def cell_cycle():
adata = utils.create_adata_dummy()
adata_int = adata.copy()
# only final score implementation
score = me.cell_cycle(adata,
adata_int,
batch_key='batch',
organism='mouse',
agg_func=np.mean,
verbose=True)
print(f"score: {score}")
assert score == 1
# get all intermediate scores
scores_df = me.cell_cycle(adata,
adata_int,
batch_key='batch',
organism='mouse',
agg_func=None,
verbose=True)
print(f"score: {scores_df}")
assert isinstance(scores_df, pd.DataFrame)
for i in scores_df['score']:
assert i == 1
def silhouette():
adata = utils.create_adata_dummy(pca=True, n_top_genes=2000)
score = me.silhouette(adata,
group_key='celltype',
embed='X_pca',
scale=True)
print(f"score: {score}")
assert score >= 0
assert score <= 1
def test_all_metrics():
adata = utils.create_adata_dummy()
adata_int = adata.copy()
for ot in ["full", "embed", "knn"]:
all_metrics(adata,
adata_int,
script="metrics.py",
type_=ot,
method="orig")
def pcr_comparison():
verbose = True
# no PCA precomputed
adata = utils.create_adata_dummy()
adata_int = adata.copy()
score = me.pcr_comparison(adata,
adata_int,
covariate='batch',
n_comps=50,
scale=True,
verbose=verbose)
print(f"no PCA precomputed: {score}")
assert score < 1e-6
# use different embedding
adata = utils.create_adata_dummy()
adata_int = adata.copy()
utils.add_emb(adata_int, type_='full')
score = me.pcr_comparison(adata,
adata_int,
covariate='batch',
embed='X_emb',
n_comps=50,
scale=True,
verbose=verbose)
print(f"using embedding: {score}")
assert score >= 0
assert score <= 1
assert score < 1e-6
# precomputed PCA
adata = utils.create_adata_dummy(pca=True, n_top_genes=2000)
adata_int = adata.copy()
score = me.pcr_comparison(adata,
adata_int,
covariate='batch',
scale=True,
verbose=verbose)
print(f"precomputed PCA: {score}")
assert score == 0 # same PCA values -> difference should be 0
def silhouette_batch():
adata = utils.create_adata_dummy(pca=True, n_top_genes=2000)
_, sil = me.silhouette_batch(adata,
batch_key='batch',
group_key='celltype',
embed='X_pca',
scale=True,
verbose=False)
score = sil['silhouette_score'].mean()
print(f"score: {score}")
assert score >= 0
assert score <= 1
def test_cluster():
adata = utils.create_adata_dummy(pca=True,
n_top_genes=2000,
neighbors=True)
_, _, score_all, clustering = cl.opt_louvain(adata,
label_key='celltype',
cluster_key='cluster',
plot=True,
inplace=False)
assert isinstance(score_all, pd.DataFrame)
assert isinstance(clustering, pd.Series)
def metrics_all_methods():
adata = utils.create_adata_dummy()
methods = {
'scanorama': runScanorama,
'trvae': runTrVae,
'seurat': runSeurat,
'harmony': runHarmony,
'mnn': runMNN,
'bbknn': runBBKNN,
'conos': runConos,
'scvi': runScvi
}
def ari():
adata = utils.create_adata_dummy(pca=True,
n_top_genes=2000,
neighbors=True)
# trivial score
score = scIB.me.ari(adata, 'celltype', 'celltype')
assert score == 1
# on cell type
cluster(adata, cluster_key='cluster', label_key='celltype')
score = me.ari(adata, group1='cluster', group2='celltype')
print(f"score: {score}")
assert score >= 0
assert score <= 1
def isolated_labels():
adata = utils.create_adata_dummy(pca=True,
n_top_genes=2000,
neighbors=True)
# test 2 different implementations of score
for impl in [True, False]:
score = me.isolated_labels(adata,
label_key='celltype',
batch_key='batch',
cluster=impl,
n=4,
verbose=True)
print(f"score: {score}")
assert score <= 1
assert score >= 0
def nmi():
adata = utils.create_adata_dummy(pca=True,
n_top_genes=2000,
neighbors=True)
# trivial score
score = scIB.me.nmi(adata, 'celltype', 'celltype')
assert score == 1
# on cell type
_, _, nmi_all = cluster(adata,
cluster_key='cluster',
label_key='celltype',
verbose=True)
for score in nmi_all['score']:
print(score)
assert score >= 0
assert score <= 1
def hvg_overlap():
adata = utils.create_adata_dummy()
adata_int = adata.copy()
score = me.hvg_overlap(adata_int, adata, batch='batch', n_hvg=500)
print(f"score: {score}")
assert score == 1
def setup_test_directory(methods):
"""
create necessary files for a test directory
TODO: use fixtures
TODO: create environments
:params methods: list of method names to be used
"""
methods = [methods] if isinstance(methods, str) else methods
data_dir = os.path.abspath(f"./pipeline-{'_'.join(methods)}")
create_if_missing(data_dir)
print(f"created {data_dir}")
# create input and output directories
input_dir = os.path.join(data_dir, "input")
create_if_missing(input_dir)
output_dir = os.path.join(data_dir, "output")
create_if_missing(output_dir)
# write data files
input_adata_file = os.path.join(input_dir, "adata_raw.h5ad")
if not os.path.isfile(input_adata_file):
adata = utils.create_adata_dummy(pca=True,
n_top_genes=2000,
neighbors=True)
adata.write(input_adata_file)
# write config file
config = {
"ROOT": output_dir,
"r_env": "benchmarking_data_integration_dev",
"py_env": "benchmarking_data_integration_dev",
"conv_env": "benchmarking_data_integration_dev",
"timing": False,
"FEATURE_SELECTION": {
"hvg": 2000,
"full_feature": 0
},
"SCALING": ["unscaled", "scaled"],
"METHODS": {k: METHODS[k]
for k in methods},
"DATA_SCENARIOS": {
"test_data": {
"batch_key": "batch",
"label_key": "celltype",
"organism": "mouse",
"assay": "expression",
"file": input_adata_file
}
}
}
config_file = os.path.join(data_dir, "config.json")
with open(config_file, 'w') as f:
f.write(json.dumps(config, indent=4))
workdir = pathlib.Path(scIB.__file__).parent.parent
return {
"workdir": workdir,
"config": config,
"configfile": config_file,
"data_dir": data_dir,
"input_dir": input_dir,
"output_dir": output_dir
}