list=gene_list,
max_adj_p=max_padj,
max_log2_fold_change=max_log2_fold_change,
min_log2_fold_change=min_log2_fold_change,
ppi_edge_min_confidence=ppi_edge_min_confidence,
current_disease_ids_path='',
disease_associations_path=phewas_path,
)
logger.info(f'Nodes {len(network.graph.vs)}')
targets = parse_gene_list(targets_file(dataset_to_disease_abv(dataset)), network)
assoc_score = assoc_path and parse_association_scores(assoc_path)
logger.debug(f'Number of targets being used for the network: {len(targets)}')
write_gat2vec_input_files(
network=network,
targets=targets,
home_dir=g2v_path,
assoc_score=assoc_score
)
results = pd.DataFrame()
for dim in [32, 64, 128, 256]:
metrics_df, _ = rank_targets(
directory=g2v_path,
network=network,
evaluation=evaluation,
dimension=dim,
)
df = pd.DataFrame()
df['auc'] = metrics_df['auc']
df['aps'] = metrics_df['aps']
df['eval'] = str(dim)
df['dge'] = dataset
df['param'] = 'Dimension'
logger.debug('df')
logger.debug(df)
results = results.append(
assemble_results_df(metrics_df, dataset, 'Dimension', dim),
ignore_index=True
)
"""num_walks: int = 10
walk_length: int = 80
dimension: int = 128
window_size: int = 5"""
for nw in [6, 10, 20, 40, 80]:
metrics_df, _ = rank_targets(
directory=g2v_path,
network=network,
evaluation=evaluation,
dimension=256,
num_walks=nw,
)
results = results.append(
assemble_results_df(metrics_df, dataset, 'Num Walks', nw),
ignore_index=True
)
for wl in [20, 40, 80, 120, 160]:
metrics_df, _ = rank_targets(
directory=g2v_path,
network=network,
evaluation=evaluation,
dimension=256,
walk_length=wl
)
results = results.append(
assemble_results_df(metrics_df, dataset, 'Walk Length', wl),
ignore_index=True
)
for ws in [3, 5, 7, 10, 20, 40]:
metrics_df, _ = rank_targets(
directory=g2v_path,
network=network,
evaluation=evaluation,
dimension=256,
window_size=ws
)
results = results.append(
assemble_results_df(metrics_df, dataset, 'Window Size', ws),
ignore_index=True
)
logger.debug('results')
logger.debug(results)
return results
def main():
# natural order: disease <-> target <-> chem
# disease - chem is what is desired
# disease - target is what is desired
# http://www.disgenet.org/static/disgenet_ap1/files/downloads/curated_gene_disease_associations.tsv.gz
results_dict = defaultdict(list)
h_network1 = generate_heterogeneous_network(ppi_path, dg_path, dch_path,
chg_file)
for use_dge, dataset in itt.product([True],
AD_DGE_DATASETS + NON_AD_DGE_DATASETS):
disease_abv = dataset_to_disease_abv(dataset)
do_id = disease_identifiers[disease_abv]
dge_params = dge_params_ad if disease_abv == 'ad' else dge_params_dis
logger.debug(
f'Running for disease {disease_abv}, with the dataset {dataset}, using the id {do_id}'
)
try:
gene_list = parse_dge(
dge_path=dge_file(dataset),
entrez_id_header=dge_params['id'],
log2_fold_change_header=dge_params['l2f'],
adj_p_header=dge_params['adjp'],
entrez_delimiter=split_char,
base_mean_header=dge_params['mean'],
)
h_network = deepcopy(h_network1)
h_network.set_up_network(genes=gene_list)
h_network.write_gat2vec_input_files(
home_dir=g2v_path,
disease_id=do_id,
filter_pleiotropic_targets=True)
# num_walks = gat2vec_config.num_walks
walk_length = gat2vec_config.walk_length
dimension = gat2vec_config.dimension
window_size = gat2vec_config.window_size
param = 'nw'
for num_walks in [6, 10, 20, 40, 80]:
start = time()
lp_results = mp_predict_links(num_walks, walk_length,
dimension, window_size)
extract_results(results_dict, lp_results, dataset, param,
num_walks)
logger.info(
f'Runtime for num_walks = {num_walks}: {time() - start}s')
# best result from num_walks
num_walks = gat2vec_config.num_walks
param = 'wl'
for walk_length in [20, 40, 80, 120, 160]:
start = time()
lp_results = mp_predict_links(num_walks, walk_length,
dimension, window_size)
extract_results(results_dict, lp_results, dataset, param,
walk_length)
logger.info(
f'Runtime for walk_length = {walk_length}: {time() - start}s'
)
# best result from num_walks
walk_length = gat2vec_config.walk_length
param = 'ws'
for window_size in [3, 5, 7, 10, 20, 40]:
start = time()
lp_results = mp_predict_links(num_walks, walk_length,
dimension, window_size)
extract_results(results_dict, lp_results, dataset, param,
window_size)
logger.info(
f'Runtime for window_size = {window_size}: {time() - start}s'
)
# best result from num_walks
window_size = gat2vec_config.window_size
param = 'd'
for dimension in [32, 64, 128, 256]:
start = time()
lp_results = mp_predict_links(num_walks, walk_length,
dimension, window_size)
extract_results(results_dict, lp_results, dataset, param,
dimension)
logger.info(
f'Runtime for dimension = {dimension}: {time() - start}s')
except ValueError:
logger.error(
f'Dataset {dataset} ({do_id}) not found in the graph.')
results = pd.DataFrame(results_dict)
# backup intermediate results in each iteration.
results.to_csv(os.path.join(g2v_path, 'results_df_nonad.tsv'),
sep='\t')
print("done")