def test_cluster_manta(self):
"""
Checks whether the correct cluster IDs are assigned.
"""
clustered_graph = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
clusters = nx.get_node_attributes(clustered_graph[0], 'cluster')
self.assertEqual(clusters['OTU_10'], clusters['OTU_6'])
def test_layout(self):
"""
Checks whether the layout function returns a dictionary of coordinates.
"""
clustered_graph = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters, min_cluster_size,
iterations, subset, ratio, edgescale, permutations, verbose)
coords = generate_layout(clustered_graph[0])
self.assertEqual(len(coords[list(coords.keys())[0]]), 2)
def perm_clusters(graph, limit, max_clusters, min_clusters, min_cluster_size,
iterations, ratio, partialperms, relperms, subset, error, verbose):
"""
Calls the rewire_graph function and robustness function
to compute robustness of cluster assignments.
Scores close to 1 imply that the scores are robust to perturbation.
Parameters
----------
:param graph: NetworkX graph of a microbial association network. Cluster assignment should be a network property.
:param limit: Percentage in error decrease until matrix is considered converged.
:param max_clusters: Maximum number of clusters to evaluate in K-means clustering.
:param min_clusters: Minimum number of clusters to evaluate in K-means clustering.
:param min_cluster_size: Minimum cluster size as fraction of network size
:param iterations: If algorithm does not converge, it stops here.
:param ratio: Ratio of scores that need to be positive or negative for a stable edge
:param partialperms: Number of permutations for partial diffusion.
:param relperms: Number of permutations for reliability testing.
:param subset: Fraction of edges used in subsetting procedure
:param error: Fraction of edges to rewire for reliability metric.
:param verbose: Verbosity level of function
:return:
"""
assignments = list()
rev_assignments = list()
for i in range(relperms):
permutation, swapfail = rewire_graph(graph, error)
if swapfail:
return
permutation, mat = cluster_graph(graph=permutation, limit=limit, max_clusters=max_clusters,
min_clusters=min_clusters, min_cluster_size=min_cluster_size,
iterations=iterations,
ratio=ratio, edgescale=0, permutations=partialperms,
subset=subset,
verbose=False)
cluster = nx.get_node_attributes(permutation, 'cluster')
# cluster.values() has same order as permutation.nodes
assignments.append(cluster)
subassignments = dict()
for k, v in cluster.items():
subassignments.setdefault(v, set()).add(k)
rev_assignments.append(subassignments)
if verbose:
logger.info('Permutation ' + str(i))
graphclusters = nx.get_node_attributes(graph, 'cluster')
clusjaccards, nodejaccards, ci_width = robustness(graphclusters, assignments)
lowerCI = dict()
upperCI = dict()
for node in nodejaccards:
lowerCI[node] = np.asscalar(nodejaccards[node][0])
upperCI[node] = np.asscalar(nodejaccards[node][1])
ci_width[node] = np.asscalar(ci_width[node])
nx.set_node_attributes(graph, lowerCI, "lowerCI")
nx.set_node_attributes(graph, upperCI, "upperCI")
nx.set_node_attributes(graph, ci_width, "widthCI")
if verbose:
logger.info("Completed estimation of node Jaccard similarities across bootstraps.")
def test_directed_manta(self):
"""
Checks whether the main function can carry out clustering
on a DiGraph.
"""
clustered_graph = cluster_graph(deepcopy(directg), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
clusters = nx.get_node_attributes(clustered_graph[0], 'cluster')
self.assertEqual(len(clusters), 10)
def test_central_node(self):
"""
Checks if, given a graph that has been tested for centrality,
no nodes are identified as hubs (actually the p-value is too low).
"""
results = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
graph = results[0]
central_edge(graph, percentile, permutations, error, verbose)
central_node(graph)
self.assertEqual(len(nx.get_node_attributes(graph, 'hub')), 0)
def test_default_manta(self):
"""
Checks whether the main function carries out both clustering and centrality estimates.
"""
clustered_graph = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
graph = clustered_graph[0]
central_edge(graph, percentile, rel,
error, verbose)
central_node(graph)
clusters = nx.get_node_attributes(graph, 'cluster')
hubs = nx.get_edge_attributes(graph, 'hub')
self.assertGreater(len(hubs), 0)
def test_bootstrap(self):
"""Checks if reliability scores for the graph are returned. """
results = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
# calculates the ratio of positive / negative weights
# note that ratios need to be adapted, because the matrix is symmetric
matrix = results[1]
negthresh = np.percentile(matrix, percentile)
posthresh = np.percentile(matrix, 100 - percentile)
neghubs = list(map(tuple, np.argwhere(matrix <= negthresh)))
poshubs = list(map(tuple, np.argwhere(matrix >= posthresh)))
bootmats = perm_edges(g, permutations, percentile, poshubs, neghubs, error=0.1)
self.assertEqual((len(poshubs) + len(neghubs)), len(bootmats))
def test_center_manta(self):
"""
Checks if the edge between 1 and 2 is identified as a positive hub.
WARNING: at the moment the test indicates that centrality measures
are not stable.
"""
results = cluster_graph(deepcopy(g), limit, max_clusters, min_clusters,
min_cluster_size, iterations, subset, ratio, edgescale, permutations, verbose)
graph = results[0]
central_edge(graph, percentile, permutations, error, verbose)
hubs = nx.get_edge_attributes(graph, 'hub')
if ('OTU_4', 'OTU_3') in hubs:
hubs['OTU_3', 'OTU_4'] = hubs['OTU_4', 'OTU_3']
# 3.5 test sometimes swaps keys around apparently
self.assertEqual(hubs[('OTU_3', 'OTU_4')], 'negative hub')
def main():
args = set_manta().parse_args(sys.argv[1:])
args = vars(args)
if args['version']:
info = VersionInfo('manta')
logger.info('Version ' + info.version_string())
exit(0)
if args['graph'] != 'demo':
filename = args['graph'].split(sep=".")
extension = filename[len(filename)-1]
# see if the file can be detected
# if not, try appending current working directory and then read.
if not os.path.isfile(args['graph']):
if os.path.isfile(os.getcwd() + '/' + args['graph']):
args['graph'] = os.getcwd() + '/'
else:
logger.error('Could not find the specified file. Is your file path correct?')
exit()
try:
if extension == 'graphml':
network = nx.read_graphml(args['graph'])
elif extension == 'txt':
network = nx.read_weighted_edgelist(args['graph'])
elif extension == 'gml':
network = nx.read_gml(args['graph'])
elif extension == 'cyjs':
network = read_cyjson(args['graph'])
else:
logger.warning('Format not accepted. '
'Please specify the filename including extension (e.g. test.graphml).', exc_info=True)
exit()
except Exception:
logger.error('Could not import network file!', exc_info=True)
exit()
# first need to convert network to undirected
elif args['graph'] == 'demo':
path = os.path.dirname(manta.__file__)
path = path + '//demo.graphml'
network = nx.read_graphml(path)
if args['direction']:
if extension == 'txt':
logger.warning('Directed networks from edge lists not supported, use graphml or cyjs! ')
exit()
else:
network = nx.to_undirected(network)
if args['bin']:
orig_edges = dict()
# store original edges for export
for edge in network.edges:
orig_edges[edge] = network.edges[edge]['weight']
network.edges[edge]['weight'] = np.sign(network.edges[edge]['weight'])
if sum(value == 0 for value in
np.any(nx.get_edge_attributes(network, 'weight').values())) > 0:
logger.error("Some edges in the network have a weight of exactly 0. \n"
"Such edges cannot be clustered. Try converting weights to 1 and -1. ")
weight_properties = nx.get_edge_attributes(network, 'weight')
if len(weight_properties) == 0:
logger.error("The imported network has no 'weight' edge property. \n"
"Please make sure you are formatting the network correctly. ")
results = cluster_graph(network, limit=args['limit'], max_clusters=args['max'],
min_clusters=args['min'], min_cluster_size=args['ms'],
iterations=args['iter'], subset=args['subset'],
ratio=args['ratio'], edgescale=args['edgescale'],
permutations=args['perm'], verbose=args['verbose'])
graph = results[0]
if args['cr']:
perm_clusters(graph=graph, limit=args['limit'], max_clusters=args['max'],
min_clusters=args['min'], min_cluster_size=args['ms'],
iterations=args['iter'], ratio=args['ratio'],
partialperms=args['perm'], relperms=args['rel'], subset=args['subset'],
error=args['error'], verbose=args['verbose'])
layout = None
if args['bin']:
for edge in network.edges:
network.edges[edge]['weight'] = orig_edges[edge]
if args['layout']:
layout = generate_layout(graph, args['tax'])
if args['fp']:
if args['f'] == 'graphml':
nx.write_graphml(graph, args['fp'] + '.graphml')
elif args['f'] == 'csv':
node_keys = graph.nodes[list(graph.nodes)[0]].keys()
properties = {}
for key in node_keys:
properties[key] = nx.get_node_attributes(graph, key)
data = pd.DataFrame(properties)
data.to_csv(args['fp'] + '.csv')
elif args['f'] == 'gml':
nx.write_gml(graph, args['fp'] + '.gml')
elif args['f'] == 'cyjs':
write_cyjson(graph=graph, filename=args['fp'] + '.cyjs', layout=layout)
logger.info('Wrote clustered network to ' + args['fp'] + '.' + args['f'])
else:
logger.error('Could not write network to disk, no file path given.')
exit(0)
def main():
args = set_manta().parse_args(sys.argv[1:])
args = vars(args)
if args['version']:
info = VersionInfo('manta')
logger.info('Version ' + info.version_string())
exit(0)
if args['graph'] != 'demo':
filename = args['graph'].split(sep=".")
extension = filename[len(filename) - 1]
# see if the file can be detected
# if not, try appending current working directory and then read.
if not os.path.isfile(args['graph']):
if os.path.isfile(os.getcwd() + '/' + args['graph']):
args['graph'] = os.getcwd() + '/'
else:
logger.error(
'Could not find the specified file. Is your file path correct?'
)
exit()
try:
if extension == 'graphml':
network = nx.read_graphml(args['graph'])
elif extension == 'txt':
network = nx.read_weighted_edgelist(args['graph'])
elif extension == 'gml':
network = nx.read_gml(args['graph'])
elif extension == 'cyjs':
network = read_cyjson(args['graph'])
else:
logger.warning(
'Format not accepted. '
'Please specify the filename including extension (e.g. test.graphml).',
exc_info=True)
exit()
except Exception:
logger.error('Could not import network file!', exc_info=True)
exit()
# first need to convert network to undirected
elif args['graph'] == 'demo':
path = os.path.dirname(manta.__file__)
path = path + '//demo.graphml'
network = nx.read_graphml(path)
if args['direction']:
if extension == 'txt':
logger.warning(
'Directed networks from edge lists not supported, use graphml or cyjs! '
)
exit()
else:
network = nx.to_undirected(network)
if args['bin']:
orig_edges = dict()
# store original edges for export
for edge in network.edges:
orig_edges[edge] = network.edges[edge]['weight']
network.edges[edge]['weight'] = np.sign(
network.edges[edge]['weight'])
results = cluster_graph(network,
limit=args['limit'],
max_clusters=args['max'],
min_clusters=args['min'],
min_cluster_size=args['ms'],
iterations=args['iter'],
subset=args['subset'],
ratio=args['ratio'],
edgescale=args['edgescale'],
permutations=args['perm'],
verbose=args['verbose'])
graph = results[0]
if args['cr']:
perm_clusters(graph=graph,
limit=args['limit'],
max_clusters=args['max'],
min_clusters=args['min'],
min_cluster_size=args['ms'],
iterations=args['iter'],
ratio=args['ratio'],
partialperms=args['perm'],
relperms=args['rel'],
subset=args['subset'],
error=args['error'],
verbose=args['verbose'])
layout = None
if args['bin']:
for edge in network.edges:
network.edges[edge]['weight'] = orig_edges[edge]
if args['layout']:
layout = generate_layout(graph, args['tax'])
if args['fp']:
if args['f'] == 'graphml':
nx.write_graphml(graph, args['fp'] + '.graphml')
elif args['f'] == 'edgelist':
nx.write_weighted_edgelist(graph, args['fp'] + '.txt')
elif args['f'] == 'gml':
nx.write_gml(graph, args['fp'] + '.gml')
elif args['f'] == 'adj':
nx.write_multiline_adjlist(graph, args['fp'] + '.txt')
elif args['f'] == 'cyjs':
write_cyjson(graph=graph,
filename=args['fp'] + '.cyjs',
layout=layout)
logger.info('Wrote clustered network to ' + args['fp'] + '.' +
args['f'])
else:
logger.error('Could not write network to disk, no file path given.')
exit(0)
