Loading deepchem/feat/__init__.py +3 −0 Original line number Diff line number Diff line Loading @@ -54,6 +54,9 @@ from deepchem.feat.material_featurizers import SineCoulombMatrix from deepchem.feat.material_featurizers import CGCNNFeaturizer from deepchem.feat.material_featurizers import ElemNetFeaturizer from deepchem.feat.atomic_conformation import AtomicConformation from deepchem.feat.atomic_conformation import AtomicConformationFeaturizer try: import transformers from transformers import BertTokenizer Loading deepchem/feat/atomic_conformation.py 0 → 100644 +175 −0 Original line number Diff line number Diff line import numpy as np from deepchem.feat.base_classes import Featurizer from deepchem.utils import get_partial_charge from typing import Sequence class AtomicConformation: """This class represents a collection of atoms arranged in 3D space. An instance of this class may represent any collection of atoms: a molecule, a fragment of a molecule, multiple interacting molecules, a material, etc. For each atom it stores a position and a list of scalar properties. Arbitrary properties are supported, but convenience methods are provided for accessing certain standard ones: atomic number, formal charge, and partial charge. Instances of this class are most often created by AtomicConformationFeaturizer. Attributes ---------- positions: ndarray the positions of all atoms in Angstroms, stored in an array of shape (N, 3) where N is the number of atoms properties: ndarray the property values for all atoms, stored in an array of shape (N, M) where N is the number of atoms and M is the number of properties property_names: ndarray an array of length M with the names of the properties """ def __init__(self, positions: np.ndarray, properties: np.ndarray, property_names: Sequence[str]): """Create an AtomicConformation for a set of atoms. Parameters ---------- positions: ndarray the positions of all atoms in Angstroms, stored in an array of shape (N, 3) where N is the number of atoms properties: ndarray the property values for all atoms, stored in an array of shape (N, M) where N is the number of atoms and M is the number of properties property_names: Sequence[str] the names of the properties """ self.positions = positions self.properties = properties self.property_names = np.array(property_names) @property def num_atoms(self) -> int: """Get the number of atoms in this object.""" return self.positions.shape[0] def get_property(self, name: str) -> np.ndarray: """Get a column of the properties array corresponding to a particular property. If there is no property with the specified name, this raises a ValueError. Parameters ---------- name: str the name of the property to get Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ indices = np.where(self.property_names == name)[0] if len(indices) == 0: raise ValueError("No property called '%s'" % name) return self.properties[:, indices[0]] @property def atomic_number(self) -> np.ndarray: """Get the column of the properties array containing atomic numbers. If there is no property with the name 'atomic number', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('atomic number') @property def formal_charge(self) -> np.ndarray: """Get the column of the properties array containing formal charges. If there is no property with the name 'formal charge', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('formal charge') @property def partial_charge(self) -> np.ndarray: """Get the column of the properties array containing partial charges. If there is no property with the name 'partial charge', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('partial charge') class AtomicConformationFeaturizer(Featurizer): """This featurizer represents each sample as an AtomicConformation object, representing a 3D arrangement of atoms. It expects each datapoint to be a string, which may be either a filename or a SMILES string. It is processed as follows. If the string ends in .pdb, .sdf, or .mol2, it is assumed to be a file in the corresponding format. The positions and elements of all atoms contained in the file are loaded. RDKit is used to compute formal and partial charges. Otherwise, it is assumed to be a SMILES string. RDKit is used to generate a 3D conformation and to compute formal and partial charges. """ def _featurize(self, datapoint: str) -> AtomicConformation: """Calculate features for a single datapoint. Parameters ---------- datapoint: str This is expected to be either a filename or a SMILES string. """ from rdkit import Chem from rdkit.Chem import AllChem if datapoint.endswith('.pdb'): mols = [Chem.MolFromPDBFile(datapoint, removeHs=False)] elif datapoint.endswith('.sdf'): supplier = Chem.SDMolSupplier(datapoint, removeHs=False) mols = [mol for mol in supplier] elif datapoint.endswith('.mol2'): mols = [Chem.MolFromMol2File(datapoint, removeHs=False)] else: mol = Chem.MolFromSmiles(datapoint) # SMILES is unique, so set a canonical order of atoms new_order = Chem.rdmolfiles.CanonicalRankAtoms(mol) mol = Chem.rdmolops.RenumberAtoms(mol, new_order) # Add hydrogens and generate a conformation. mol = Chem.AddHs(mol) AllChem.EmbedMolecule(mol, AllChem.ETKDGv3()) mols = [mol] # Record properties of the molecules. positions = [] properties = [] for mol in mols: positions.append(mol.GetConformer(0).GetPositions()) AllChem.ComputeGasteigerCharges(mol) for atom in mol.GetAtoms(): atomic_num = atom.GetAtomicNum() formal_charge = atom.GetFormalCharge() partial_charge = get_partial_charge(atom) properties.append([atomic_num, formal_charge, partial_charge]) # Create the output object. positions = np.concatenate(positions).astype(np.float32) properties = np.array(properties, dtype=np.float32) names = ['atomic number', 'formal charge', 'partial charge'] return AtomicConformation(positions, properties, names) deepchem/feat/tests/test_atomic_conformation.py 0 → 100644 +51 −0 Original line number Diff line number Diff line import deepchem as dc import numpy as np import os import unittest class TestAtomicConformation(unittest.TestCase): def test_featurize(self): current_dir = os.path.dirname(os.path.realpath(__file__)) sdf_file = os.path.join(current_dir, 'data', 'water.sdf') pdb_file = os.path.join(current_dir, 'data', '3zso_ligand_hyd.pdb') smiles = 'CCC' featurizer = dc.feat.AtomicConformationFeaturizer() features = featurizer.featurize([sdf_file, pdb_file, smiles]) assert len(features) == 3 # Check the SDF file. assert features[0].num_atoms == 60 assert features[0].atomic_number[0] == 8 assert features[0].atomic_number[1] == 1 assert np.all(features[0].formal_charge == 0) for i in range(60): assert (features[0].partial_charge[i] < 0) == (i % 3 == 0) # Check the PDB file. assert features[1].num_atoms == 47 assert features[1].atomic_number[0] == 6 assert features[1].atomic_number[35] == 7 assert features[1].atomic_number[46] == 1 for i in range(47): if i == 36: assert features[1].formal_charge[i] == 1 else: assert features[1].formal_charge[i] == 0 if features[1].atomic_number[i] in (7, 8): # N and O should be negative assert features[1].partial_charge[i] < 0 elif features[1].atomic_number[i] == 1: # H should be positive assert features[1].partial_charge[i] > 0 # Check the SMILES string. assert features[2].num_atoms == 11 assert sum(features[2].atomic_number == 6) == 3 assert sum(features[2].atomic_number == 1) == 8 assert np.all(features[2].formal_charge == 0) for i in range(11): assert (features[2].partial_charge[i] < 0) == (features[2].atomic_number[i] == 6) Loading
deepchem/feat/__init__.py +3 −0 Original line number Diff line number Diff line Loading @@ -54,6 +54,9 @@ from deepchem.feat.material_featurizers import SineCoulombMatrix from deepchem.feat.material_featurizers import CGCNNFeaturizer from deepchem.feat.material_featurizers import ElemNetFeaturizer from deepchem.feat.atomic_conformation import AtomicConformation from deepchem.feat.atomic_conformation import AtomicConformationFeaturizer try: import transformers from transformers import BertTokenizer Loading
deepchem/feat/atomic_conformation.py 0 → 100644 +175 −0 Original line number Diff line number Diff line import numpy as np from deepchem.feat.base_classes import Featurizer from deepchem.utils import get_partial_charge from typing import Sequence class AtomicConformation: """This class represents a collection of atoms arranged in 3D space. An instance of this class may represent any collection of atoms: a molecule, a fragment of a molecule, multiple interacting molecules, a material, etc. For each atom it stores a position and a list of scalar properties. Arbitrary properties are supported, but convenience methods are provided for accessing certain standard ones: atomic number, formal charge, and partial charge. Instances of this class are most often created by AtomicConformationFeaturizer. Attributes ---------- positions: ndarray the positions of all atoms in Angstroms, stored in an array of shape (N, 3) where N is the number of atoms properties: ndarray the property values for all atoms, stored in an array of shape (N, M) where N is the number of atoms and M is the number of properties property_names: ndarray an array of length M with the names of the properties """ def __init__(self, positions: np.ndarray, properties: np.ndarray, property_names: Sequence[str]): """Create an AtomicConformation for a set of atoms. Parameters ---------- positions: ndarray the positions of all atoms in Angstroms, stored in an array of shape (N, 3) where N is the number of atoms properties: ndarray the property values for all atoms, stored in an array of shape (N, M) where N is the number of atoms and M is the number of properties property_names: Sequence[str] the names of the properties """ self.positions = positions self.properties = properties self.property_names = np.array(property_names) @property def num_atoms(self) -> int: """Get the number of atoms in this object.""" return self.positions.shape[0] def get_property(self, name: str) -> np.ndarray: """Get a column of the properties array corresponding to a particular property. If there is no property with the specified name, this raises a ValueError. Parameters ---------- name: str the name of the property to get Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ indices = np.where(self.property_names == name)[0] if len(indices) == 0: raise ValueError("No property called '%s'" % name) return self.properties[:, indices[0]] @property def atomic_number(self) -> np.ndarray: """Get the column of the properties array containing atomic numbers. If there is no property with the name 'atomic number', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('atomic number') @property def formal_charge(self) -> np.ndarray: """Get the column of the properties array containing formal charges. If there is no property with the name 'formal charge', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('formal charge') @property def partial_charge(self) -> np.ndarray: """Get the column of the properties array containing partial charges. If there is no property with the name 'partial charge', this raises a ValueError. Returns ------- a numpy array containing the requested column of the properties array. This is a 1D array of length num_atoms. """ return self.get_property('partial charge') class AtomicConformationFeaturizer(Featurizer): """This featurizer represents each sample as an AtomicConformation object, representing a 3D arrangement of atoms. It expects each datapoint to be a string, which may be either a filename or a SMILES string. It is processed as follows. If the string ends in .pdb, .sdf, or .mol2, it is assumed to be a file in the corresponding format. The positions and elements of all atoms contained in the file are loaded. RDKit is used to compute formal and partial charges. Otherwise, it is assumed to be a SMILES string. RDKit is used to generate a 3D conformation and to compute formal and partial charges. """ def _featurize(self, datapoint: str) -> AtomicConformation: """Calculate features for a single datapoint. Parameters ---------- datapoint: str This is expected to be either a filename or a SMILES string. """ from rdkit import Chem from rdkit.Chem import AllChem if datapoint.endswith('.pdb'): mols = [Chem.MolFromPDBFile(datapoint, removeHs=False)] elif datapoint.endswith('.sdf'): supplier = Chem.SDMolSupplier(datapoint, removeHs=False) mols = [mol for mol in supplier] elif datapoint.endswith('.mol2'): mols = [Chem.MolFromMol2File(datapoint, removeHs=False)] else: mol = Chem.MolFromSmiles(datapoint) # SMILES is unique, so set a canonical order of atoms new_order = Chem.rdmolfiles.CanonicalRankAtoms(mol) mol = Chem.rdmolops.RenumberAtoms(mol, new_order) # Add hydrogens and generate a conformation. mol = Chem.AddHs(mol) AllChem.EmbedMolecule(mol, AllChem.ETKDGv3()) mols = [mol] # Record properties of the molecules. positions = [] properties = [] for mol in mols: positions.append(mol.GetConformer(0).GetPositions()) AllChem.ComputeGasteigerCharges(mol) for atom in mol.GetAtoms(): atomic_num = atom.GetAtomicNum() formal_charge = atom.GetFormalCharge() partial_charge = get_partial_charge(atom) properties.append([atomic_num, formal_charge, partial_charge]) # Create the output object. positions = np.concatenate(positions).astype(np.float32) properties = np.array(properties, dtype=np.float32) names = ['atomic number', 'formal charge', 'partial charge'] return AtomicConformation(positions, properties, names)
deepchem/feat/tests/test_atomic_conformation.py 0 → 100644 +51 −0 Original line number Diff line number Diff line import deepchem as dc import numpy as np import os import unittest class TestAtomicConformation(unittest.TestCase): def test_featurize(self): current_dir = os.path.dirname(os.path.realpath(__file__)) sdf_file = os.path.join(current_dir, 'data', 'water.sdf') pdb_file = os.path.join(current_dir, 'data', '3zso_ligand_hyd.pdb') smiles = 'CCC' featurizer = dc.feat.AtomicConformationFeaturizer() features = featurizer.featurize([sdf_file, pdb_file, smiles]) assert len(features) == 3 # Check the SDF file. assert features[0].num_atoms == 60 assert features[0].atomic_number[0] == 8 assert features[0].atomic_number[1] == 1 assert np.all(features[0].formal_charge == 0) for i in range(60): assert (features[0].partial_charge[i] < 0) == (i % 3 == 0) # Check the PDB file. assert features[1].num_atoms == 47 assert features[1].atomic_number[0] == 6 assert features[1].atomic_number[35] == 7 assert features[1].atomic_number[46] == 1 for i in range(47): if i == 36: assert features[1].formal_charge[i] == 1 else: assert features[1].formal_charge[i] == 0 if features[1].atomic_number[i] in (7, 8): # N and O should be negative assert features[1].partial_charge[i] < 0 elif features[1].atomic_number[i] == 1: # H should be positive assert features[1].partial_charge[i] > 0 # Check the SMILES string. assert features[2].num_atoms == 11 assert sum(features[2].atomic_number == 6) == 3 assert sum(features[2].atomic_number == 1) == 8 assert np.all(features[2].formal_charge == 0) for i in range(11): assert (features[2].partial_charge[i] < 0) == (features[2].atomic_number[i] == 6)
