Loading deepchem/feat/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -27,3 +27,4 @@ from deepchem.feat.adjacency_fingerprints import AdjacencyFingerprint from deepchem.feat.smiles_featurizers import SmilesToSeq, SmilesToImage from deepchem.feat.materials_featurizers import ElementPropertyFingerprint, SineCoulombMatrix, StructureGraphFeaturizer from deepchem.feat.contact_fingerprints import ContactCircularFingerprint from deepchem.feat.contact_fingerprints import ContactCircularVoxelizer deepchem/feat/atomic_coordinates.py +9 −0 Original line number Diff line number Diff line Loading @@ -237,6 +237,11 @@ class AtomicConvFeaturizer(ComplexFeaturizer): z-matrix for the molecule which is an array of shape (N_atoms,) that contains the atomic number of that atom. If `reduce_to_contacts` is set to `True`, only those atoms which are in a contact region of the molecular complex are considered. This can dramatically reduce the number of atoms that need to be featurized so this feature is set to True by default. Since the featurization computes these three quantities for each of the two molecules and the complex, a total of 9 quantities are returned for each complex. Note that for Loading Loading @@ -310,6 +315,10 @@ class AtomicConvFeaturizer(ComplexFeaturizer): if self.reduce_to_contacts: fragments = reduce_molecular_complex_to_contacts(fragments, self.cutoff) coords = [frag[0] for frag in fragments] ################################################ print("[coord.shape for coord in coords]") print([coord.shape for coord in coords]) ################################################ mols = [frag[1] for frag in fragments] #system_mol = rdkit_util.merge_molecules(mols) system_mol = rdkit_util.merge_molecular_fragments(mols) Loading deepchem/feat/contact_fingerprints.py +48 −37 Original line number Diff line number Diff line Loading @@ -90,14 +90,6 @@ class ContactCircularFingerprint(ComplexFeaturizer): """ Compute featurization for a single mol/protein complex TODO(rbharath): This is very not ergonomic. I'd much prefer returning an vector instead of a list of two vectors. In addition, there's a question of efficiency. RdkitGridFeaturizer caches rotated versions etc internally. To make things work out of box, we are accepting that kludgey input. This needs to be cleaned up before full merge. Parameters ---------- molecular_complex: Object Loading Loading @@ -171,20 +163,35 @@ class ContactCircularVoxelizer(ComplexFeaturizer): self.voxel_width = voxel_width self.voxels_per_edge = int(self.box_width / self.voxel_width) def _featurize_complex(self, mol, protein): def _featurize_complex(self, molecular_complex): """ Compute featurization for a single mol/protein complex Parameters ---------- mol: object Representation of the molecule protein: object Representation of the protein molecular_complex: Object A representation of a molecular complex, produced by `rdkit_util.load_complex`. """ (lig_xyz, lig_rdk), (prot_xyz, prot_rdk) = mol, protein distances = compute_pairwise_distances(prot_xyz, lig_xyz) return [ # TODO(rbharath): This is a little tricky in the generalized # regime, but we need to find a way to compute the centroid. My # idea is that we can compute the centroid of the contact # atoms.and use this to recenter all the fragments. try: fragments = rdkit_util.load_complex(molecular_complex, add_hydrogens=False) except MoleculeLoadException: logger.warning("This molecule cannot be loaded by Rdkit. Returning None") return None pairwise_features = [] # We compute pairwise contact fingerprints for (frag1, frag2) in itertools.combinations(fragments, 2): distances = compute_pairwise_distances(frag1[0], frag2[0]) xyzs = [frag1[0], frag2[0]] #(lig_xyz, lig_rdk), (prot_xyz, prot_rdk) = mol, protein #distances = compute_pairwise_distances(prot_xyz, lig_xyz) pairwise_features.append( sum([ voxelize( convert_atom_to_voxel, Loading @@ -195,14 +202,18 @@ class ContactCircularVoxelizer(ComplexFeaturizer): xyz, feature_dict=ecfp_dict, nb_channel=self.size) for xyz, ecfp_dict in zip((prot_xyz, lig_xyz), featurize_binding_pocket_ecfp( prot_xyz, prot_rdk, lig_xyz, lig_rdk, for xyz, ecfp_dict in zip(xyzs, featurize_contacts_ecfp( frag1, frag2, distances, cutoff=self.cutoff, ecfp_degree=self.radius)) ]) ] ) ############################################ print("[feat.shape for feat in pairwise_features]") print([feat.shape for feat in pairwise_features]) ############################################ # Features are of shape (voxels_per_edge, voxels_per_edge, voxels_per_edge, num_feat) so we should concatenate on the last axis. return np.concatenate(pairwise_features, axis=-1) deepchem/utils/rdkit_util.py +27 −2 Original line number Diff line number Diff line Loading @@ -648,11 +648,36 @@ def get_contact_atom_indices(fragments, cutoff=4.5): #return atoms_to_keep def compute_contact_centroid(molecular_complex, cutoff=4.5): """Computes the (x,y,z) centroid of the contact regions of this molecular complex. For a molecular complex, it's necessary for various featurizations that compute voxel grids to find a reasonable center for the voxelization. This function computes the centroid of all the contact atoms, defined as an atom that's within `cutoff` Angstroms of an atom from a different molecule. Parameters ---------- molecular_complex: Object A representation of a molecular complex, produced by `rdkit_util.load_complex`. cutoff: float, optional The distance in Angstroms considered for computing contacts. """ fragments = reduce_molecular_complex_to_contacts(molecular_complex, cutoff) coords = [frag[0] for frag in fragments] contact_coords = merge_molecules_xyz(coords) centroid = np.mean(contact_coords, axis=0) return (centroid) def compute_centroid(coordinates): """Compute the x,y,z centroid of provided coordinates """Compute the (x,y,z) centroid of provided coordinates Parameters ---------- coordinates: np.ndarray Shape (N, 3), where N is number atoms. Shape `(N, 3)`, where `N` is the number of atoms. """ centroid = np.mean(coordinates, axis=0) return (centroid) Loading examples/pdbbind/featurize_atomic_conv.py +9 −2 Original line number Diff line number Diff line Loading @@ -4,14 +4,21 @@ import logging from deepchem.molnet.load_function.pdbbind_datasets import get_pdbbind_molecular_complex_files complex_files = get_pdbbind_molecular_complex_files(subset="core", version="v2015", interactions="protein-ligand", load_binding_pocket=False) core_subset = complex_files[:2] n_complexes = len(complex_files) n_featurize = n_complexes core_subset = complex_files[:n_featurize] max_num_neighbors = 4 # Cutoff in angstroms neighbor_cutoff = 4 featurizer = dc.feat.AtomicConvFeaturizer( frag_max_atoms=70, frag_max_atoms=[100, 50], max_num_neighbors=max_num_neighbors, neighbor_cutoff=neighbor_cutoff) features, failures = featurizer.featurize_complexes( core_subset, parallelize=False) print("features.shape") print(features.shape) print("%d failures" % len(failures)) assert features.shape == (n_featurize -len(failures), 2) Loading
deepchem/feat/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -27,3 +27,4 @@ from deepchem.feat.adjacency_fingerprints import AdjacencyFingerprint from deepchem.feat.smiles_featurizers import SmilesToSeq, SmilesToImage from deepchem.feat.materials_featurizers import ElementPropertyFingerprint, SineCoulombMatrix, StructureGraphFeaturizer from deepchem.feat.contact_fingerprints import ContactCircularFingerprint from deepchem.feat.contact_fingerprints import ContactCircularVoxelizer
deepchem/feat/atomic_coordinates.py +9 −0 Original line number Diff line number Diff line Loading @@ -237,6 +237,11 @@ class AtomicConvFeaturizer(ComplexFeaturizer): z-matrix for the molecule which is an array of shape (N_atoms,) that contains the atomic number of that atom. If `reduce_to_contacts` is set to `True`, only those atoms which are in a contact region of the molecular complex are considered. This can dramatically reduce the number of atoms that need to be featurized so this feature is set to True by default. Since the featurization computes these three quantities for each of the two molecules and the complex, a total of 9 quantities are returned for each complex. Note that for Loading Loading @@ -310,6 +315,10 @@ class AtomicConvFeaturizer(ComplexFeaturizer): if self.reduce_to_contacts: fragments = reduce_molecular_complex_to_contacts(fragments, self.cutoff) coords = [frag[0] for frag in fragments] ################################################ print("[coord.shape for coord in coords]") print([coord.shape for coord in coords]) ################################################ mols = [frag[1] for frag in fragments] #system_mol = rdkit_util.merge_molecules(mols) system_mol = rdkit_util.merge_molecular_fragments(mols) Loading
deepchem/feat/contact_fingerprints.py +48 −37 Original line number Diff line number Diff line Loading @@ -90,14 +90,6 @@ class ContactCircularFingerprint(ComplexFeaturizer): """ Compute featurization for a single mol/protein complex TODO(rbharath): This is very not ergonomic. I'd much prefer returning an vector instead of a list of two vectors. In addition, there's a question of efficiency. RdkitGridFeaturizer caches rotated versions etc internally. To make things work out of box, we are accepting that kludgey input. This needs to be cleaned up before full merge. Parameters ---------- molecular_complex: Object Loading Loading @@ -171,20 +163,35 @@ class ContactCircularVoxelizer(ComplexFeaturizer): self.voxel_width = voxel_width self.voxels_per_edge = int(self.box_width / self.voxel_width) def _featurize_complex(self, mol, protein): def _featurize_complex(self, molecular_complex): """ Compute featurization for a single mol/protein complex Parameters ---------- mol: object Representation of the molecule protein: object Representation of the protein molecular_complex: Object A representation of a molecular complex, produced by `rdkit_util.load_complex`. """ (lig_xyz, lig_rdk), (prot_xyz, prot_rdk) = mol, protein distances = compute_pairwise_distances(prot_xyz, lig_xyz) return [ # TODO(rbharath): This is a little tricky in the generalized # regime, but we need to find a way to compute the centroid. My # idea is that we can compute the centroid of the contact # atoms.and use this to recenter all the fragments. try: fragments = rdkit_util.load_complex(molecular_complex, add_hydrogens=False) except MoleculeLoadException: logger.warning("This molecule cannot be loaded by Rdkit. Returning None") return None pairwise_features = [] # We compute pairwise contact fingerprints for (frag1, frag2) in itertools.combinations(fragments, 2): distances = compute_pairwise_distances(frag1[0], frag2[0]) xyzs = [frag1[0], frag2[0]] #(lig_xyz, lig_rdk), (prot_xyz, prot_rdk) = mol, protein #distances = compute_pairwise_distances(prot_xyz, lig_xyz) pairwise_features.append( sum([ voxelize( convert_atom_to_voxel, Loading @@ -195,14 +202,18 @@ class ContactCircularVoxelizer(ComplexFeaturizer): xyz, feature_dict=ecfp_dict, nb_channel=self.size) for xyz, ecfp_dict in zip((prot_xyz, lig_xyz), featurize_binding_pocket_ecfp( prot_xyz, prot_rdk, lig_xyz, lig_rdk, for xyz, ecfp_dict in zip(xyzs, featurize_contacts_ecfp( frag1, frag2, distances, cutoff=self.cutoff, ecfp_degree=self.radius)) ]) ] ) ############################################ print("[feat.shape for feat in pairwise_features]") print([feat.shape for feat in pairwise_features]) ############################################ # Features are of shape (voxels_per_edge, voxels_per_edge, voxels_per_edge, num_feat) so we should concatenate on the last axis. return np.concatenate(pairwise_features, axis=-1)
deepchem/utils/rdkit_util.py +27 −2 Original line number Diff line number Diff line Loading @@ -648,11 +648,36 @@ def get_contact_atom_indices(fragments, cutoff=4.5): #return atoms_to_keep def compute_contact_centroid(molecular_complex, cutoff=4.5): """Computes the (x,y,z) centroid of the contact regions of this molecular complex. For a molecular complex, it's necessary for various featurizations that compute voxel grids to find a reasonable center for the voxelization. This function computes the centroid of all the contact atoms, defined as an atom that's within `cutoff` Angstroms of an atom from a different molecule. Parameters ---------- molecular_complex: Object A representation of a molecular complex, produced by `rdkit_util.load_complex`. cutoff: float, optional The distance in Angstroms considered for computing contacts. """ fragments = reduce_molecular_complex_to_contacts(molecular_complex, cutoff) coords = [frag[0] for frag in fragments] contact_coords = merge_molecules_xyz(coords) centroid = np.mean(contact_coords, axis=0) return (centroid) def compute_centroid(coordinates): """Compute the x,y,z centroid of provided coordinates """Compute the (x,y,z) centroid of provided coordinates Parameters ---------- coordinates: np.ndarray Shape (N, 3), where N is number atoms. Shape `(N, 3)`, where `N` is the number of atoms. """ centroid = np.mean(coordinates, axis=0) return (centroid) Loading
examples/pdbbind/featurize_atomic_conv.py +9 −2 Original line number Diff line number Diff line Loading @@ -4,14 +4,21 @@ import logging from deepchem.molnet.load_function.pdbbind_datasets import get_pdbbind_molecular_complex_files complex_files = get_pdbbind_molecular_complex_files(subset="core", version="v2015", interactions="protein-ligand", load_binding_pocket=False) core_subset = complex_files[:2] n_complexes = len(complex_files) n_featurize = n_complexes core_subset = complex_files[:n_featurize] max_num_neighbors = 4 # Cutoff in angstroms neighbor_cutoff = 4 featurizer = dc.feat.AtomicConvFeaturizer( frag_max_atoms=70, frag_max_atoms=[100, 50], max_num_neighbors=max_num_neighbors, neighbor_cutoff=neighbor_cutoff) features, failures = featurizer.featurize_complexes( core_subset, parallelize=False) print("features.shape") print(features.shape) print("%d failures" % len(failures)) assert features.shape == (n_featurize -len(failures), 2)
