Loading deepchem/feat/graph_features.py +58 −24 Original line number Diff line number Diff line Loading @@ -455,7 +455,7 @@ def max_pair_distance_pairs(mol: RDKitMol, sum_adj += np.linalg.matrix_power(adj, power) nonzero_locs = np.where(sum_adj != 0) num_pairs = len(nonzero_locs[0]) # This creates a mstrix of shape (2, num_pairs) # This creates a matrix of shape (2, num_pairs) pair_edges = np.reshape(np.array(list(zip(nonzero_locs))), (2, num_pairs)) return pair_edges Loading Loading @@ -489,13 +489,14 @@ def pair_features( The number of different bond types to consider. graph_distance: bool, optional (default True) If true, use graph distance between molecules. Else use euclidean distance. The specified `mol` must have a conformer. Atomic positions will be retrieved by calling `mol.getConformer(0)`. distance. The specified `mol` must have a conformer. Atomic positions will be retrieved by calling `mol.getConformer(0)`. max_pair_distance: Union[int, str], (default 'infinity') This value can be a positive integer or the string 'infinity'. This parameter determines the maximum graph distance at which pair features are computed. For example, if `max_pair_distance==2`, then pair features are computed only for atoms at most graph distance 2 apart. This value can be a positive integer or the string 'infinity'. This parameter determines the maximum graph distance at which pair features are computed. For example, if `max_pair_distance==2`, then pair features are computed only for atoms at most graph distance 2 apart. Note ---- Loading @@ -504,8 +505,13 @@ def pair_features( Returns ------- features: np.ndarray Of shape `(N, N, bt_len + max_distance + 1)`. This is the array of pairwise features for all atom pairs. Of shape `(N_edges, bt_len + max_distance + 1)`. This is the array of pairwise features for all atom pairs, where N_edges is the number of edges within max_pair_distance of one another in this molecules. pair_edges: np.ndarray Of shape `(2, num_pairs)` where `num_pairs` is the total number of pairs within `max_pair_distance` of one another. """ if graph_distance: max_distance = 7 Loading @@ -514,27 +520,55 @@ def pair_features( N = mol.GetNumAtoms() pair_edges = max_pair_distance_pairs(mol, max_pair_distance) num_pairs = pair_edges.shape[1] # TODO(rbharath): Figure out how to rewrite this to use the pair_edges correctly if max_pair_distance == "infinity": features = np.zeros((N, N, bt_len + max_distance + 1)) N_edges = pair_edges.shape[1] features = np.zeros((N_edges, bt_len + max_distance + 1)) #if max_pair_distance == "infinity": # features = np.zeros((N, N, bt_len + max_distance + 1)) #else: # Get mapping mapping = {} for n in range(N_edges): a1, a2 = pair_edges[:, n] mapping[(int(a1), int(a2))] = n num_atoms = mol.GetNumAtoms() rings = mol.GetRingInfo().AtomRings() for a1 in range(num_atoms): for a2 in bond_adj_list[a1]: # first `bt_len` features are bond features(if applicable) features[a1, a2, :bt_len] = np.asarray( if (int(a1), int(a2)) not in mapping: raise ValueError( "Malformed molecule with bonds not in specified graph distance.") else: n = mapping[(int(a1), int(a2))] features[n, :bt_len] = np.asarray( bond_features_map[tuple(sorted((a1, a2)))], dtype=float) for ring in rings: if a1 in ring: for a2 in ring: if (int(a1), int(a2)) not in mapping: # For ring pairs outside max pairs distance continue continue else: n = mapping[(int(a1), int(a2))] # `bt_len`-th feature is if the pair of atoms are in the same ring features[a1, ring, bt_len] = 1 features[a1, a1, bt_len] = 0. if a2 == a1: features[n, bt_len] = 0 else: features[n, bt_len] = 1 #features[a1, ring, bt_len] = 1 #features[a1, a1, bt_len] = 0. # graph distance between two atoms if graph_distance: # distance is a matrix of 1-hot encoded distances # distance is a matrix of 1-hot encoded distances for all atoms distance = find_distance( a1, num_atoms, bond_adj_list, max_distance=max_distance) features[a1, :, bt_len + 1:] = distance for a2 in range(num_atoms): if (int(a1), int(a2)) not in mapping: # For ring pairs outside max pairs distance continue continue else: n = mapping[(int(a1), int(a2))] features[n, bt_len + 1:] = distance[a2] # Euclidean distance between atoms if not graph_distance: coords = np.zeros((N, 3)) Loading @@ -545,11 +579,11 @@ def pair_features( np.stack([coords] * N, axis=1) - \ np.stack([coords] * N, axis=0)), axis=2)) if max_pair_distance == "infinity": features = np.reshape(features, (N * N, bt_len + max_distance + 1)) return features #if max_pair_distance == "infinity": # features = np.reshape(features, (N * N, bt_len + max_distance + 1)) # return features return features return features, pair_edges def find_distance(a1: RDKitAtom, num_atoms: int, bond_adj_list, Loading Loading @@ -824,7 +858,7 @@ class WeaveFeaturizer(MolecularFeaturizer): bond_adj_list[bond[1]].append(bond[0]) # Calculate pair features pairs = pair_features( pairs, pair_edges = pair_features( mol, bond_features_map, bond_adj_list, Loading @@ -832,7 +866,7 @@ class WeaveFeaturizer(MolecularFeaturizer): graph_distance=self.graph_distance, max_pair_distance=self.max_pair_distance) return WeaveMol(nodes, pairs) return WeaveMol(nodes, pairs, pair_edges) class AtomicConvFeaturizer(ComplexNeighborListFragmentAtomicCoordinates): Loading deepchem/feat/mol_graphs.py +5 −1 Original line number Diff line number Diff line Loading @@ -379,11 +379,15 @@ class WeaveMol(object): .. [1] Kearnes, Steven, et al. "Molecular graph convolutions: moving beyond fingerprints." Journal of computer-aided molecular design 30.8 (2016): 595-608. """ def __init__(self, nodes, pairs): def __init__(self, nodes, pairs, pair_edges): self.nodes = nodes self.pairs = pairs self.num_atoms = self.nodes.shape[0] self.n_features = self.nodes.shape[1] self.pair_edges = pair_edges def get_pair_edges(self): return self.pair_edges def get_pair_features(self): return self.pairs Loading deepchem/feat/tests/test_weave.py +25 −20 Original line number Diff line number Diff line Loading @@ -33,6 +33,24 @@ def test_max_pair_distance_pairs(): assert pair_edges.shape == (2, 9) def test_max_pair_distance_infinity(): """Test that max pair distance pairs are computed properly with infinity distance.""" from rdkit import Chem # Test alkane mol = Chem.MolFromSmiles('CCC') # Test distance infinity pair_edges = max_pair_distance_pairs(mol, "infinity") # Everything is connected at this distance assert pair_edges.shape == (2, 9) # Test pentane mol = Chem.MolFromSmiles('CCCCC') # Test distance infinity pair_edges = max_pair_distance_pairs(mol, "infinity") # Everything is connected at this distance assert pair_edges.shape == (2, 25) def test_weave_single_carbon(): """Test that single carbon atom is featurized properly.""" mols = ['C'] Loading Loading @@ -73,8 +91,10 @@ def test_weave_alkane_max_pairs(): """Test on simple alkane with max pairs distance cutoff""" mols = ['CCC'] featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) mol_list = featurizer.featurize(mols) mol = mol_list[0] #mol_list = featurizer.featurize(mols) #mol = mol_list[0] from rdkit import Chem mol = featurizer._featurize(Chem.MolFromSmiles(mols[0])) # 3 carbonds in alkane assert mol.get_num_atoms() == 3 Loading @@ -82,7 +102,9 @@ def test_weave_alkane_max_pairs(): # Test feature sizes assert mol.get_num_features() == 75 # Should be a 3x3 interaction grid # Should be a 7x14 interaction grid since there are 7 pairs within graph # distance 1 (3 self interactions plus 2 bonds counted twice because of # symmetry) assert mol.get_pair_features().shape == (7, 14) Loading @@ -105,20 +127,3 @@ def test_carbon_nitrogen(): # Should be a 3x3 interaction grid assert mol.get_pair_features().shape == (5 * 5, 14) #def test_alkane_max_pair_distance(): # """Test on simple alkane with max_pair_distance < infinity""" # mols = ['CCC'] # featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) # mol_list = featurizer.featurize(mols) # mol = mol_list[0] # # # 3 carbonds in alkane # assert mol.get_num_atoms() == 3 # # # Test feature sizes # assert mol.get_num_features() == 75 # # # Should be a 3x3 interaction grid # assert mol.get_pair_features().shape == (3, 1, 14) deepchem/models/graph_models.py +66 −35 Original line number Diff line number Diff line Loading @@ -197,7 +197,6 @@ class WeaveModel(KerasModel): self.n_classes = n_classes # Build the model. atom_features = Input(shape=(self.n_atom_feat[0],)) pair_features = Input(shape=(self.n_pair_feat[0],)) pair_split = Input(shape=tuple(), dtype=tf.int32) Loading Loading @@ -277,6 +276,71 @@ class WeaveModel(KerasModel): super(WeaveModel, self).__init__( model, loss, output_types=output_types, batch_size=batch_size, **kwargs) def compute_features_on_batch(self, X_b): """Compute tensors that will be input into the model from featurized representation. The featurized input to `WeaveModel` is instances of `WeaveMol` created by `WeaveFeaturizer`. This method converts input `WeaveMol` objects into tensors used by the Keras implementation to compute `WeaveModel` outputs. Parameters ---------- X_b: np.ndarray A numpy array with dtype=object where elements are `WeaveMol` objects. Returns ------- atom_feat: np.ndarray Of shape `(N_atoms, N_atom_feat)`. pair_feat: np.ndarray Of shape `(N_pairs, N_pair_feat)`. Note that `N_pairs` will depend on the number of pairs being considered. If `max_pair_distance` is "infinity", then this will be `N_atoms**2`. Else it will be the number of pairs within the specifed graph distance. pair_split: np.ndarray Of shape `(N_pairs,)`. The i-th entry in this array will tell you the originating atom for this pair (the "source"). Note that pairs are symmetric so for a pair `(a, b)`, both `a` and `b` will separately be sources at different points in this array. atom_split: np.ndarray Of shape `(N_atoms,)`. The i-th entry in this array will be the molecule with the i-th atom belongs to. atom_to_pair: np.ndarray Of shape `(N_pairs, 2)`. The i-th row in this array will be the array `[a, b]` if `(a, b)` is a pair to be considered. (Note by symmetry, this implies some other row will contain `[b, a]`. """ atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] start = 0 for im, mol in enumerate(X_b): n_atoms = mol.get_num_atoms() # pair_edges is of shape (2, N) pair_edges = mol.get_pair_edges() N_pairs = pair_edges[1] # number of atoms in each molecule atom_split.extend([im] * n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append(pair_edges.T + start) # Get starting pair atoms pair_starts = pair_edges.T[:, 0] # number of pairs for each atom pair_split.extend(pair_starts + start) start = start + n_atoms # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append(mol.get_pair_features()) return (np.concatenate(atom_feat, axis=0), np.concatenate( pair_feat, axis=0), np.array(pair_split), np.array(atom_split), np.concatenate(atom_to_pair, axis=0)) def default_generator( self, dataset: Dataset, Loading Loading @@ -313,40 +377,7 @@ class WeaveModel(KerasModel): if self.mode == 'classification': y_b = to_one_hot(y_b.flatten(), self.n_classes).reshape( -1, self.n_tasks, self.n_classes) atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] start = 0 for im, mol in enumerate(X_b): n_atoms = mol.get_num_atoms() # number of atoms in each molecule atom_split.extend([im] * n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append( np.transpose( np.array([C1.flatten() + start, C0.flatten() + start]))) # number of pairs for each atom pair_split.extend(C1.flatten() + start) start = start + n_atoms # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append( np.reshape(mol.get_pair_features(), (n_atoms * n_atoms, self.n_pair_feat[0]))) inputs = [ np.concatenate(atom_feat, axis=0), np.concatenate(pair_feat, axis=0), np.array(pair_split), np.array(atom_split), np.concatenate(atom_to_pair, axis=0) ] inputs = self.compute_features_on_batch(X_b) yield (inputs, [y_b], [w_b]) Loading deepchem/models/tests/test_weave_models.py +116 −5 Original line number Diff line number Diff line Loading @@ -39,6 +39,88 @@ def get_dataset(mode='classification', featurizer='GraphConv', num_tasks=2): return tasks, ds, transformers, metric def test_compute_features_on_infinity_distance(): """Test that WeaveModel correctly transforms WeaveMol objects into tensors with infinite max_pair_distance.""" featurizer = dc.feat.WeaveFeaturizer(max_pair_distance="infinity") X = featurizer(["C", "CCC"]) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) atom_feat, pair_feat, pair_split, atom_split, atom_to_pair = model.compute_features_on_batch( X) # There are 4 atoms each of which have 75 atom features assert atom_feat.shape == (4, 75) # There are 10 pairs with infinity distance and 14 pair features assert pair_feat.shape == (10, 14) # 4 atoms in total assert atom_split.shape == (4,) assert np.all(atom_split == np.array([0, 1, 1, 1])) # 10 pairs in total assert pair_split.shape == (10,) assert np.all(pair_split == np.array([0, 1, 1, 1, 2, 2, 2, 3, 3, 3])) # 10 pairs in total each with start/finish assert atom_to_pair.shape == (10, 2) assert np.all( atom_to_pair == np.array([[0, 0], [1, 1], [1, 2], [1, 3], [2, 1], [2, 2], [2, 3], [3, 1], [3, 2], [3, 3]])) def test_compute_features_on_distance_1(): """Test that WeaveModel correctly transforms WeaveMol objects into tensors with finite max_pair_distance.""" featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) X = featurizer(["C", "CCC"]) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) atom_feat, pair_feat, pair_split, atom_split, atom_to_pair = model.compute_features_on_batch( X) # There are 4 atoms each of which have 75 atom features assert atom_feat.shape == (4, 75) # There are 8 pairs with distance 1 and 14 pair features. (To see why 8, # there's the self pair for "C". For "CCC" there are 7 pairs including self # connections and accounting for symmetry.) assert pair_feat.shape == (8, 14) # 4 atoms in total assert atom_split.shape == (4,) assert np.all(atom_split == np.array([0, 1, 1, 1])) # 10 pairs in total assert pair_split.shape == (8,) print("pair_split") print(pair_split) print("atom_to_pair") print(atom_to_pair) # The center atom is self connected and to both neighbors so it appears # thrice. The canonical ranking used in MolecularFeaturizer means this # central atom is ranked last in ordering. assert np.all(pair_split == np.array([0, 1, 1, 2, 2, 3, 3, 3])) # 10 pairs in total each with start/finish assert atom_to_pair.shape == (8, 2) assert np.all(atom_to_pair == np.array([[0, 0], [1, 1], [1, 3], [2, 2], [2, 3], [3, 1], [3, 2], [3, 3]])) @flaky @pytest.mark.slow def test_weave_model(): Loading Loading @@ -84,16 +166,15 @@ def test_weave_regression_model(): assert scores['mean_absolute_error'] < 0.1 def test_weave_fit_simple(): featurizer = dc.feat.WeaveFeaturizer() def test_weave_fit_simple_infinity_distance(): featurizer = dc.feat.WeaveFeaturizer(max_pair_distance="infinity") X = featurizer(["C", "CCC"]) y = np.random.randint(2, size=(2,)) y = np.array([0, 1.]) dataset = dc.data.NumpyDataset(X, y) tasks, dataset, transformers, metric = get_dataset('classification', 'Weave') batch_size = 20 model = WeaveModel( len(tasks), 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], Loading @@ -105,5 +186,35 @@ def test_weave_fit_simple(): }, learning_rage=0.0005) model.fit(dataset, nb_epoch=200) transformers = [] metric = dc.metrics.Metric( dc.metrics.roc_auc_score, np.mean, mode="classification") scores = model.evaluate(dataset, [metric], transformers) assert scores['mean-roc_auc_score'] >= 0.9 def test_weave_fit_simple_distance_1(): featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) X = featurizer(["C", "CCC"]) y = np.array([0, 1.]) dataset = dc.data.NumpyDataset(X, y) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) model.fit(dataset, nb_epoch=200) transformers = [] metric = dc.metrics.Metric( dc.metrics.roc_auc_score, np.mean, mode="classification") scores = model.evaluate(dataset, [metric], transformers) assert scores['mean-roc_auc_score'] >= 0.9 Loading
deepchem/feat/graph_features.py +58 −24 Original line number Diff line number Diff line Loading @@ -455,7 +455,7 @@ def max_pair_distance_pairs(mol: RDKitMol, sum_adj += np.linalg.matrix_power(adj, power) nonzero_locs = np.where(sum_adj != 0) num_pairs = len(nonzero_locs[0]) # This creates a mstrix of shape (2, num_pairs) # This creates a matrix of shape (2, num_pairs) pair_edges = np.reshape(np.array(list(zip(nonzero_locs))), (2, num_pairs)) return pair_edges Loading Loading @@ -489,13 +489,14 @@ def pair_features( The number of different bond types to consider. graph_distance: bool, optional (default True) If true, use graph distance between molecules. Else use euclidean distance. The specified `mol` must have a conformer. Atomic positions will be retrieved by calling `mol.getConformer(0)`. distance. The specified `mol` must have a conformer. Atomic positions will be retrieved by calling `mol.getConformer(0)`. max_pair_distance: Union[int, str], (default 'infinity') This value can be a positive integer or the string 'infinity'. This parameter determines the maximum graph distance at which pair features are computed. For example, if `max_pair_distance==2`, then pair features are computed only for atoms at most graph distance 2 apart. This value can be a positive integer or the string 'infinity'. This parameter determines the maximum graph distance at which pair features are computed. For example, if `max_pair_distance==2`, then pair features are computed only for atoms at most graph distance 2 apart. Note ---- Loading @@ -504,8 +505,13 @@ def pair_features( Returns ------- features: np.ndarray Of shape `(N, N, bt_len + max_distance + 1)`. This is the array of pairwise features for all atom pairs. Of shape `(N_edges, bt_len + max_distance + 1)`. This is the array of pairwise features for all atom pairs, where N_edges is the number of edges within max_pair_distance of one another in this molecules. pair_edges: np.ndarray Of shape `(2, num_pairs)` where `num_pairs` is the total number of pairs within `max_pair_distance` of one another. """ if graph_distance: max_distance = 7 Loading @@ -514,27 +520,55 @@ def pair_features( N = mol.GetNumAtoms() pair_edges = max_pair_distance_pairs(mol, max_pair_distance) num_pairs = pair_edges.shape[1] # TODO(rbharath): Figure out how to rewrite this to use the pair_edges correctly if max_pair_distance == "infinity": features = np.zeros((N, N, bt_len + max_distance + 1)) N_edges = pair_edges.shape[1] features = np.zeros((N_edges, bt_len + max_distance + 1)) #if max_pair_distance == "infinity": # features = np.zeros((N, N, bt_len + max_distance + 1)) #else: # Get mapping mapping = {} for n in range(N_edges): a1, a2 = pair_edges[:, n] mapping[(int(a1), int(a2))] = n num_atoms = mol.GetNumAtoms() rings = mol.GetRingInfo().AtomRings() for a1 in range(num_atoms): for a2 in bond_adj_list[a1]: # first `bt_len` features are bond features(if applicable) features[a1, a2, :bt_len] = np.asarray( if (int(a1), int(a2)) not in mapping: raise ValueError( "Malformed molecule with bonds not in specified graph distance.") else: n = mapping[(int(a1), int(a2))] features[n, :bt_len] = np.asarray( bond_features_map[tuple(sorted((a1, a2)))], dtype=float) for ring in rings: if a1 in ring: for a2 in ring: if (int(a1), int(a2)) not in mapping: # For ring pairs outside max pairs distance continue continue else: n = mapping[(int(a1), int(a2))] # `bt_len`-th feature is if the pair of atoms are in the same ring features[a1, ring, bt_len] = 1 features[a1, a1, bt_len] = 0. if a2 == a1: features[n, bt_len] = 0 else: features[n, bt_len] = 1 #features[a1, ring, bt_len] = 1 #features[a1, a1, bt_len] = 0. # graph distance between two atoms if graph_distance: # distance is a matrix of 1-hot encoded distances # distance is a matrix of 1-hot encoded distances for all atoms distance = find_distance( a1, num_atoms, bond_adj_list, max_distance=max_distance) features[a1, :, bt_len + 1:] = distance for a2 in range(num_atoms): if (int(a1), int(a2)) not in mapping: # For ring pairs outside max pairs distance continue continue else: n = mapping[(int(a1), int(a2))] features[n, bt_len + 1:] = distance[a2] # Euclidean distance between atoms if not graph_distance: coords = np.zeros((N, 3)) Loading @@ -545,11 +579,11 @@ def pair_features( np.stack([coords] * N, axis=1) - \ np.stack([coords] * N, axis=0)), axis=2)) if max_pair_distance == "infinity": features = np.reshape(features, (N * N, bt_len + max_distance + 1)) return features #if max_pair_distance == "infinity": # features = np.reshape(features, (N * N, bt_len + max_distance + 1)) # return features return features return features, pair_edges def find_distance(a1: RDKitAtom, num_atoms: int, bond_adj_list, Loading Loading @@ -824,7 +858,7 @@ class WeaveFeaturizer(MolecularFeaturizer): bond_adj_list[bond[1]].append(bond[0]) # Calculate pair features pairs = pair_features( pairs, pair_edges = pair_features( mol, bond_features_map, bond_adj_list, Loading @@ -832,7 +866,7 @@ class WeaveFeaturizer(MolecularFeaturizer): graph_distance=self.graph_distance, max_pair_distance=self.max_pair_distance) return WeaveMol(nodes, pairs) return WeaveMol(nodes, pairs, pair_edges) class AtomicConvFeaturizer(ComplexNeighborListFragmentAtomicCoordinates): Loading
deepchem/feat/mol_graphs.py +5 −1 Original line number Diff line number Diff line Loading @@ -379,11 +379,15 @@ class WeaveMol(object): .. [1] Kearnes, Steven, et al. "Molecular graph convolutions: moving beyond fingerprints." Journal of computer-aided molecular design 30.8 (2016): 595-608. """ def __init__(self, nodes, pairs): def __init__(self, nodes, pairs, pair_edges): self.nodes = nodes self.pairs = pairs self.num_atoms = self.nodes.shape[0] self.n_features = self.nodes.shape[1] self.pair_edges = pair_edges def get_pair_edges(self): return self.pair_edges def get_pair_features(self): return self.pairs Loading
deepchem/feat/tests/test_weave.py +25 −20 Original line number Diff line number Diff line Loading @@ -33,6 +33,24 @@ def test_max_pair_distance_pairs(): assert pair_edges.shape == (2, 9) def test_max_pair_distance_infinity(): """Test that max pair distance pairs are computed properly with infinity distance.""" from rdkit import Chem # Test alkane mol = Chem.MolFromSmiles('CCC') # Test distance infinity pair_edges = max_pair_distance_pairs(mol, "infinity") # Everything is connected at this distance assert pair_edges.shape == (2, 9) # Test pentane mol = Chem.MolFromSmiles('CCCCC') # Test distance infinity pair_edges = max_pair_distance_pairs(mol, "infinity") # Everything is connected at this distance assert pair_edges.shape == (2, 25) def test_weave_single_carbon(): """Test that single carbon atom is featurized properly.""" mols = ['C'] Loading Loading @@ -73,8 +91,10 @@ def test_weave_alkane_max_pairs(): """Test on simple alkane with max pairs distance cutoff""" mols = ['CCC'] featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) mol_list = featurizer.featurize(mols) mol = mol_list[0] #mol_list = featurizer.featurize(mols) #mol = mol_list[0] from rdkit import Chem mol = featurizer._featurize(Chem.MolFromSmiles(mols[0])) # 3 carbonds in alkane assert mol.get_num_atoms() == 3 Loading @@ -82,7 +102,9 @@ def test_weave_alkane_max_pairs(): # Test feature sizes assert mol.get_num_features() == 75 # Should be a 3x3 interaction grid # Should be a 7x14 interaction grid since there are 7 pairs within graph # distance 1 (3 self interactions plus 2 bonds counted twice because of # symmetry) assert mol.get_pair_features().shape == (7, 14) Loading @@ -105,20 +127,3 @@ def test_carbon_nitrogen(): # Should be a 3x3 interaction grid assert mol.get_pair_features().shape == (5 * 5, 14) #def test_alkane_max_pair_distance(): # """Test on simple alkane with max_pair_distance < infinity""" # mols = ['CCC'] # featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) # mol_list = featurizer.featurize(mols) # mol = mol_list[0] # # # 3 carbonds in alkane # assert mol.get_num_atoms() == 3 # # # Test feature sizes # assert mol.get_num_features() == 75 # # # Should be a 3x3 interaction grid # assert mol.get_pair_features().shape == (3, 1, 14)
deepchem/models/graph_models.py +66 −35 Original line number Diff line number Diff line Loading @@ -197,7 +197,6 @@ class WeaveModel(KerasModel): self.n_classes = n_classes # Build the model. atom_features = Input(shape=(self.n_atom_feat[0],)) pair_features = Input(shape=(self.n_pair_feat[0],)) pair_split = Input(shape=tuple(), dtype=tf.int32) Loading Loading @@ -277,6 +276,71 @@ class WeaveModel(KerasModel): super(WeaveModel, self).__init__( model, loss, output_types=output_types, batch_size=batch_size, **kwargs) def compute_features_on_batch(self, X_b): """Compute tensors that will be input into the model from featurized representation. The featurized input to `WeaveModel` is instances of `WeaveMol` created by `WeaveFeaturizer`. This method converts input `WeaveMol` objects into tensors used by the Keras implementation to compute `WeaveModel` outputs. Parameters ---------- X_b: np.ndarray A numpy array with dtype=object where elements are `WeaveMol` objects. Returns ------- atom_feat: np.ndarray Of shape `(N_atoms, N_atom_feat)`. pair_feat: np.ndarray Of shape `(N_pairs, N_pair_feat)`. Note that `N_pairs` will depend on the number of pairs being considered. If `max_pair_distance` is "infinity", then this will be `N_atoms**2`. Else it will be the number of pairs within the specifed graph distance. pair_split: np.ndarray Of shape `(N_pairs,)`. The i-th entry in this array will tell you the originating atom for this pair (the "source"). Note that pairs are symmetric so for a pair `(a, b)`, both `a` and `b` will separately be sources at different points in this array. atom_split: np.ndarray Of shape `(N_atoms,)`. The i-th entry in this array will be the molecule with the i-th atom belongs to. atom_to_pair: np.ndarray Of shape `(N_pairs, 2)`. The i-th row in this array will be the array `[a, b]` if `(a, b)` is a pair to be considered. (Note by symmetry, this implies some other row will contain `[b, a]`. """ atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] start = 0 for im, mol in enumerate(X_b): n_atoms = mol.get_num_atoms() # pair_edges is of shape (2, N) pair_edges = mol.get_pair_edges() N_pairs = pair_edges[1] # number of atoms in each molecule atom_split.extend([im] * n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append(pair_edges.T + start) # Get starting pair atoms pair_starts = pair_edges.T[:, 0] # number of pairs for each atom pair_split.extend(pair_starts + start) start = start + n_atoms # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append(mol.get_pair_features()) return (np.concatenate(atom_feat, axis=0), np.concatenate( pair_feat, axis=0), np.array(pair_split), np.array(atom_split), np.concatenate(atom_to_pair, axis=0)) def default_generator( self, dataset: Dataset, Loading Loading @@ -313,40 +377,7 @@ class WeaveModel(KerasModel): if self.mode == 'classification': y_b = to_one_hot(y_b.flatten(), self.n_classes).reshape( -1, self.n_tasks, self.n_classes) atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] start = 0 for im, mol in enumerate(X_b): n_atoms = mol.get_num_atoms() # number of atoms in each molecule atom_split.extend([im] * n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append( np.transpose( np.array([C1.flatten() + start, C0.flatten() + start]))) # number of pairs for each atom pair_split.extend(C1.flatten() + start) start = start + n_atoms # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append( np.reshape(mol.get_pair_features(), (n_atoms * n_atoms, self.n_pair_feat[0]))) inputs = [ np.concatenate(atom_feat, axis=0), np.concatenate(pair_feat, axis=0), np.array(pair_split), np.array(atom_split), np.concatenate(atom_to_pair, axis=0) ] inputs = self.compute_features_on_batch(X_b) yield (inputs, [y_b], [w_b]) Loading
deepchem/models/tests/test_weave_models.py +116 −5 Original line number Diff line number Diff line Loading @@ -39,6 +39,88 @@ def get_dataset(mode='classification', featurizer='GraphConv', num_tasks=2): return tasks, ds, transformers, metric def test_compute_features_on_infinity_distance(): """Test that WeaveModel correctly transforms WeaveMol objects into tensors with infinite max_pair_distance.""" featurizer = dc.feat.WeaveFeaturizer(max_pair_distance="infinity") X = featurizer(["C", "CCC"]) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) atom_feat, pair_feat, pair_split, atom_split, atom_to_pair = model.compute_features_on_batch( X) # There are 4 atoms each of which have 75 atom features assert atom_feat.shape == (4, 75) # There are 10 pairs with infinity distance and 14 pair features assert pair_feat.shape == (10, 14) # 4 atoms in total assert atom_split.shape == (4,) assert np.all(atom_split == np.array([0, 1, 1, 1])) # 10 pairs in total assert pair_split.shape == (10,) assert np.all(pair_split == np.array([0, 1, 1, 1, 2, 2, 2, 3, 3, 3])) # 10 pairs in total each with start/finish assert atom_to_pair.shape == (10, 2) assert np.all( atom_to_pair == np.array([[0, 0], [1, 1], [1, 2], [1, 3], [2, 1], [2, 2], [2, 3], [3, 1], [3, 2], [3, 3]])) def test_compute_features_on_distance_1(): """Test that WeaveModel correctly transforms WeaveMol objects into tensors with finite max_pair_distance.""" featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) X = featurizer(["C", "CCC"]) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) atom_feat, pair_feat, pair_split, atom_split, atom_to_pair = model.compute_features_on_batch( X) # There are 4 atoms each of which have 75 atom features assert atom_feat.shape == (4, 75) # There are 8 pairs with distance 1 and 14 pair features. (To see why 8, # there's the self pair for "C". For "CCC" there are 7 pairs including self # connections and accounting for symmetry.) assert pair_feat.shape == (8, 14) # 4 atoms in total assert atom_split.shape == (4,) assert np.all(atom_split == np.array([0, 1, 1, 1])) # 10 pairs in total assert pair_split.shape == (8,) print("pair_split") print(pair_split) print("atom_to_pair") print(atom_to_pair) # The center atom is self connected and to both neighbors so it appears # thrice. The canonical ranking used in MolecularFeaturizer means this # central atom is ranked last in ordering. assert np.all(pair_split == np.array([0, 1, 1, 2, 2, 3, 3, 3])) # 10 pairs in total each with start/finish assert atom_to_pair.shape == (8, 2) assert np.all(atom_to_pair == np.array([[0, 0], [1, 1], [1, 3], [2, 2], [2, 3], [3, 1], [3, 2], [3, 3]])) @flaky @pytest.mark.slow def test_weave_model(): Loading Loading @@ -84,16 +166,15 @@ def test_weave_regression_model(): assert scores['mean_absolute_error'] < 0.1 def test_weave_fit_simple(): featurizer = dc.feat.WeaveFeaturizer() def test_weave_fit_simple_infinity_distance(): featurizer = dc.feat.WeaveFeaturizer(max_pair_distance="infinity") X = featurizer(["C", "CCC"]) y = np.random.randint(2, size=(2,)) y = np.array([0, 1.]) dataset = dc.data.NumpyDataset(X, y) tasks, dataset, transformers, metric = get_dataset('classification', 'Weave') batch_size = 20 model = WeaveModel( len(tasks), 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], Loading @@ -105,5 +186,35 @@ def test_weave_fit_simple(): }, learning_rage=0.0005) model.fit(dataset, nb_epoch=200) transformers = [] metric = dc.metrics.Metric( dc.metrics.roc_auc_score, np.mean, mode="classification") scores = model.evaluate(dataset, [metric], transformers) assert scores['mean-roc_auc_score'] >= 0.9 def test_weave_fit_simple_distance_1(): featurizer = dc.feat.WeaveFeaturizer(max_pair_distance=1) X = featurizer(["C", "CCC"]) y = np.array([0, 1.]) dataset = dc.data.NumpyDataset(X, y) batch_size = 20 model = WeaveModel( 1, batch_size=batch_size, mode='classification', fully_connected_layer_sizes=[2000, 1000], batch_normalize=True, batch_normalize_kwargs={ "fused": False, "trainable": True, "renorm": True }, learning_rage=0.0005) model.fit(dataset, nb_epoch=200) transformers = [] metric = dc.metrics.Metric( dc.metrics.roc_auc_score, np.mean, mode="classification") scores = model.evaluate(dataset, [metric], transformers) assert scores['mean-roc_auc_score'] >= 0.9
