def test__parallel_fingerprinter(self):
fps_from_sgl = \
CircularFPFeaturizer(n_jobs=1, fp_mode="binary_folded", output_format="dense").fit_transform(self.smis)
fps_from_par = \
CircularFPFeaturizer(n_jobs=4, fp_mode="binary_folded", output_format="dense").fit_transform(self.smis)
self.assertEqual((self.n_mols, 2048), fps_from_par.shape)
np.testing.assert_equal(fps_from_sgl, fps_from_par)
def test__determine_not_fitted_yet(self) -> None:
fprinter = CircularFPFeaturizer(only_freq_subs=True)
with self.assertRaises(NotFittedError):
len(fprinter)
with self.assertRaises(NotFittedError):
fprinter.transform(self.mols)
self.assertEqual(len(fprinter.fit(self.mols)), 86)
def test__string_output_format__only_freq_subs(self) -> None:
fprintr = CircularFPFeaturizer(output_format="sparse_string", only_freq_subs=True)
fps_str = fprintr.fit_transform(self.smis) # using SMILES
# Output shape
self.assertEqual(self.n_mols, len(fps_str))
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
for k in eval("{" + fps_str[i] + "}"):
self.assertTrue(0 <= k < len(fprintr))
def test__count_and_filter_hashes__ints_vs_floats(self):
d = [
{"A": 0, "B": 0, "C": 0, "D": 0},
{"A": 0 , "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{"A": 0, "B": 0, "C": 0, "D": 0},
{ "D": 0}
] # A = 3 / 8, B = 6 / 8 and C = 7 / 8
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]), 4)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 1.0)[0]), 1)
def test__error_when_parsing_smiles(self) -> None:
with self.assertRaises(RuntimeError):
CircularFPFeaturizer().fit_transform([
"O=C(O)C1OC(Oc2c(-c3ccc(O)c(O)c3)oc3cc(O)cc(O)c3c2=O)C(O)C(O)C1O",
"Oc1cc(O)c2c(c1)OC1(c3ccc(O)c(O)c3)Oc3cc(O)c4c(c3C2C1O)OC(c1ccc(O)c(O)c1)C(O)C4",
"COc1cc(O)c2c(=O)c(O)c(-c3ccc(O)c(O)c3)oc2c1",
"CaC1asfOC(O)C(O)C(O)C1O"])
def test__string_output_format(self) -> None:
fprintr = CircularFPFeaturizer(output_format="sparse_string")
fps_str = fprintr.fit_transform(self.smis) # using SMILES
# Output shape
self.assertEqual(self.n_mols, len(fps_str))
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
fps_ref = GetMorganFingerprint(mol, radius=fprintr.radius, useFeatures=fprintr.use_features_,
useChirality=fprintr.use_chirality, useCounts=fprintr.use_counts_)
fp_i_from_str = eval("{" + fps_str[i] + "}")
for hash, cnt in fps_ref.GetNonzeroElements().items():
self.assertEqual(fp_i_from_str[hash], cnt)
def test__to_dense_output(self) -> None:
# Output to large to be converted to a dense matrix
fpr_mat = CircularFPFeaturizer(fp_mode="binary_folded", output_format="dense", n_bits_folded=2048,
max_n_bits_for_dense_output=2048).fit_transform(self.mols)
self.assertFalse(isspmatrix_csr(fpr_mat))
self.assertTrue(isinstance(fpr_mat, np.ndarray))
# Output to large to be converted to a dense matrix
fpr_mat = CircularFPFeaturizer(fp_mode="binary_folded", output_format="dense", n_bits_folded=2048,
max_n_bits_for_dense_output=100).fit_transform(self.mols)
self.assertTrue(isspmatrix_csr(fpr_mat))
self.assertFalse(isinstance(fpr_mat, np.ndarray))
# Save-guard works for hashed fingerprints
fpr_mat = CircularFPFeaturizer(output_format="dense").fit_transform(self.mols)
self.assertTrue(isspmatrix_csr(fpr_mat))
self.assertFalse(isinstance(fpr_mat, np.ndarray))
def test__string_output_format__binary(self) -> None:
fprintr = CircularFPFeaturizer(output_format="sparse_string", fp_mode="binary_folded")
fps_str = fprintr.fit_transform(self.smis) # using SMILES
# Output shape
self.assertEqual(self.n_mols, len(fps_str))
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
fps_ref = GetMorganFingerprintAsBitVect(
mol, radius=fprintr.radius, useFeatures=fprintr.use_features_, useChirality=fprintr.use_chirality,
nBits=fprintr.n_bits_
)
fp_i_from_str = eval("{" + fps_str[i] + "}")
for idx in fps_ref.GetOnBits():
self.assertIn(idx, fp_i_from_str)
def test__hashed_counting_fingerprints__fcfp(self) -> None:
fprintr = CircularFPFeaturizer(fp_type="FCFP")
fps_mat_smi = fprintr.fit_transform(self.smis) # using SMILES
fps_mat_mol = fprintr.fit_transform(self.mols) # using Mol objects
# Output shape
self.assertEqual(fps_mat_smi.shape[0], self.n_mols)
self.assertEqual(fps_mat_smi.shape[1], fprintr.max_hash_value_)
self.assertEqual(fps_mat_mol.shape[0], self.n_mols)
self.assertEqual(fps_mat_mol.shape[1], fprintr.max_hash_value_)
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
fps_ref = GetMorganFingerprint(mol, radius=fprintr.radius, useFeatures=fprintr.use_features_,
useChirality=fprintr.use_chirality, useCounts=fprintr.use_counts_)
for hash, cnt in fps_ref.GetNonzeroElements().items():
self.assertEqual(fps_mat_smi[i, hash], cnt)
self.assertEqual(fps_mat_mol[i, hash], cnt)
def test__hashed_binary_fingerprints__ecfp(self) -> None:
fprintr = CircularFPFeaturizer(fp_mode="binary")
fps_mat_smi = fprintr.fit_transform(self.smis) # using SMILES
fps_mat_mol = fprintr.fit_transform(self.mols) # using Mol objects
# Output shape
self.assertEqual(fps_mat_smi.shape[0], self.n_mols)
self.assertEqual(fps_mat_smi.shape[1], fprintr.max_hash_value_)
self.assertEqual(fps_mat_mol.shape[0], self.n_mols)
self.assertEqual(fps_mat_mol.shape[1], fprintr.max_hash_value_)
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
fps_ref = GetMorganFingerprint(mol, radius=fprintr.radius, useFeatures=fprintr.use_features_,
useChirality=fprintr.use_chirality, useCounts=fprintr.use_counts_)
for hash in fps_ref.GetNonzeroElements():
self.assertTrue(fps_mat_smi[i, hash])
self.assertTrue(fps_mat_mol[i, hash])
# No other elements are set
self.assertEqual(np.sum(fps_mat_smi[i, :].data), len(fps_ref.GetNonzeroElements()))
self.assertEqual(np.sum(fps_mat_mol[i, :].data), len(fps_ref.GetNonzeroElements()))
def test__folded_binary_fingerprints__ecfp(self) -> None:
fprintr = CircularFPFeaturizer(fp_mode="binary_folded", n_bits_folded=512)
fps_mat_smi = fprintr.fit_transform(self.smis) # using SMILES
fps_mat_mol = fprintr.fit_transform(self.mols) # using Mol objects
# Output shape
self.assertEqual(fps_mat_smi.shape[0], self.n_mols)
self.assertEqual(fps_mat_smi.shape[1], fprintr.n_bits_folded)
self.assertEqual(fps_mat_mol.shape[0], self.n_mols)
self.assertEqual(fps_mat_mol.shape[1], fprintr.n_bits_folded)
# Fingerprint matrix structure
for i, mol in enumerate(self.mols):
fps_ref = GetMorganFingerprintAsBitVect(mol, radius=fprintr.radius, useFeatures=fprintr.use_features_,
useChirality=fprintr.use_chirality, nBits=fprintr.n_bits_folded)
on_bits = list(fps_ref.GetOnBits())
for j in range(fprintr.n_bits_folded):
if j in on_bits:
self.assertTrue(fps_mat_smi[i, j])
self.assertTrue(fps_mat_mol[i, j])
else:
self.assertFalse(fps_mat_smi[i, j])
self.assertFalse(fps_mat_mol[i, j])
def test__train_with_frequent_substrset(self) -> None:
# appears on ALL molecules
self.assertEqual(len(CircularFPFeaturizer(only_freq_subs=True, min_subs_freq=1.0).fit(self.mols)), 0)
# All data for fit and transform
n_old = np.inf
for freq in np.arange(0, 1.01, 0.01):
fprinter = CircularFPFeaturizer(only_freq_subs=True, min_subs_freq=freq).fit(self.mols)
# set of frequent pattern should get smaller when we require the patterns to be appear in more molecules
n_new = len(fprinter)
self.assertTrue(n_new <= n_old)
n_old = n_new
# Check dimension of transformed output
fps_mat = fprinter.transform(self.mols)
self.assertEqual(len(fprinter), fps_mat.shape[1])
# Check frequency of substructures in the output
freq_hash_set_inv = {v: k for k, v in fprinter.freq_hash_set_.items()}
for j in range(len(fprinter)):
h = freq_hash_set_inv[j]
self.assertTrue(np.sum(fps_mat[:, j].data > 0) / self.n_mols >= fprinter.hash_cnts_filtered_[h][1])
# Half of the data for fit and the other half for transform
n_old = np.inf
for freq in np.arange(0, 1.01, 0.01):
fprinter = CircularFPFeaturizer(only_freq_subs=True, min_subs_freq=freq).fit(self.mols[:7])
# set of frequent pattern should get smaller when we require the patterns to be appear in more molecules
n_new = len(fprinter)
self.assertTrue(n_new <= n_old)
n_old = n_new
# Check dimension of transformed output
fps_mat = fprinter.transform(self.mols[7:])
self.assertEqual(len(fprinter), fps_mat.shape[1])
def test__rdkit_parameters_are_correct(self) -> None:
fprintr = CircularFPFeaturizer(fp_type="ECFP", fp_mode="count").fit(None)
self.assertTrue(fprintr.use_counts_)
self.assertFalse(fprintr.use_features_)
fprintr = CircularFPFeaturizer(fp_type="FCFP", fp_mode="count").fit(None)
self.assertTrue(fprintr.use_counts_)
self.assertTrue(fprintr.use_features_)
fprintr = CircularFPFeaturizer(fp_type="ECFP", fp_mode="binary").fit(None)
self.assertFalse(fprintr.use_counts_)
self.assertFalse(fprintr.use_features_)
fprintr = CircularFPFeaturizer(fp_type="FCFP", fp_mode="binary").fit(None)
self.assertFalse(fprintr.use_counts_)
self.assertTrue(fprintr.use_features_)
fprintr = CircularFPFeaturizer(fp_type="ECFP", fp_mode="binary_folded").fit(None)
self.assertFalse(fprintr.use_counts_)
self.assertFalse(fprintr.use_features_)
fprintr = CircularFPFeaturizer(fp_type="FCFP", fp_mode="binary_folded").fit(None)
self.assertFalse(fprintr.use_counts_)
self.assertTrue(fprintr.use_features_)
def test__sklearn_clone(self):
fprinter = CircularFPFeaturizer()
_ = clone(fprinter)
def test__sklearn_get_params(self):
fprinter = CircularFPFeaturizer()
print(fprinter.get_params())
def test__count_and_filter_hashes__with_ints(self) -> None:
d = [
{"A": 0, "B": 0, "C": 0, "D": 0},
{"A": 0 , "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{"A": 0, "B": 0, "C": 0, "D": 0},
{ "D": 0}
] # A = 3 / 8, B = 6 / 8 and C = 7 / 8
# Must appear in at least 1 time --> result should contain all keys
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 1)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0].keys(), {"A", "B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0].keys(), {"A", "B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 4)[0]), 3)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4)[0].keys(), {"B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4)[0]["B"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4)[0]["C"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4)[0]["D"], 2)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 7)[0]), 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7)[0].keys(), {"C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7)[0]["C"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7)[0]["D"], 1)
"COc1cc(C=Cc2cc(O)cc(OC3OC(CO)C(O)C(O)C3O)c2)ccc1O",
"O=c1c(OC2OC(CO)C(O)C(O)C2O)c(-c2ccc(O)cc2)oc2cc(O)cc(O)c12",
"CC1OC(OCC2OC(Oc3c(-c4ccc(O)cc4)oc4cc(O)cc(O)c4c3=O)C(O)C(O)C2O)C(O)C(O)C1O",
"O=c1cc(-c2ccc(O)c(O)c2)oc2cc(OC3OC(CO)C(O)C(O)C3O)cc(O)c12",
"COC(=O)c1cc(O)c(O)c(O)c1",
"O=c1c(O)c(-c2cc(O)c(O)c(O)c2)oc2cc(O)cc(O)c12",
"O=C1CC(c2ccc(O)cc2)Oc2cc(O)cc(O)c21",
"O=C1CC(c2ccc(O)c(O)c2)c2c(cc(O)c3c2OC(c2ccc(O)c(O)c2)C(O)C3)O1",
"Oc1ccc(C=Cc2cc(O)cc3c2C(c2cc(O)cc(O)c2)C(c2ccc(O)cc2)O3)cc1",
"O=C1CC(c2ccc(O)cc2)Oc2cc(OC3OC(CO)C(O)C(O)C3O)cc(O)c21",
"O=C(O)C=Cc1ccccc1O",
"COc1cc(-c2oc3cc(O)cc4oc(=O)cc(c2OC2OC(CO)C(O)C(O)C2O)c34)cc(OC)c1O",
"Oc1ccc(C2c3c(O)cc(O)cc3C3C(c4ccc(O)cc4)c4c(O)cc(O)cc4C23)cc1",
"O=C(CCc1ccc(O)cc1)c1c(O)cc(O)cc1OC1OC(CO)C(O)C(O)C1O",
"Oc1cc(O)cc(C=Cc2ccc(O)c(O)c2)c1",
"Oc1cc(O)c2c(c1)OC(c1ccc(O)c(O)c1)C(O)C2c1c(O)cc(O)c2c1OC(c1ccc(O)c(O)c1)C(O)C2"]
# Get kernel matrix from fingerprints without substructure learning
fps_mat = CircularFPFeaturizer(fp_mode="count").transform(smis)
print("Is instance of 'csr_matrix': %d" % sp.isspmatrix_csr(fps_mat))
print(fps_mat.shape)
times = timeit.repeat(lambda: _min_max_sparse_csr(fps_mat, fps_mat, n_jobs=4), number=1, repeat=3)
print("min time:", np.min(times))
# Now with substructure learning
fps_mat = CircularFPFeaturizer(fp_mode="count", only_freq_subs=True, output_format=True).fit_transform(smis)
print("Is instance of 'csr_matrix': %d" % sp.isspmatrix_csr(fps_mat))
print(fps_mat.shape)
times = timeit.repeat(lambda: _min_max_dense(fps_mat, fps_mat), number=1, repeat=3)
print("min time:", np.min(times))
def test__count_and_filter_hashes(self) -> None:
d = [
{"A": 0, "B": 0, "C": 0, "D": 0},
{"A": 0 , "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{ "B": 0, "C": 0, "D": 0},
{"A": 0, "B": 0, "C": 0, "D": 0},
{ "D": 0}
] # A = 3 / 8, B = 6 / 8 and C = 7 / 8
# Must appear in at least 101% --> should result in an empty output
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 1.01)[0]), 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 1.01)[0], OrderedDict())
# Must appear in at least 0%
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0].keys(), {"A", "B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 0)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0].keys(), {"A", "B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 2 / 8)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0]), 4)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0].keys(), {"A", "B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0]["A"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0]["B"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0]["C"], 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 3 / 8)[0]["D"], 3)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 4 / 8)[0]), 3)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4 / 8)[0].keys(), {"B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4 / 8)[0]["B"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4 / 8)[0]["C"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 4 / 8)[0]["D"], 2)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 6 / 8)[0]), 3)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 6 / 8)[0].keys(), {"B", "C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 6 / 8)[0]["B"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 6 / 8)[0]["C"], 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 6 / 8)[0]["D"], 2)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 7 / 8)[0]), 2)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7 / 8)[0].keys(), {"C", "D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7 / 8)[0]["C"], 0)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 7 / 8)[0]["D"], 1)
self.assertEqual(len(CircularFPFeaturizer._count_and_filter_hashes(d, 8 / 8)[0]), 1)
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 8 / 8)[0].keys(), {"D"})
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes(d, 8 / 8)[0]["D"], 0)
# Empty input
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 0)[0], OrderedDict())
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 0.5)[0], OrderedDict())
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 1)[0], OrderedDict())
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 0)[1], OrderedDict())
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 0.5)[1], OrderedDict())
self.assertEqual(CircularFPFeaturizer._count_and_filter_hashes({}, 1)[1], OrderedDict())
