def test9ToNumpy(self):
import numpy
for typ in (DataStructs.ExplicitBitVect, ):
bv = typ(32)
bv.SetBit(0)
bv.SetBit(1)
bv.SetBit(17)
bv.SetBit(23)
bv.SetBit(31)
arr = numpy.zeros((32, ), 'i')
DataStructs.ConvertToNumpyArray(bv, arr)
for i in range(bv.GetNumBits()):
self.assertEqual(bv[i], arr[i])
for typ in (DataStructs.IntSparseIntVect,
DataStructs.LongSparseIntVect,
DataStructs.UIntSparseIntVect,
DataStructs.ULongSparseIntVect):
iv = typ(32)
iv[0] = 1
iv[1] = 1
iv[17] = 1
iv[23] = 1
iv[31] = 1
arr = numpy.zeros((32, ), 'i')
DataStructs.ConvertToNumpyArray(iv, arr)
for i in range(iv.GetLength()):
self.assertEqual(iv[i], arr[i])
def rf_validate(enzymes, success, candidate):
# generate fingeprints: Morgan fingerprint with radius 2
mols = []
for i in enzymes:
mols.append( Chem.MolFromSmiles(i) )
fps = [AllChem.GetMorganFingerprintAsBitVect(m, 2) for m in mols]
# convert the RDKit explicit vectors into numpy arrays
np_fps = []
for fp in fps:
arr = numpy.zeros((1,))
DataStructs.ConvertToNumpyArray(fp, arr)
np_fps.append(arr)
# get a random forest classifiert with 100 trees
rf = RandomForestClassifier(n_estimators=100, random_state=1123)
# train the random forest
# with the first two molecules being actives (class 1) and
# the last two being inactives (class 0)
ys_fit = success
rf.fit(np_fps, ys_fit)
# use the random forest to predict a new molecule
m6 = Chem.MolFromSmiles(candidate)
fp = numpy.zeros((1,))
DataStructs.ConvertToNumpyArray(AllChem.GetMorganFingerprintAsBitVect(m6, 2), fp)
return rf.predict_proba(fp)[0][1]
def create_rxn_Morgan2FP(rsmi,
psmi,
rxnfpsize=2048,
pfpsize=2048,
useFeatures=False,
calculate_rfp=True):
"""Create a rxn Morgan (r=2) fingerprint as bit vector from SMILES string lists of reactants and products"""
# Modified from Schneider's code (2014)
if calculate_rfp is True:
rsmi = rsmi.encode('utf-8')
try:
mol = Chem.MolFromSmiles(rsmi)
except Exception as e:
return
try:
fp_bit = AllChem.GetMorganFingerprintAsBitVect(mol=mol,
radius=2,
nBits=rxnfpsize,
useFeatures=False,
useChirality=True)
fp = np.empty(rxnfpsize, dtype=np.bool)
DataStructs.ConvertToNumpyArray(fp_bit, fp)
except Exception as e:
print("Cannot build reactant fp due to {}".format(e))
return
rfp = fp
else:
rfp = None
psmi = psmi.encode('utf-8')
try:
mol = Chem.MolFromSmiles(psmi)
except Exception as e:
print(psmi)
return
try:
fp_bit = AllChem.GetMorganFingerprintAsBitVect(mol=mol,
radius=2,
nBits=pfpsize,
useFeatures=False,
useChirality=True)
fp = np.empty(pfpsize, dtype=np.bool)
DataStructs.ConvertToNumpyArray(fp_bit, fp)
except Exception as e:
print("Cannot build product fp due to {}".format(e))
return
pfp = fp
return [pfp, rfp]
def mol_train_test(dataset,
labels,
test_size=0.1,
random_state=2019,
nbits=1024):
# TAKING WRONG INCHIS
all_mols = [
Chem.MolFromSmiles(SMILES_string)
for SMILES_string in dataset['SMILES']
]
drop_index = [i for i, mol in enumerate(all_mols)
if mol == None] # FINDING WRONG INCHIS
# DROP FROM MOLS, lABELS, AND DATASET
if len(drop_index) != 0:
labels = labels.drop(drop_index).reset_index(drop=True)
dataset = dataset.drop(drop_index).reset_index(drop=True)
all_mols = [
Chem.MolFromSmiles(SMILES_string)
for SMILES_string in dataset['SMILES']
] ### FIND BETTER WAY TO NOT CALCULATE AGAIN!!!!
# TRAIN-TEST SPLITS
train_mols, test_mols, y_train, y_test = train_test_split(all_mols, labels, test_size=test_size\
, random_state=random_state)
# CONVERT TRAINING MOLECULES INTO FINGERPRINT AS 256BITS VECTORS
bi = {}
fps = [rdMolDescriptors.GetMorganFingerprintAsBitVect(m, radius=2, bitInfo= bi, nBits=nbits) \
for m in train_mols]
# PUT ALL EACH OF THE CORRESPONDING 256BITS FINGERPRINTS INTO A LIST
train_fps_array = []
for fp in fps:
arr = np.zeros((1, ), dtype=int)
DataStructs.ConvertToNumpyArray(fp, arr)
train_fps_array.append(arr)
# CONVERT InChi STRINGS INTO MOLECULES FOR TEST DATA
test_fps = [rdMolDescriptors.GetMorganFingerprintAsBitVect(test_m, radius=2, bitInfo= bi, nBits=nbits) \
for test_m in test_mols]
#Convert testing fingerprints into binary, and put all testing binaries into arrays
test_np_fps_array = []
for test_fp in test_fps:
test_arr = np.zeros((1, ), dtype=int)
DataStructs.ConvertToNumpyArray(test_fp, test_arr)
test_np_fps_array.append(test_arr)
return dataset, labels, all_mols, y_train, y_test, train_fps_array, test_np_fps_array
def SMILES_2_ECFP(smiles, radius=3, bit_len=4096, index=None):
"""
This function transforms a list of SMILES strings into a list of ECFP with
radius 3.
----------
smiles: List of SMILES strings to transform
Returns
-------
This function return the SMILES strings transformed into a vector of 4096 elements
"""
fps = np.zeros((len(smiles), bit_len))
for i, smile in enumerate(smiles):
mol = Chem.MolFromSmiles(smile)
arr = np.zeros((1, ))
try:
mol = MurckoScaffold.GetScaffoldForMol(mol)
fp = AllChem.GetMorganFingerprintAsBitVect(mol,
radius,
nBits=bit_len)
DataStructs.ConvertToNumpyArray(fp, arr)
fps[i, :] = arr
except:
print(smile)
fps[i, :] = [0] * bit_len
return pd.DataFrame(fps, index=(smiles if index is None else index))
def calculate_ecfp4(mol, nBits=1024):
rdmol = to_rdkit_Mol(mol)
Chem.rdmolops.SanitizeMol(rdmol)
fp = Chem.rdMolDescriptors.GetMorganFingerprintAsBitVect(rdmol, 2, nBits=nBits)
arr = np.array([])
DataStructs.ConvertToNumpyArray(fp, arr)
return arr
def __call__(self, smiles, radius=3, bit_len=4096, scaffold=0):
fps = np.zeros((len(smiles), bit_len))
for i, smile in enumerate(smiles):
mol = Chem.MolFromSmiles(smile)
arr = np.zeros((1, ))
try:
if scaffold == 1:
mol = MurckoScaffold.GetScaffoldForMol(mol)
elif scaffold == 2:
mol = MurckoScaffold.MakeScaffoldGeneric(mol)
if not mol:
raise Exception(
f'Failed to calculate Morgan fingerprint (creating RDKit instance from smiles failed: {smile})'
)
fp = AllChem.GetMorganFingerprintAsBitVect(mol,
radius,
nBits=bit_len)
DataStructs.ConvertToNumpyArray(fp, arr)
fps[i, :] = arr
except Exception as exp:
# TODO: use a more specific exception related to descriptor errors
# traceback.print_exc()
self.builder.errors.append(exp)
fps[i, :] = [0] * bit_len
return pd.DataFrame(fps)
def generate_fingerprints(smile):
mol = MolFromSmiles(smile)
fp = AllChem.GetMorganFingerprintAsBitVect(mol, radius=3, nBits=1024)
array = np.zeros((0,), dtype=np.int8)
DataStructs.ConvertToNumpyArray(fp, array)
#print(array)
return array
def main():
"""Run the main function."""
outfile = open(sys.argv[1] + "_predictions.txt", 'w')
outfile.write("Chemical Name\tPrediction\n")
model = load_model("pparg_ligand_model.h5")
dataframe = pandas.read_csv(sys.argv[1], sep="\t")
mols = []
fps = []
for index, row in dataframe.iterrows():
mol = Chem.MolFromSmiles(row['SMILES'])
fp = AllChem.GetMorganFingerprintAsBitVect(mol, 2)
mols.append(mol)
fps.append(fp)
np_fps = []
for fp in fps:
arr = numpy.zeros((1, ))
DataStructs.ConvertToNumpyArray(fp, arr)
np_fps.append(arr)
np_fps_array = numpy.array(np_fps)
predictions = model.predict(np_fps_array, batch_size=5)
i = 0
for prediction in predictions:
y_prediction = ''
if (prediction < 0.50):
y_prediction = "ligand"
else:
y_prediction = "not_ligand"
outfile.write(dataframe['Chemical_Name'][i] + "\t" + y_prediction +
"\n")
i += 1
outfile.close()
def morgan_fingp(fname):
nbits = 1024
radius = 2
#fp = []
fsplit = fname.split('/')[-1]
#to_skip = done_dict[fsplit]
ref2 = open(fp + '/' + fn + '/' + fsplit, 'a')
#print(fname,to_skip)
with open(fname, 'r') as ref:
ref.readline()
#for count in range(to_skip):
# ref.readline()
for line in ref:
smile, zin_id = line.rstrip().split()
arg = np.zeros((1, ))
try:
DataStructs.ConvertToNumpyArray(
AllChem.GetMorganFingerprintAsBitVect(
Chem.MolFromSmiles(smile), radius, nBits=nbits), arg)
ref2.write(
(',').join([zin_id] +
[str(elem) for elem in np.where(arg == 1)[0]]))
ref2.write('\n')
except:
print(line)
pass
def fp_matrix(self, fp):
matrix_fp = []
for f in fp:
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(f, arr)
matrix_fp.append(arr)
return matrix_fp
def computeFP(x):
#compute depth-2 morgan fingerprint hashed to 2048 bits
fp = Chem.GetMorganFingerprintAsBitVect(x, 2, nBits=2048)
res = numpy.zeros(len(fp), numpy.int32)
#convert the fingerprint to a numpy array and wrap it into the dummy container
DataStructs.ConvertToNumpyArray(fp, res)
return FP(res)
def compute_morgan_fingerprints(smiles, fingerprint_length, fingerprint_radius):
"""Get Morgan Fingerprint of a specific SMILES string.
Adapted from: <https://github.com/google-research/google-research/blob/
dfac4178ccf521e8d6eae45f7b0a33a6a5b691ee/mol_dqn/chemgraph/dqn/deep_q_networks.py#L750>
Args:
smiles: String. The SMILES string of the molecule.
fingerprint_length (int): Bit-length of fingerprint
fingerprint_radius (int): Radius used to compute fingerprint
Returns:
np.array. shape = [hparams.fingerprint_length]. The Morgan fingerprint.
"""
if smiles is None: # No smiles string
return np.zeros((fingerprint_length,))
molecule = Chem.MolFromSmiles(smiles)
if molecule is None: # Invalid smiles string
return np.zeros((fingerprint_length,))
# Compute the fingerprint
fingerprint = AllChem.GetMorganFingerprintAsBitVect(
molecule, fingerprint_radius, fingerprint_length)
arr = np.zeros((1,))
# ConvertToNumpyArray takes ~ 0.19 ms, while
# np.asarray takes ~ 4.69 ms
DataStructs.ConvertToNumpyArray(fingerprint, arr)
return arr
def Xyfromdf(df, return_y):
"""Generate X (design matrix) and y (labels) for training a machine learning model from a dataframe of
structures. The structures are encoded using their Morgan fingerprint.
Args:
df: a Pandas DataFrame with columns "Compound ID", "Structure", "IC50"
return_y: whether or not to return labels y.
Returns:
2-tuple (X,y) where X is a dataframe and y is a series if return_y is True. Otherwise just X
"""
# generate fingeprints: Morgan fingerprint with radius 2
fps = [
AllChem.GetMorganFingerprintAsBitVect(Chem.MolFromSmiles(smile), 2)
for smile in df["Structure"]
]
# convert the RDKit explicit vectors into numpy arrays
np_fps = [np.zeros((1, )) for fp in fps]
for i, fp in enumerate(fps):
DataStructs.ConvertToNumpyArray(fp, np_fps[i])
X = pd.DataFrame(np.array(np_fps))
if return_y:
y = np.log(df["IC50"])
assert y.isna().sum() == 0
return X, y
else:
return X
def morgan(self, radius, size=None):
"""
Calculates circular fingerprints and renders them as a DataFrame, so that it is easier to handle.
: radius (int): radius = 2 ~ ECFP4, radius = 3 ~ ECFP6
: size (int, optional): number of bits to generate. If None, it will be assigned
the standard value of 2048
"""
if size is None:
size = 2048
fps = [
AllChem.GetMorganFingerprintAsBitVect(m, radius, size)
for m in self.smiles_converted
]
np_fps = []
for fp in fps:
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(fp, arr)
np_fps.append(arr)
df = pd.DataFrame(np_fps)
return df
def get_morgan_fp(self,
mol: Mol,
radius: int = 2,
nBits: int = 1024,
invariants: List[AtomPairsParameters] = [],
fromAtoms: List[AtomPairsParameters] = [],
useChirality: bool = False,
useBondTypes: bool = True,
useFeatures: bool = False,
bitInfo: AtomPairsParameters = {},
includeRedundantEnvironments: bool = False) -> array:
"""
Function to generate a set of fingerprints from a single molecule
Parameters:
Same parameters as: https://www.rdkit.org/docs/source/rdkit.Chem.rdMolDescriptors.html#rdkit.Chem.rdMolDescriptors.GetMorganFingerprintAsBitVect
Returns:
fingerprints_vector (array): numpy array containing fingerprints for the molecule
"""
arr = np.zeros((1, ))
fp = AllChem.GetMorganFingerprintAsBitVect(mol=mol,
radius=radius,
nBits=nBits,
invariants=invariants,
fromAtoms=fromAtoms,
useChirality=useChirality,
useBondTypes=useBondTypes,
useFeatures=useFeatures,
bitInfo=bitInfo)
DataStructs.ConvertToNumpyArray(fp, arr)
arr = np.array(
[len(bitInfo[x]) if x in bitInfo else 0 for x in range(nBits)])
return arr
def convert_bitvec_to_array(bitvec: list) -> np.ndarray:
"""Convert bit vector fingerprint to numpy array."""
features = np.zeros(1, )
DataStructs.ConvertToNumpyArray(bitvec, features)
return features
def fingerprints_from_mol(cls, mol): # use ECFP4
features_vec = AllChem.GetMorganFingerprintAsBitVect(mol,
2,
nBits=2048)
features = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(features_vec, features)
return features.reshape(1, -1)
def featurizer(folder, filename):
folderpath = folder
data = pd.read_csv(folderpath + filename, sep=',')
alias = []
ic50 = []
fps = []
#convert smiles to rkd objects
for i in range(data.shape[0]):
try:
arr = np.zeros((1, ))
compound = Chem.MolFromSmiles(data.ix[i, 1])
fp = AllChem.GetMorganFingerprintAsBitVect(compound, 3, 4096)
DataStructs.ConvertToNumpyArray(fp, arr)
fps.append(arr)
alias.append(data.ix[i, 0])
ic50.append(data.ix[i, 2])
except:
print(i)
print(data.ix[i, 0])
#create dataframe to store fingerprinters and write to csv file
df = pd.DataFrame(fps, index=alias)
df.insert(0, "ic50", ic50)
df.to_csv(folderpath + 'fingerprinters_4096.csv')
def getNumpy(inlist):
outlist = []
for i in inlist:
arr = numpy.zeros((3, ), tree.DTYPE)
DataStructs.ConvertToNumpyArray(i[1], arr)
outlist.append(arr)
return outlist
def get_circular_fp(smile, radius=6, fp_len=128):
mol = Chem.MolFromSmiles(smile)
fingerprint = Chem.AllChem.GetMorganFingerprintAsBitVect(
mol, radius, fp_len)
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(fingerprint, arr)
return arr
def smile2fp(smile: str) -> Any:
"""
Calculates one fingerprint from a SMILE
:param smile: Input SMILE
:return: List of bits if conversion is successfull,
None otherwise
"""
# generate morgan fp (circular, ecfp)
# smile = df['smiles'][1]
# mol = Chem.MolFromSmiles(smile)
# from rdkit.Chem import AllChem
# morgan = AllChem.GetMorganFingerprintAsBitVect(mol, 2)
# npa = np.zeros((0,), dtype=np.bool)
# from rdkit import DataStructs
# DataStructs.ConvertToNumpyArray(morgan, npa)
npa = np.zeros((0,), dtype=np.bool)
try:
DataStructs.ConvertToNumpyArray(
AllChem.GetMorganFingerprintAsBitVect(Chem.MolFromSmiles(smile), 2, nBits=fp_size),
npa)
return npa
except:
return None
def main(smiles, output, top = 10, model = '10uM'):
if model == '1uM':
morgan_nb = joblib.load('Data/models_23/1uM/mNB_1uM_all.pkl')
else:
morgan_nb = joblib.load('Data/models_23/10uM/mNB_10uM_all.pkl')
classes = list(morgan_nb.targets)
mol = Chem.MolFromSmiles(smiles)
fp = AllChem.GetMorganFingerprintAsBitVect(mol, 2, nBits = 2048)
res = np.zeros(len(fp), np.int32)
DataStructs.ConvertToNumpyArray(fp, res)
probas = list(morgan_nb.predict_proba(res.reshape(1,-1))[0])
predictions = pd.DataFrame(list(zip(classes, probas)), columns=['id','probas'])
top_pred = predictions.sort_values(by='probas', ascending = False).head(top)
plist = []
for i, e in enumerate(top_pred['id']):
plist.append(fetch_WS(e))
target_info = pd.DataFrame(plist, columns=['id', 'name', 'organism'])
result = pd.merge(top_pred, target_info)
result.to_csv(output)
def rdkit_numpy_convert(fp):
output = []
for f in fp:
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(f, arr)
output.append(arr)
return np.asarray(output)
def morgan_fingerprint(
df: pd.DataFrame,
mols_col: str,
radius: int = 3,
nbits: int = 2048,
kind: str = "counts",
):
"""
Convert a column of RDKIT Mol objects into Morgan Fingerprints.
Returns a new dataframe without any of the original data. This is
intentional, as Morgan fingerprints are usually high-dimensional
features.
Method chaining usage:
.. code-block:: python
df = pd.DataFrame(...)
morgans = df.morgan_fingerprint(mols_col='mols', radius=3, nbits=2048)
If you wish to join the Morgans back into the original dataframe, this
can be accomplished by doing a `join`, becuase the indices are
preserved:
..code-block:: python
joined = df.join(morgans)
:param df: A pandas DataFrame.
:param mols_col: The name of the column that has the RDKIT mol objects
:param radius: Radius of Morgan fingerprints. Defaults to 3.
:param nbits: The length of the fingerprints. Defaults to 2048.
:param kind: Whether to return counts or bits. Defaults to counts.
:returns: A pandas DataFrame
"""
acceptable_kinds = ["counts", "bits"]
if kind not in acceptable_kinds:
raise ValueError(f"`kind` must be one of {acceptable_kinds}")
if kind == "bits":
fps = [
GetMorganFingerprintAsBitVect(m, radius, nbits)
for m in df[mols_col]
]
elif kind == "counts":
fps = [
GetHashedMorganFingerprint(m, radius, nbits) for m in df[mols_col]
]
np_fps = []
for fp in fps:
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(fp, arr)
np_fps.append(arr)
np_fps = np.vstack(np_fps)
fpdf = pd.DataFrame(np_fps)
fpdf.index = df.index
return fpdf
def fp_to_pandas(fp, drug_names):
fp_np = []
for fp in fp:
arr = np.zeros((1,))
DataStructs.ConvertToNumpyArray(fp, arr)
fp_np.append(arr)
fp_df = pd.DataFrame(fp_np, index=drug_names)
return fp_df
def fingerprintsToNPArr(fps):
# print fps
np_fps = []
for fp in fps:
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(fp, arr)
np_fps.append(arr)
return np_fps
def get_numpy_fingerprint_from_smiles(smiles):
"""Get Morgan Fingerprint as NumPy vector from SMILES string"""
mol = Chem.MolFromSmiles(smiles)
fingerprint = AllChem.GetMorganFingerprintAsBitVect(mol, 3)
finger_container = np.empty(fingerprint.GetNumBits())
DataStructs.ConvertToNumpyArray(fingerprint, finger_container)
return finger_container
def _maccs_keys(self, molecules: List, parameters: {}):
fingerprints = []
fps = [MACCSkeys.GenMACCSKeys(mol) for mol in molecules]
for fp in fps:
fp_np = np.zeros((1, ), dtype=np.int32)
DataStructs.ConvertToNumpyArray(fp, fp_np)
fingerprints.append(fp_np)
return fingerprints
def bitvect_as_np_array(self):
"""Transforms the calculated 2048-bit bitvector into a np array.
Returns a 1 x 2048 array.
"""
import numpy as np
arr = np.zeros((1, ))
DataStructs.ConvertToNumpyArray(self.bitvect, arr)
return arr
