def test_model_train():
mols = list(oddt.toolkit.readfile('sdf', actives_sdf))[:10]
list(map(lambda x: x.addh(), mols))
rec = next(oddt.toolkit.readfile('pdb', receptor_pdb))
rec.protein = True
rec.addh()
data_dir = os.path.join(test_data_dir, 'data')
home_dir = mkdtemp()
pdbbind_versions = (2007, 2013, 2016)
pdbbind_dir = os.path.join(data_dir, 'pdbbind')
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if not os.path.isdir(version_dir):
os.symlink(pdbbind_dir, version_dir)
for model in [nnscore(n_jobs=1)] + [rfscore(version=v, n_jobs=1)
for v in [1, 2, 3]]:
with NamedTemporaryFile(suffix='.pickle') as f:
model.gen_training_data(data_dir, pdbbind_versions=pdbbind_versions,
home_dir=home_dir)
model.train(home_dir=home_dir, sf_pickle=f.name)
model.set_protein(rec)
preds = model.predict(mols)
assert len(preds) == 10
assert preds.dtype == np.float
assert model.score(mols, preds) == 1.0
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if os.path.islink(version_dir):
os.unlink(version_dir)
def test_rfscore_desc():
"""Test RFScore v1-3 descriptors generators"""
mols = list(oddt.toolkit.readfile('sdf', actives_sdf))
list(map(lambda x: x.addh(), mols))
rec = next(oddt.toolkit.readfile('pdb', receptor_pdb))
rec.protein = True
rec.addh()
# Delete molecule which has differences in Acceptor-Donor def in RDK and OB
del mols[65]
for v in [1, 2, 3]:
descs = rfscore(version=v, protein=rec).descriptor_generator.build(mols)
# save correct results (for future use)
# np.savetxt(os.path.join(results, 'rfscore_v%i_descs.csv' % v),
# descs,
# fmt='%.16g',
# delimiter=',')
descs_correct = np.loadtxt(
os.path.join(results, 'rfscore_v%i_descs.csv' % v),
delimiter=',')
# help debug errors
for i in range(descs.shape[1]):
mask = np.abs(descs[:, i] - descs_correct[:, i]) > 1e-4
if mask.sum() > 1:
print(i, np.vstack((descs[mask, i], descs_correct[mask, i])))
assert_array_almost_equal(descs, descs_correct, decimal=4)
def test_ensemble_descriptor():
mols = list(oddt.toolkit.readfile('sdf', actives_sdf))[:10]
list(map(lambda x: x.addh(), mols))
rec = next(oddt.toolkit.readfile('pdb', receptor_pdb))
rec.protein = True
rec.addh()
desc1 = rfscore(version=1).descriptor_generator
desc2 = oddt_vina_descriptor()
ensemble = ensemble_descriptor((desc1, desc2))
ensemble.set_protein(rec)
assert len(ensemble) == len(desc1) + len(desc2)
# set protein
assert desc1.protein == rec
assert desc2.protein == rec
ensemble_scores = ensemble.build(mols)
scores1 = desc1.build(mols)
scores2 = desc2.build(mols)
assert_array_almost_equal(ensemble_scores, np.hstack((scores1, scores2)))
def test_vs_scoring():
protein = next(oddt.toolkit.readfile('pdb', xiap_protein))
protein.protein = True
data_dir = os.path.join(test_data_dir, 'data')
home_dir = mkdtemp()
pdbbind_versions = (2007, 2013, 2016)
pdbbind_dir = os.path.join(data_dir, 'pdbbind')
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if not os.path.isdir(version_dir):
os.symlink(pdbbind_dir, version_dir)
filenames = []
# train mocked SFs
for model in [nnscore(n_jobs=1)
] + [rfscore(version=v, n_jobs=1) for v in [1, 2, 3]]:
model.gen_training_data(data_dir,
pdbbind_versions=pdbbind_versions,
home_dir=home_dir)
filenames.append(model.train(home_dir=home_dir))
vs = virtualscreening(n_cpu=-1, chunksize=10)
vs.load_ligands('sdf', xiap_actives_docked)
# error if no protein is fed
with pytest.raises(ValueError):
vs.score('nnscore')
# bad sf name
with pytest.raises(ValueError):
vs.score('bad_sf', protein=protein)
vs.score('nnscore', protein=xiap_protein)
vs.score('nnscore_pdbbind2016', protein=protein)
vs.score('rfscore_v1', protein=protein)
vs.score('rfscore_v1_pdbbind2016', protein=protein)
vs.score('rfscore_v2', protein=protein)
vs.score('rfscore_v3', protein=protein)
# use pickle directly
vs.score(filenames[0], protein=protein)
# pass SF object directly
vs.score(scorer.load(filenames[0]), protein=protein)
# pass wrong object (sum is not an instance of scorer)
with pytest.raises(ValueError):
vs.score(sum, protein=protein)
mols = list(vs.fetch())
assert len(mols) == 100
mol_data = mols[0].data
assert 'nnscore' in mol_data
assert 'rfscore_v1' in mol_data
assert 'rfscore_v2' in mol_data
assert 'rfscore_v3' in mol_data
vs = virtualscreening(n_cpu=-1, chunksize=10)
vs.load_ligands('sdf', xiap_actives_docked)
vs.score('nnscore', protein=protein)
vs.score('rfscore_v1', protein=protein)
vs.score('rfscore_v2', protein=protein)
vs.score('rfscore_v3', protein=protein)
with NamedTemporaryFile('w', suffix='.sdf') as molfile:
with NamedTemporaryFile('w', suffix='.csv') as csvfile:
vs.write('sdf', molfile.name, csv_filename=csvfile.name)
data = pd.read_csv(csvfile.name)
assert 'nnscore' in data.columns
assert 'rfscore_v1' in data.columns
assert 'rfscore_v2' in data.columns
assert 'rfscore_v3' in data.columns
mols = list(oddt.toolkit.readfile('sdf', molfile.name))
assert len(mols) == 100
vs.write_csv(
csvfile.name,
fields=['nnscore', 'rfscore_v1', 'rfscore_v2', 'rfscore_v3'])
data = pd.read_csv(csvfile.name)
assert len(data.columns) == 4
assert 'nnscore' in data.columns
assert 'rfscore_v1' in data.columns
assert 'rfscore_v2' in data.columns
assert 'rfscore_v3' in data.columns
# remove files
for f in filenames:
os.unlink(f)
# remove symlinks
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if os.path.islink(version_dir):
os.unlink(version_dir)
'nnscore_descs_rdk.csv'),
delimiter=',')
# help debug errors
for i in range(descs.shape[1]):
mask = np.abs(descs[:, i] - descs_correct[:, i]) > 1e-4
if mask.sum() > 1:
print(i, gen.titles[i], mask.sum())
print(np.vstack((descs[mask, i], descs_correct[mask, i])))
assert_array_almost_equal(descs, descs_correct, decimal=4)
models = ([
PLECscore(n_jobs=1, version=v, size=2048) for v in ['linear', 'nn', 'rf']
] + [nnscore(n_jobs=1)] + [rfscore(version=v, n_jobs=1) for v in [1, 2, 3]])
@pytest.mark.parametrize('model', models)
def test_model_train(model):
mols = list(oddt.toolkit.readfile('sdf', actives_sdf))[:10]
list(map(lambda x: x.addh(), mols))
rec = next(oddt.toolkit.readfile('pdb', receptor_pdb))
rec.protein = True
rec.addh()
data_dir = os.path.join(test_data_dir, 'data')
home_dir = mkdtemp()
pdbbind_versions = (2007, 2013, 2016)
def init_rfscore(files):
global RFSCORES
for file in files:
RFSCORES.append(functions.rfscore().load(file))
def test_vs_scoring():
protein = next(oddt.toolkit.readfile('pdb', xiap_protein))
protein.protein = True
data_dir = os.path.join(test_data_dir, 'data')
home_dir = mkdtemp()
pdbbind_versions = (2007, 2013, 2016)
pdbbind_dir = os.path.join(data_dir, 'pdbbind')
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if not os.path.isdir(version_dir):
os.symlink(pdbbind_dir, version_dir)
filenames = []
# train mocked SFs
for model in [nnscore(n_jobs=1)] + [rfscore(version=v, n_jobs=1)
for v in [1, 2, 3]]:
model.gen_training_data(data_dir, pdbbind_versions=pdbbind_versions,
home_dir=home_dir)
filenames.append(model.train(home_dir=home_dir))
vs = virtualscreening(n_cpu=-1, chunksize=10)
vs.load_ligands('sdf', xiap_actives_docked)
# error if no protein is fed
with pytest.raises(ValueError):
vs.score('nnscore')
# bad sf name
with pytest.raises(ValueError):
vs.score('bad_sf', protein=protein)
vs.score('nnscore', protein=xiap_protein)
vs.score('nnscore_pdbbind2016', protein=protein)
vs.score('rfscore_v1', protein=protein)
vs.score('rfscore_v1_pdbbind2016', protein=protein)
vs.score('rfscore_v2', protein=protein)
vs.score('rfscore_v3', protein=protein)
vs.score('pleclinear', protein=protein)
vs.score('pleclinear_p5_l1_s65536_pdbbind2016', protein=protein)
# use pickle directly
vs.score(filenames[0], protein=protein)
# pass SF object directly
vs.score(scorer.load(filenames[0]), protein=protein)
# pass wrong object (sum is not an instance of scorer)
with pytest.raises(ValueError):
vs.score(sum, protein=protein)
mols = list(vs.fetch())
assert len(mols) == 100
mol_data = mols[0].data
assert 'nnscore' in mol_data
assert 'rfscore_v1' in mol_data
assert 'rfscore_v2' in mol_data
assert 'rfscore_v3' in mol_data
assert 'PLEClinear_p5_l1_s65536' in mol_data
vs = virtualscreening(n_cpu=-1, chunksize=10)
vs.load_ligands('sdf', xiap_actives_docked)
vs.score('nnscore', protein=protein)
vs.score('rfscore_v1', protein=protein)
vs.score('rfscore_v2', protein=protein)
vs.score('rfscore_v3', protein=protein)
with NamedTemporaryFile('w', suffix='.sdf') as molfile:
with NamedTemporaryFile('w', suffix='.csv') as csvfile:
vs.write('sdf', molfile.name, csv_filename=csvfile.name)
data = pd.read_csv(csvfile.name)
assert 'nnscore' in data.columns
assert 'rfscore_v1' in data.columns
assert 'rfscore_v2' in data.columns
assert 'rfscore_v3' in data.columns
mols = list(oddt.toolkit.readfile('sdf', molfile.name))
assert len(mols) == 100
vs.write_csv(csvfile.name, fields=['nnscore', 'rfscore_v1',
'rfscore_v2', 'rfscore_v3'])
data = pd.read_csv(csvfile.name)
assert len(data.columns) == 4
assert 'nnscore' in data.columns
assert 'rfscore_v1' in data.columns
assert 'rfscore_v2' in data.columns
assert 'rfscore_v3' in data.columns
# remove files
for f in filenames:
os.unlink(f)
# remove symlinks
for pdbbind_v in pdbbind_versions:
version_dir = os.path.join(data_dir, 'v%s' % pdbbind_v)
if os.path.islink(version_dir):
os.unlink(version_dir)
