Utilities¶

DeepChem has a broad collection of utility functions. Many of these maybe be of independent interest to users since they deal with some tricky aspects of processing scientific datatypes.

Data Utilities¶

Array Utilities¶

pad_array(x: numpy.ndarray, shape: Union[Tuple, int], fill: float = 0.0, both: bool = False) → numpy.ndarray[source]¶

Pad an array with a fill value.

Parameters

x (np.ndarray) – A numpy array.
shape (Tuple or int) – Desired shape. If int, all dimensions are padded to that size.
fill (float, optional (default 0.0)) – The padded value.
both (bool, optional (default False)) – If True, split the padding on both sides of each axis. If False, padding is applied to the end of each axis.

Returns

A padded numpy array

Return type

np.ndarray

Data Directory¶

The DeepChem data directory is where downloaded MoleculeNet datasets are stored.

get_data_dir() → str[source]¶

Get the DeepChem data directory.

Returns: The default path to store DeepChem data. If you want to change this path, please set your own path to DEEPCHEM_DATA_DIR as an environment variable.
Return type: str

URL Handling¶

download_url(url: str, dest_dir: str = '/tmp', name: Optional[str] = None)[source]¶

Download a file to disk.

Parameters

url (str) – The URL to download from
dest_dir (str) – The directory to save the file in
name (str) – The file name to save it as. If omitted, it will try to extract a file name from the URL

File Handling¶

untargz_file(file: str, dest_dir: str = '/tmp', name: Optional[str] = None)[source]¶

Untar and unzip a .tar.gz file to disk.

Parameters

file (str) – The filepath to decompress
dest_dir (str) – The directory to save the file in
name (str) – The file name to save it as. If omitted, it will use the file name

unzip_file(file: str, dest_dir: str = '/tmp', name: Optional[str] = None)[source]¶

Unzip a .zip file to disk.

Parameters

file (str) – The filepath to decompress
dest_dir (str) – The directory to save the file in
name (str) – The directory name to unzip it to. If omitted, it will use the file name

load_data(input_files: List[str], shard_size: Optional[int] = None) → Iterator[Any][source]¶

Loads data from files.

Parameters

input_files (List[str]) – List of filenames.
shard_size (int, default None) – Size of shard to yield

Returns

Iterator which iterates over provided files.

Return type

Iterator[Any]

Notes

The supported file types are SDF, CSV and Pickle.

load_sdf_files(input_files: List[str], clean_mols: bool = True, tasks: List[str] = [], shard_size: Optional[int] = None) → Iterator[pandas.core.frame.DataFrame][source]¶

Load SDF file into dataframe.

Parameters

input_files (List[str]) – List of filenames
clean_mols (bool, default True) – Whether to sanitize molecules.
tasks (List[str], default []) – Each entry in tasks is treated as a property in the SDF file and is retrieved with mol.GetProp(str(task)) where mol is the RDKit mol loaded from a given SDF entry.
shard_size (int, default None) – The shard size to yield at one time.

Returns

Generator which yields the dataframe which is the same shard size.

Return type

Iterator[pd.DataFrame]

Notes

This function requires RDKit to be installed.

load_csv_files(input_files: List[str], shard_size: Optional[int] = None) → Iterator[pandas.core.frame.DataFrame][source]¶

Load data as pandas dataframe from CSV files.

Parameters

input_files (List[str]) – List of filenames
shard_size (int, default None) – The shard size to yield at one time.

Returns

Generator which yields the dataframe which is the same shard size.

Return type

Iterator[pd.DataFrame]

load_json_files(input_files: List[str], shard_size: Optional[int] = None) → Iterator[pandas.core.frame.DataFrame][source]¶

Load data as pandas dataframe.

Parameters

input_files (List[str]) – List of json filenames.
shard_size (int, default None) – Chunksize for reading json files.

Returns

Generator which yields the dataframe which is the same shard size.

Return type

Iterator[pd.DataFrame]

Notes

To load shards from a json file into a Pandas dataframe, the file must be originally saved with df.to_json('filename.json', orient='records', lines=True)

load_pickle_files(input_files: List[str]) → Iterator[Any][source]¶

Load dataset from pickle files.

Parameters: input_files (List[str]) – The list of filenames of pickle file. This function can load from gzipped pickle file like XXXX.pkl.gz.
Returns: Generator which yields the objects which is loaded from each pickle file.
Return type: Iterator[Any]

load_from_disk(filename: str) → Any[source]¶

Load a dataset from file.

Parameters: filename (str) – A filename you want to load data.
Returns: A loaded object from file.
Return type: Any

save_to_disk(dataset: Any, filename: str, compress: int = 3)[source]¶

Save a dataset to file.

Parameters

dataset (str) – A data saved
filename (str) – Path to save data.
compress (int, default 3) – The compress option when dumping joblib file.

load_dataset_from_disk(save_dir: str) → Tuple[bool, Optional[Tuple[deepchem.data.datasets.DiskDataset, deepchem.data.datasets.DiskDataset, deepchem.data.datasets.DiskDataset]], List[transformers.Transformer]][source]¶

Loads MoleculeNet train/valid/test/transformers from disk.

Expects that data was saved using save_dataset_to_disk below. Expects the following directory structure for save_dir: save_dir/

—> train_dir/ | —> valid_dir/ | —> test_dir/ | —> transformers.pkl

Parameters

save_dir (str) – Directory name to load datasets.

Returns

loaded (bool) – Whether the load succeeded
all_dataset (Tuple[DiskDataset, DiskDataset, DiskDataset]) – The train, valid, test datasets
transformers (Transformer) – The transformers used for this dataset

Molecular Utilities¶

class ConformerGenerator(max_conformers: int = 1, rmsd_threshold: float = 0.5, force_field: str = 'uff', pool_multiplier: int = 10)[source]¶

Generate molecule conformers.

Notes

Procedure 1. Generate a pool of conformers. 2. Minimize conformers. 3. Prune conformers using an RMSD threshold.

Note that pruning is done _after_ minimization, which differs from the protocol described in the references 1 2.

References

1: http://rdkit.org/docs/GettingStartedInPython.html#working-with-3d-molecules
2: http://pubs.acs.org/doi/full/10.1021/ci2004658

Notes

This class requires RDKit to be installed.

__init__(max_conformers: int = 1, rmsd_threshold: float = 0.5, force_field: str = 'uff', pool_multiplier: int = 10)[source]¶

Parameters

max_conformers (int, optional (default 1)) – Maximum number of conformers to generate (after pruning).
rmsd_threshold (float, optional (default 0.5)) – RMSD threshold for pruning conformers. If None or negative, no pruning is performed.
force_field (str, optional (default 'uff')) – Force field to use for conformer energy calculation and minimization. Options are ‘uff’, ‘mmff94’, and ‘mmff94s’.
pool_multiplier (int, optional (default 10)) – Factor to multiply by max_conformers to generate the initial conformer pool. Since conformers are pruned after energy minimization, increasing the size of the pool increases the chance of identifying max_conformers unique conformers.

generate_conformers(mol: Any) → Any[source]¶

Generate conformers for a molecule.

This function returns a copy of the original molecule with embedded conformers.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object
Returns: mol – A new RDKit Mol object containing the chosen conformers, sorted by increasing energy.
Return type: rdkit.Chem.rdchem.Mol

embed_molecule(mol: Any) → Any[source]¶

Generate conformers, possibly with pruning.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object
Returns: mol – RDKit Mol object with embedded multiple conformers.
Return type: rdkit.Chem.rdchem.Mol

get_molecule_force_field(mol: Any, conf_id: Optional[int] = None, **kwargs) → Any[source]¶

Get a force field for a molecule.

Parameters

mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object with embedded conformers.
conf_id (int, optional) – ID of the conformer to associate with the force field.
kwargs (dict, optional) – Keyword arguments for force field constructor.

Returns

ff – RDKit force field instance for a molecule.

Return type

rdkit.ForceField.rdForceField.ForceField

minimize_conformers(mol: Any) → None[source]¶

Minimize molecule conformers.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object with embedded conformers.

get_conformer_energies(mol: Any) → numpy.ndarray[source]¶

Calculate conformer energies.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object with embedded conformers.
Returns: energies – Minimized conformer energies.
Return type: np.ndarray

prune_conformers(mol: Any) → Any[source]¶

Prune conformers from a molecule using an RMSD threshold, starting with the lowest energy conformer.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object
Returns: new_mol – A new rdkit.Chem.rdchem.Mol containing the chosen conformers, sorted by increasing energy.
Return type: rdkit.Chem.rdchem.Mol

static get_conformer_rmsd(mol: Any) → numpy.ndarray[source]¶

Calculate conformer-conformer RMSD.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit Mol object
Returns: rmsd – A conformer-conformer RMSD value. The shape is (NumConformers, NumConformers)
Return type: np.ndarray

class MoleculeLoadException(*args, **kwargs)[source]¶

__init__(*args, **kwargs)[source]¶

get_xyz_from_mol(mol)[source]¶

Extracts a numpy array of coordinates from a molecules.

Returns a (N, 3) numpy array of 3d coords of given rdkit molecule

Parameters: mol (rdkit Molecule) – Molecule to extract coordinates for
Return type: Numpy ndarray of shape (N, 3) where N = mol.GetNumAtoms().

add_hydrogens_to_mol(mol, is_protein=False)[source]¶

Add hydrogens to a molecule object

Parameters

mol (Rdkit Mol) – Molecule to hydrogenate
is_protein (bool, optional (default False)) – Whether this molecule is a protein.

Return type

Rdkit Mol

Note

This function requires RDKit and PDBFixer to be installed.

compute_charges(mol)[source]¶

Attempt to compute Gasteiger Charges on Mol

This also has the side effect of calculating charges on mol. The mol passed into this function has to already have been sanitized

Parameters: mol (rdkit molecule) –
Return type: No return since updates in place.

Note

This function requires RDKit to be installed.

load_molecule(molecule_file, add_hydrogens=True, calc_charges=True, sanitize=True, is_protein=False)[source]¶

Converts molecule file to (xyz-coords, obmol object)

Given molecule_file, returns a tuple of xyz coords of molecule and an rdkit object representing that molecule in that order (xyz, rdkit_mol). This ordering convention is used in the code in a few places.

Parameters

molecule_file (str) – filename for molecule
add_hydrogens (bool, optional (default True)) – If True, add hydrogens via pdbfixer
calc_charges (bool, optional (default True)) – If True, add charges via rdkit
sanitize (bool, optional (default False)) – If True, sanitize molecules via rdkit
is_protein (bool, optional (default False)) – If True`, this molecule is loaded as a protein. This flag will affect some of the cleanup procedures applied.

Returns

Tuple (xyz, mol) if file contains single molecule. Else returns a
list of the tuples for the separate molecules in this list.

Note

This function requires RDKit to be installed.

write_molecule(mol, outfile, is_protein=False)[source]¶

Write molecule to a file

This function writes a representation of the provided molecule to the specified outfile. Doesn’t return anything.

Parameters

mol (rdkit Mol) – Molecule to write
outfile (str) – Filename to write mol to
is_protein (bool, optional) – Is this molecule a protein?

Note

This function requires RDKit to be installed.

Raises: ValueError – if outfile isn’t of a supported format.:

Molecular Fragment Utilities¶

It’s often convenient to manipulate subsets of a molecule. The MolecularFragment class aids in such manipulations.

class MolecularFragment(atoms: Sequence[Any], coords: numpy.ndarray)[source]¶

A class that represents a fragment of a molecule.

It’s often convenient to represent a fragment of a molecule. For example, if two molecules form a molecular complex, it may be useful to create two fragments which represent the subsets of each molecule that’s close to the other molecule (in the contact region).

Ideally, we’d be able to do this in RDKit direct, but manipulating molecular fragments doesn’t seem to be supported functionality.

Examples

>>> import numpy as np
>>> from rdkit import Chem
>>> mol = Chem.MolFromSmiles("C")
>>> coords = np.array([[0.0, 0.0, 0.0]])
>>> atom = mol.GetAtoms()[0]
>>> fragment = MolecularFragment([atom], coords)

__init__(atoms: Sequence[Any], coords: numpy.ndarray)[source]¶

Initialize this object.

Parameters

atoms (Iterable[rdkit.Chem.rdchem.Atom]) – Each entry in this list should be a RDKit Atom.
coords (np.ndarray) – Array of locations for atoms of shape (N, 3) where N == len(atoms).

GetAtoms() → List[deepchem.utils.fragment_utils.AtomShim][source]¶

Returns the list of atoms

Returns: list of atoms in this fragment.
Return type: List[AtomShim]

GetNumAtoms() → int[source]¶

Returns the number of atoms

Returns: Number of atoms in this fragment.
Return type: int

GetCoords() → numpy.ndarray[source]¶

Returns 3D coordinates for this fragment as numpy array.

Returns: A numpy array of shape (N, 3) with coordinates for this fragment. Here, N is the number of atoms.
Return type: np.ndarray

class AtomShim(atomic_num: int, partial_charge: float, atom_coords: numpy.ndarray)[source]¶

This is a shim object wrapping an atom.

We use this class instead of raw RDKit atoms since manipulating a large number of rdkit Atoms seems to result in segfaults. Wrapping the basic information in an AtomShim seems to avoid issues.

__init__(atomic_num: int, partial_charge: float, atom_coords: numpy.ndarray)[source]¶

Initialize this object

Parameters

atomic_num (int) – Atomic number for this atom.
partial_charge (float) – The partial Gasteiger charge for this atom
atom_coords (np.ndarray) – Of shape (3,) with the coordinates of this atom

GetAtomicNum() → int[source]¶

Returns atomic number for this atom.

Returns: Atomic number for this atom.
Return type: int

GetPartialCharge() → float[source]¶

Returns partial charge for this atom.

Returns: A partial Gasteiger charge for this atom.
Return type: float

GetCoords() → numpy.ndarray[source]¶

Returns 3D coordinates for this atom as numpy array.

Returns: Numpy array of shape (3,) with coordinates for this atom.
Return type: np.ndarray

strip_hydrogens(coords: numpy.ndarray, mol: Union[Any, deepchem.utils.fragment_utils.MolecularFragment]) → Tuple[numpy.ndarray, deepchem.utils.fragment_utils.MolecularFragment][source]¶

Strip the hydrogens from input molecule

Parameters

coords (np.ndarray) – The coords must be of shape (N, 3) and correspond to coordinates of mol.
mol (rdkit.Chem.rdchem.Mol or MolecularFragment) – The molecule to strip

Returns

A tuple of (coords, mol_frag) where coords is a numpy array of coordinates with hydrogen coordinates. mol_frag is a MolecularFragment.

Return type

Tuple[np.ndarray, MolecularFragment]

Notes

This function requires RDKit to be installed.

merge_molecular_fragments(molecules: List[deepchem.utils.fragment_utils.MolecularFragment]) → Optional[deepchem.utils.fragment_utils.MolecularFragment][source]¶

Helper method to merge two molecular fragments.

Parameters: molecules (List[MolecularFragment]) – List of MolecularFragment objects.
Returns: Returns a merged MolecularFragment
Return type: Optional[MolecularFragment]

get_contact_atom_indices(fragments: List[Tuple[numpy.ndarray, Any]], cutoff: float = 4.5) → List[List[int]][source]¶

Compute that atoms close to contact region.

Molecular complexes can get very large. This can make it unwieldy to compute functions on them. To improve memory usage, it can be very useful to trim out atoms that aren’t close to contact regions. This function computes pairwise distances between all pairs of molecules in the molecular complex. If an atom is within cutoff distance of any atom on another molecule in the complex, it is regarded as a contact atom. Otherwise it is trimmed.

Parameters

fragments (List[Tuple[np.ndarray, rdkit.Chem.rdchem.Mol]]) – As returned by rdkit_utils.load_complex, a list of tuples of (coords, mol) where coords is a (N_atoms, 3) array and mol is the rdkit molecule object.
cutoff (float, optional (default 4.5)) – The cutoff distance in angstroms.

Returns

A list of length len(molecular_complex). Each entry in this list is a list of atom indices from that molecule which should be kept, in sorted order.

Return type

List[List[int]]

reduce_molecular_complex_to_contacts(fragments: List[Tuple[numpy.ndarray, Any]], cutoff: float = 4.5) → List[Tuple[numpy.ndarray, deepchem.utils.fragment_utils.MolecularFragment]][source]¶

Reduce a molecular complex to only those atoms near a contact.

Molecular complexes can get very large. This can make it unwieldy to compute functions on them. To improve memory usage, it can be very useful to trim out atoms that aren’t close to contact regions. This function takes in a molecular complex and returns a new molecular complex representation that contains only contact atoms. The contact atoms are computed by calling get_contact_atom_indices under the hood.

Parameters

fragments (List[Tuple[np.ndarray, rdkit.Chem.rdchem.Mol]]) – As returned by rdkit_utils.load_complex, a list of tuples of (coords, mol) where coords is a (N_atoms, 3) array and mol is the rdkit molecule object.
cutoff (float) – The cutoff distance in angstroms.

Returns

A list of length len(molecular_complex). Each entry in this list is a tuple of (coords, MolecularFragment). The coords is stripped down to (N_contact_atoms, 3) where N_contact_atoms is the number of contact atoms for this complex. MolecularFragment is used since it’s tricky to make a RDKit sub-molecule.

Return type

List[Tuple[np.ndarray, MolecularFragment]]

Coordinate Box Utilities¶

class CoordinateBox(x_range: Tuple[float, float], y_range: Tuple[float, float], z_range: Tuple[float, float])[source]¶

A coordinate box that represents a block in space.

Molecular complexes are typically represented with atoms as coordinate points. Each complex is naturally associated with a number of different box regions. For example, the bounding box is a box that contains all atoms in the molecular complex. A binding pocket box is a box that focuses in on a binding region of a protein to a ligand. A interface box is the region in which two proteins have a bulk interaction.

The CoordinateBox class is designed to represent such regions of space. It consists of the coordinates of the box, and the collection of atoms that live in this box alongside their coordinates.

__init__(x_range: Tuple[float, float], y_range: Tuple[float, float], z_range: Tuple[float, float])[source]¶

Initialize this box.

Parameters

x_range (Tuple[float, float]) – A tuple of (x_min, x_max) with max and min x-coordinates.
y_range (Tuple[float, float]) – A tuple of (y_min, y_max) with max and min y-coordinates.
z_range (Tuple[float, float]) – A tuple of (z_min, z_max) with max and min z-coordinates.

Raises

ValueError –

__contains__(point: Sequence[float]) → bool[source]¶

Check whether a point is in this box.

Parameters: point (Sequence[float]) – 3-tuple or list of length 3 or np.ndarray of shape (3,). The (x, y, z) coordinates of a point in space.
Returns: True if other is contained in this box.
Return type: bool

center() → Tuple[float, float, float][source]¶

Computes the center of this box.

Returns: (x, y, z) the coordinates of the center of the box.
Return type: Tuple[float, float, float]

Examples

>>> box = CoordinateBox((0, 1), (0, 1), (0, 1))
>>> box.center()
(0.5, 0.5, 0.5)

volume() → float[source]¶

Computes and returns the volume of this box.

Returns: The volume of this box. Can be 0 if box is empty
Return type: float

Examples

>>> box = CoordinateBox((0, 1), (0, 1), (0, 1))
>>> box.volume()
1

contains(other: deepchem.utils.coordinate_box_utils.CoordinateBox) → bool[source]¶

Test whether this box contains another.

This method checks whether other is contained in this box.

Parameters: other (CoordinateBox) – The box to check is contained in this box.
Returns: True if other is contained in this box.
Return type: bool
Raises: ValueError –

intersect_interval(interval1: Tuple[float, float], interval2: Tuple[float, float]) → Tuple[float, float][source]¶

Computes the intersection of two intervals.

Parameters

interval1 (Tuple[float, float]) – Should be (x1_min, x1_max)
interval2 (Tuple[float, float]) – Should be (x2_min, x2_max)

Returns

x_intersect – Should be the intersection. If the intersection is empty returns (0, 0) to represent the empty set. Otherwise is (max(x1_min, x2_min), min(x1_max, x2_max)).

Return type

Tuple[float, float]

union(box1: deepchem.utils.coordinate_box_utils.CoordinateBox, box2: deepchem.utils.coordinate_box_utils.CoordinateBox) → deepchem.utils.coordinate_box_utils.CoordinateBox[source]¶

Merges provided boxes to find the smallest union box.

This method merges the two provided boxes.

Parameters

box1 (CoordinateBox) – First box to merge in
box2 (CoordinateBox) – Second box to merge into this box

Returns

Smallest CoordinateBox that contains both box1 and box2

Return type

CoordinateBox

merge_overlapping_boxes(boxes: List[deepchem.utils.coordinate_box_utils.CoordinateBox], threshold: float = 0.8) → List[deepchem.utils.coordinate_box_utils.CoordinateBox][source]¶

Merge boxes which have an overlap greater than threshold.

Parameters

boxes (list[CoordinateBox]) – A list of CoordinateBox objects.
threshold (float, default 0.8) – The volume fraction of the boxes that must overlap for them to be merged together.

Returns

List[CoordinateBox] of merged boxes. This list will have length less than or equal to the length of boxes.

Return type

List[CoordinateBox]

get_face_boxes(coords: numpy.ndarray, pad: float = 5.0) → List[deepchem.utils.coordinate_box_utils.CoordinateBox][source]¶

For each face of the convex hull, compute a coordinate box around it.

The convex hull of a macromolecule will have a series of triangular faces. For each such triangular face, we construct a bounding box around this triangle. Think of this box as attempting to capture some binding interaction region whose exterior is controlled by the box. Note that this box will likely be a crude approximation, but the advantage of this technique is that it only uses simple geometry to provide some basic biological insight into the molecule at hand.

The pad parameter is used to control the amount of padding around the face to be used for the coordinate box.

Parameters

coords (np.ndarray) – A numpy array of shape (N, 3). The coordinates of a molecule.
pad (float, optional (default 5.0)) – The number of angstroms to pad.

Returns

boxes – List of CoordinateBox

Return type

List[CoordinateBox]

Examples

>>> coords = np.array([[0, 0, 0], [1, 0, 0], [0, 1, 0], [0, 0, 1]])
>>> boxes = get_face_boxes(coords, pad=5)

Evaluation Utils¶

class Evaluator(model, dataset: deepchem.data.datasets.Dataset, transformers: List[transformers.Transformer])[source]¶

Class that evaluates a model on a given dataset.

The evaluator class is used to evaluate a dc.models.Model class on a given dc.data.Dataset object. The evaluator is aware of dc.trans.Transformer objects so will automatically undo any transformations which have been applied.

Examples

Evaluators allow for a model to be evaluated directly on a Metric for sklearn. Let’s do a bit of setup constructing our dataset and model.

>>> import deepchem as dc
>>> import numpy as np
>>> X = np.random.rand(10, 5)
>>> y = np.random.rand(10, 1)
>>> dataset = dc.data.NumpyDataset(X, y)
>>> model = dc.models.MultitaskRegressor(1, 5)
>>> transformers = []

Then you can evaluate this model as follows >>> import sklearn >>> evaluator = Evaluator(model, dataset, transformers) >>> multitask_scores = evaluator.compute_model_performance( … sklearn.metrics.mean_absolute_error)

Evaluators can also be used with dc.metrics.Metric objects as well in case you want to customize your metric further.

>>> evaluator = Evaluator(model, dataset, transformers)
>>> metric = dc.metrics.Metric(dc.metrics.mae_score)
>>> multitask_scores = evaluator.compute_model_performance(metric)

__init__(model, dataset: deepchem.data.datasets.Dataset, transformers: List[transformers.Transformer])[source]¶

Initialize this evaluator

Parameters

model (Model) – Model to evaluate. Note that this must be a regression or classification model and not a generative model.
dataset (Dataset) – Dataset object to evaluate model on.
transformers (List[Transformer]) – List of dc.trans.Transformer objects. These transformations must have been applied to dataset previously. The dataset will be untransformed for metric evaluation.

output_statistics(scores: Dict[str, float], stats_out: str)[source]¶

Write computed stats to file.

Parameters

scores (dict) – Dictionary mapping names of metrics to scores.
stats_out (str) – Name of file to write scores to.

output_predictions(y_preds: numpy.ndarray, csv_out: str)[source]¶

Writes predictions to file.

Writes predictions made on self.dataset to a specified file on disk. self.dataset.ids are used to format predictions.

Parameters

y_preds (np.ndarray) – Predictions to output
csv_out (str) – Name of file to write predictions to.

compute_model_performance(metrics: Union[deepchem.metrics.metric.Metric, Callable[[...], Any], List[deepchem.metrics.metric.Metric], List[Callable[[...], Any]]], csv_out: Optional[str] = None, stats_out: Optional[str] = None, per_task_metrics: bool = False, use_sample_weights: bool = False, n_classes: int = 2) → Union[Dict[str, float], Tuple[Dict[str, float], Dict[str, float]]][source]¶

Computes statistics of model on test data and saves results to csv.

Parameters

metrics (dc.metrics.Metric/list[dc.metrics.Metric]/function) – The set of metrics provided. This class attempts to do some intelligent handling of input. If a single dc.metrics.Metric object is provided or a list is provided, it will evaluate self.model on these metrics. If a function is provided, it is assumed to be a metric function that this method will attempt to wrap in a dc.metrics.Metric object. A metric function must accept two arguments, y_true, y_pred both of which are np.ndarray objects and return a floating point score. The metric function may also accept a keyword argument sample_weight to account for per-sample weights.
csv_out (str, optional (DEPRECATED)) – Filename to write CSV of model predictions.
stats_out (str, optional (DEPRECATED)) – Filename to write computed statistics.
per_task_metrics (bool, optional) – If true, return computed metric for each task on multitask dataset.
use_sample_weights (bool, optional (default False)) – If set, use per-sample weights w.
n_classes (int, optional (default None)) – If specified, will use n_classes as the number of unique classes in self.dataset. Note that this argument will be ignored for regression metrics.

Returns

multitask_scores (dict) – Dictionary mapping names of metrics to metric scores.
all_task_scores (dict, optional) – If per_task_metrics == True, then returns a second dictionary of scores for each task separately.

class GeneratorEvaluator(model, generator: Iterable[Tuple[Any, Any, Any]], transformers: List[transformers.Transformer], labels: Optional[List] = None, weights: Optional[List] = None)[source]¶

Evaluate models on a stream of data.

This class is a partner class to Evaluator. Instead of operating over datasets this class operates over a generator which yields batches of data to feed into provided model.

Examples

>>> import deepchem as dc
>>> import numpy as np
>>> X = np.random.rand(10, 5)
>>> y = np.random.rand(10, 1)
>>> dataset = dc.data.NumpyDataset(X, y)
>>> model = dc.models.MultitaskRegressor(1, 5)
>>> generator = model.default_generator(dataset, pad_batches=False)
>>> transformers = []

Then you can evaluate this model as follows

>>> import sklearn
>>> evaluator = GeneratorEvaluator(model, generator, transformers)
>>> multitask_scores = evaluator.compute_model_performance(
...     sklearn.metrics.mean_absolute_error)

Evaluators can also be used with dc.metrics.Metric objects as well in case you want to customize your metric further. (Note that a given generator can only be used once so we have to redefine the generator here.)

>>> generator = model.default_generator(dataset, pad_batches=False)
>>> evaluator = GeneratorEvaluator(model, generator, transformers)
>>> metric = dc.metrics.Metric(dc.metrics.mae_score)
>>> multitask_scores = evaluator.compute_model_performance(metric)

__init__(model, generator: Iterable[Tuple[Any, Any, Any]], transformers: List[transformers.Transformer], labels: Optional[List] = None, weights: Optional[List] = None)[source]¶

Parameters

model (Model) – Model to evaluate.
generator (generator) – Generator which yields batches to feed into the model. For a KerasModel, it should be a tuple of the form (inputs, labels, weights). The “correct” way to create this generator is to use model.default_generator as shown in the example above.
transformers (List[Transformer]) – Tranformers to “undo” when applied to the models outputs
labels (list of Layer) – layers which are keys in the generator to compare to outputs
weights (list of Layer) – layers which are keys in the generator for weight matrices

compute_model_performance(metrics: Union[deepchem.metrics.metric.Metric, Callable[[...], Any], List[deepchem.metrics.metric.Metric], List[Callable[[...], Any]]], per_task_metrics: bool = False, use_sample_weights: bool = False, n_classes: int = 2) → Union[Dict[str, float], Tuple[Dict[str, float], Dict[str, float]]][source]¶

Computes statistics of model on test data and saves results to csv.

Parameters

metrics (dc.metrics.Metric/list[dc.metrics.Metric]/function) – The set of metrics provided. This class attempts to do some intelligent handling of input. If a single dc.metrics.Metric object is provided or a list is provided, it will evaluate self.model on these metrics. If a function is provided, it is assumed to be a metric function that this method will attempt to wrap in a dc.metrics.Metric object. A metric function must accept two arguments, y_true, y_pred both of which are np.ndarray objects and return a floating point score.
per_task_metrics (bool, optional) – If true, return computed metric for each task on multitask dataset.
use_sample_weights (bool, optional (default False)) – If set, use per-sample weights w.
n_classes (int, optional (default None)) – If specified, will assume that all metrics are classification metrics and will use n_classes as the number of unique classes in self.dataset.

Returns

multitask_scores (dict) – Dictionary mapping names of metrics to metric scores.
all_task_scores (dict, optional) – If per_task_metrics == True, then returns a second dictionary of scores for each task separately.

relative_difference(x: numpy.ndarray, y: numpy.ndarray) → numpy.ndarray[source]¶

Compute the relative difference between x and y

The two argument arrays must have the same shape.

Parameters

x (np.ndarray) – First input array
y (np.ndarray) – Second input array

Returns

z – We will have z == np.abs(x-y) / np.abs(max(x, y)).

Return type

np.ndarray

Genomic Utilities¶

seq_one_hot_encode(sequences, letters: str = 'ATCGN') → numpy.ndarray[source]¶

One hot encodes list of genomic sequences.

Sequences encoded have shape (N_sequences, N_letters, sequence_length, 1). These sequences will be processed as images with one color channel.

Parameters

sequences (np.ndarray or Iterator[Bio.SeqRecord]) – Iterable object of genetic sequences
letters (str, optional (default "ATCGN")) – String with the set of possible letters in the sequences.

Raises

ValueError: – If sequences are of different lengths.

Returns

A numpy array of shape (N_sequences, N_letters, sequence_length, 1).

Return type

np.ndarray

encode_bio_sequence(fname: str, file_type: str = 'fasta', letters: str = 'ATCGN') → numpy.ndarray[source]¶

Loads a sequence file and returns an array of one-hot sequences.

Parameters

fname (str) – Filename of fasta file.
file_type (str, optional (default "fasta")) – The type of file encoding to process, e.g. fasta or fastq, this is passed to Biopython.SeqIO.parse.
letters (str, optional (default "ATCGN")) – The set of letters that the sequences consist of, e.g. ATCG.

Returns

A numpy array of shape (N_sequences, N_letters, sequence_length, 1).

Return type

np.ndarray

Notes

This function requires BioPython to be installed.

hhblits(dataset_path, database=None, data_dir=None, evalue=0.001, num_iterations=2, num_threads=4)[source]¶

Run hhblits multisequence alignment search on a dataset. This function requires the hhblits binary to be installed and in the path. This function also requires a Hidden Markov Model reference database to be provided. Both can be found here: https://github.com/soedinglab/hh-suite

The database should be in the deepchem data directory or specified as an argument. To set the deepchem data directory, run this command in your environment:

export DEEPCHEM_DATA_DIR=<path to data directory>

Parameters

dataset_path (str) – Path to single sequence or multiple sequence alignment (MSA) dataset. Results will be saved in this directory.
database (str) – Name of database to search against. Note this is not the path, but the name of the database.
data_dir (str) – Path to database directory.
evalue (float) – E-value cutoff.
num_iterations (int) – Number of iterations.
num_threads (int) – Number of threads.

Returns

results (.a3m file) – MSA file containing the results of the hhblits search.
results (.hhr file) – hhsuite results file containing the results of the hhblits search.

Examples

>>> from deepchem.utils.sequence_utils import hhblits
>>> msa_path = hhblits('test/data/example.fasta', database='example_db', data_dir='test/data/', evalue=0.001, num_iterations=2, num_threads=4)

hhsearch(dataset_path, database=None, data_dir=None, evalue=0.001, num_iterations=2, num_threads=4)[source]¶

Run hhsearch multisequence alignment search on a dataset. This function requires the hhblits binary to be installed and in the path. This function also requires a Hidden Markov Model reference database to be provided. Both can be found here: https://github.com/soedinglab/hh-suite

The database should be in the deepchem data directory or specified as an argument. To set the deepchem data directory, run this command in your environment:

export DEEPCHEM_DATA_DIR=<path to data directory>

Examples

>>> from deepchem.utils.sequence_utils import hhsearch
>>> msa_path = hhsearch('test/data/example.fasta', database='example_db', data_dir='test/data/', evalue=0.001, num_iterations=2, num_threads=4)

Parameters

dataset_path (str) – Path to multiple sequence alignment dataset. Results will be saved in this directory.
database (str) – Name of database to search against. Note this is not the path, but the name of the database.
data_dir (str) – Path to database directory.
evalue (float) – E-value cutoff.
num_iterations (int) – Number of iterations.
num_threads (int) – Number of threads.

Returns

results (.a3m file) – MSA file containing the results of the hhblits search.
results (.hhr file) – hhsuite results file containing the results of the hhblits search.

MSA_to_dataset(msa_path)[source]¶: Convert a multiple sequence alignment to a NumpyDataset object.

Geometry Utilities¶

unit_vector(vector: numpy.ndarray) → numpy.ndarray[source]¶

Returns the unit vector of the vector.

Parameters: vector (np.ndarray) – A numpy array of shape (3,), where 3 is (x,y,z).
Returns: A numpy array of shape (3,). The unit vector of the input vector.
Return type: np.ndarray

angle_between(vector_i: numpy.ndarray, vector_j: numpy.ndarray) → float[source]¶

Returns the angle in radians between vectors “vector_i” and “vector_j”

Note that this function always returns the smaller of the two angles between the vectors (value between 0 and pi).

Parameters

vector_i (np.ndarray) – A numpy array of shape (3,), where 3 is (x,y,z).
vector_j (np.ndarray) – A numpy array of shape (3,), where 3 is (x,y,z).

Returns

The angle in radians between the two vectors.

Return type

np.ndarray

Examples

>>> print("%0.06f" % angle_between((1, 0, 0), (0, 1, 0)))
1.570796
>>> print("%0.06f" % angle_between((1, 0, 0), (1, 0, 0)))
0.000000
>>> print("%0.06f" % angle_between((1, 0, 0), (-1, 0, 0)))
3.141593

generate_random_unit_vector() → numpy.ndarray[source]¶

Generate a random unit vector on the sphere S^2.

Citation: http://mathworld.wolfram.com/SpherePointPicking.html

Pseudocode:

Choose random theta element [0, 2*pi]
Choose random z element [-1, 1]
Compute output vector u: (x,y,z) = (sqrt(1-z^2)*cos(theta), sqrt(1-z^2)*sin(theta),z)

Returns: u – A numpy array of shape (3,). u is an unit vector
Return type: np.ndarray

generate_random_rotation_matrix() → numpy.ndarray[source]¶

Generates a random rotation matrix.

Generate a random unit vector u, randomly sampled from the

unit sphere (see function generate_random_unit_vector() for details)

Generate a second random unit vector v

If absolute value of u dot v > 0.99, repeat.

(This is important for numerical stability. Intuition: we want them to be as linearly independent as possible or else the orthogonalized version of v will be much shorter in magnitude compared to u. I assume in Stack they took this from Gram-Schmidt orthogonalization?)

v” = v - (u dot v)*u, i.e. subtract out the component of

v that’s in u’s direction

normalize v” (this isn”t in Stack but I assume it must be

done)

find w = u cross v”
u, v”, and w will form the columns of a rotation matrix, R.

The intuition is that u, v” and w are, respectively, what the standard basis vectors e1, e2, and e3 will be mapped to under the transformation.

Returns: R – A numpy array of shape (3, 3). R is a rotation matrix.
Return type: np.ndarray

is_angle_within_cutoff(vector_i: numpy.ndarray, vector_j: numpy.ndarray, angle_cutoff: float) → bool[source]¶

A utility function to compute whether two vectors are within a cutoff from 180 degrees apart.

Parameters

vector_i (np.ndarray) – A numpy array of shape (3,)`, where 3 is (x,y,z).
vector_j (np.ndarray) – A numpy array of shape (3,), where 3 is (x,y,z).
cutoff (float) – The deviation from 180 (in degrees)

Returns

Whether two vectors are within a cutoff from 180 degrees apart

Return type

bool

Graph Utilities¶

fourier_encode_dist(x, num_encodings=4, include_self=True)[source]¶

Fourier encode the input tensor x based on the specified number of encodings.

This function applies a Fourier encoding to the input tensor x by dividing it by a range of scales (2^i for i in range(num_encodings)) and then concatenating the sine and cosine of the scaled values. Optionally, the original input tensor can be included in the output.

Parameters

x (torch.Tensor) – Input tensor to be Fourier encoded.
num_encodings (int, optional, default=4) – Number of Fourier encodings to apply.
include_self (bool, optional, default=True) – Whether to include the original input tensor in the output.

Returns

Fourier encoded tensor.

Return type

torch.Tensor

Examples

>>> import torch
>>> x = torch.tensor([1.0, 2.0, 3.0])
>>> encoded_x = fourier_encode_dist(x, num_encodings=4, include_self=True)

aggregate_mean(h, **kwargs)[source]¶

Compute the mean of the input tensor along the second to last dimension.

Parameters: h (torch.Tensor) – Input tensor.
Returns: Mean of the input tensor along the second to last dimension.
Return type: torch.Tensor

aggregate_max(h, **kwargs)[source]¶

Compute the max of the input tensor along the second to last dimension.

Parameters: h (torch.Tensor) – Input tensor.
Returns: Max of the input tensor along the second to last dimension.
Return type: torch.Tensor

aggregate_min(h, **kwargs)[source]¶

Compute the min of the input tensor along the second to last dimension.

Parameters

h (torch.Tensor) – Input tensor.
**kwargs – Additional keyword arguments.

Returns

Min of the input tensor along the second to last dimension.

Return type

torch.Tensor

aggregate_std(h, **kwargs)[source]¶

Compute the standard deviation of the input tensor along the second to last dimension.

Parameters: h (torch.Tensor) – Input tensor.
Returns: Standard deviation of the input tensor along the second to last dimension.
Return type: torch.Tensor

aggregate_var(h, **kwargs)[source]¶

Compute the variance of the input tensor along the second to last dimension.

Parameters: h (torch.Tensor) – Input tensor.
Returns: Variance of the input tensor along the second to last dimension.
Return type: torch.Tensor

aggregate_moment(h, n=3, **kwargs)[source]¶

Compute the nth moment of the input tensor along the second to last dimension.

Parameters

h (torch.Tensor) – Input tensor.
n (int, optional, default=3) – The order of the moment to compute.

Returns

Nth moment of the input tensor along the second to last dimension.

Return type

torch.Tensor

aggregate_sum(h, **kwargs)[source]¶

Compute the sum of the input tensor along the second to last dimension.

Parameters: h (torch.Tensor) – Input tensor.
Returns: Sum of the input tensor along the second to last dimension.
Return type: torch.Tensor

scale_identity(h, D=None, avg_d=None)[source]¶

Identity scaling function.

Parameters

h (torch.Tensor) – Input tensor.
D (torch.Tensor, optional) – Degree tensor.
avg_d (dict, optional) – Dictionary containing averages over the training set.

Returns

Scaled input tensor.

Return type

torch.Tensor

scale_amplification(h, D, avg_d)[source]¶

Amplification scaling function. log(D + 1) / d * h where d is the average of the log(D + 1) in the training set

Parameters

h (torch.Tensor) – Input tensor.
D (torch.Tensor) – Degree tensor.
avg_d (dict) – Dictionary containing averages over the training set.

Returns

Scaled input tensor.

Return type

torch.Tensor

scale_attenuation(h, D, avg_d)[source]¶

Attenuation scaling function. (log(D + 1))^-1 / d * X where d is the average of the log(D + 1))^-1 in the training set

Parameters

h (torch.Tensor) – Input tensor.
D (torch.Tensor) – Degree tensor.
avg_d (dict) – Dictionary containing averages over the training set.

Returns

Scaled input tensor.

Return type

torch.Tensor

Hash Function Utilities¶

hash_ecfp(ecfp: str, size: int = 1024) → int[source]¶

Returns an int < size representing given ECFP fragment.

Input must be a string. This utility function is used for various ECFP based fingerprints.

Parameters

ecfp (str) – String to hash. Usually an ECFP fragment.
size (int, optional (default 1024)) – Hash to an int in range [0, size)

Returns

ecfp_hash – An int < size representing given ECFP fragment

Return type

int

hash_ecfp_pair(ecfp_pair: Tuple[str, str], size: int = 1024) → int[source]¶

Returns an int < size representing that ECFP pair.

Input must be a tuple of strings. This utility is primarily used for spatial contact featurizers. For example, if a protein and ligand have close contact region, the first string could be the protein’s fragment and the second the ligand’s fragment. The pair could be hashed together to achieve one hash value for this contact region.

Parameters

ecfp_pair (Tuple[str, str]) – Pair of ECFP fragment strings
size (int, optional (default 1024)) – Hash to an int in range [0, size)

Returns

ecfp_hash – An int < size representing given ECFP pair.

Return type

int

vectorize(hash_function: Callable[[Any, int], int], feature_dict: Optional[Dict[int, str]] = None, size: int = 1024, feature_list: Optional[List] = None) → numpy.ndarray[source]¶

Helper function to vectorize a spatial description from a hash.

Hash functions are used to perform spatial featurizations in DeepChem. However, it’s necessary to convert backwards from the hash function to feature vectors. This function aids in this conversion procedure. It creates a vector of zeros of length size. It then loops through feature_dict, uses hash_function to hash the stored value to an integer in range [0, size) and bumps that index.

Parameters

hash_function (Function, Callable[[str, int], int]) – Should accept two arguments, feature, and size and return a hashed integer. Here feature is the item to hash, and size is an int. For example, if size=1024, then hashed values must fall in range [0, 1024).
feature_dict (Dict, optional (default None)) – Maps unique keys to features computed.
size (int (default 1024)) – Length of generated bit vector
feature_list (List, optional (default None)) – List of features.

Returns

feature_vector – A numpy array of shape (size,)

Return type

np.ndarray

Voxel Utils¶

convert_atom_to_voxel(coordinates: numpy.ndarray, atom_index: int, box_width: float, voxel_width: float) → numpy.ndarray[source]¶

Converts atom coordinates to an i,j,k grid index.

This function offsets molecular atom coordinates by (box_width/2, box_width/2, box_width/2) and then divides by voxel_width to compute the voxel indices.

Parameters

coordinates (np.ndarray) – Array with coordinates of all atoms in the molecule, shape (N, 3).
atom_index (int) – Index of an atom in the molecule.
box_width (float) – Size of the box in Angstroms.
voxel_width (float) – Size of a voxel in Angstroms

Returns

indices – A 1D numpy array of length 3 with [i, j, k], the voxel coordinates of specified atom.

Return type

np.ndarray

convert_atom_pair_to_voxel(coordinates_tuple: Tuple[numpy.ndarray, numpy.ndarray], atom_index_pair: Tuple[int, int], box_width: float, voxel_width: float) → numpy.ndarray[source]¶

Converts a pair of atoms to i,j,k grid indexes.

Parameters

coordinates_tuple (Tuple[np.ndarray, np.ndarray]) – A tuple containing two molecular coordinate arrays of shapes (N, 3) and (M, 3).
atom_index_pair (Tuple[int, int]) – A tuple of indices for the atoms in the two molecules.
box_width (float) – Size of the box in Angstroms.
voxel_width (float) – Size of a voxel in Angstroms

Returns

indices_list – A numpy array of shape (2, 3), where 3 is [i, j, k] of the voxel coordinates of specified atom.

Return type

np.ndarray

voxelize(get_voxels: Callable[[...], Any], coordinates: Any, box_width: float = 16.0, voxel_width: float = 1.0, hash_function: Optional[Callable[[...], Any]] = None, feature_dict: Optional[Dict[Any, Any]] = None, feature_list: Optional[List[Union[int, Tuple[int]]]] = None, nb_channel: int = 16, dtype: str = 'int') → numpy.ndarray[source]¶

Helper function to voxelize inputs.

This helper function helps convert a hash function which specifies spatial features of a molecular complex into a voxel tensor. This utility is used by various featurizers that generate voxel grids.

Parameters

get_voxels (Function) – Function that voxelizes inputs
coordinates (Any) – Contains the 3D coordinates of a molecular system. This should have whatever type get_voxels() expects as its first argument.
box_width (float, optional (default 16.0)) – Size of a box in which voxel features are calculated. Box is centered on a ligand centroid.
voxel_width (float, optional (default 1.0)) – Size of a 3D voxel in a grid in Angstroms.
hash_function (Function) – Used to map feature choices to voxel channels.
feature_dict (Dict, optional (default None)) – Keys are atom indices or tuples of atom indices, the values are computed features. If hash_function is not None, then the values are hashed using the hash function into [0, nb_channels) and this channel at the voxel for the given key is incremented by 1 for each dictionary entry. If hash_function is None, then the value must be a vector of size (n_channels,) which is added to the existing channel values at that voxel grid.
feature_list (List, optional (default None)) – List of atom indices or tuples of atom indices. This can only be used if nb_channel==1. Increments the voxels corresponding to these indices by 1 for each entry.
nb_channel (int, , optional (default 16)) – The number of feature channels computed per voxel. Should be a power of 2.
dtype (str ('int' or 'float'), optional (default 'int')) – The type of the numpy ndarray created to hold features.

Returns

feature_tensor – The voxel of the input with the shape (voxels_per_edge, voxels_per_edge, voxels_per_edge, nb_channel).

Return type

np.ndarray

Graph Convolution Utilities¶

one_hot_encode(val: Union[int, str], allowable_set: Union[List[str], List[int]], include_unknown_set: bool = False) → List[float][source]¶

One hot encoder for elements of a provided set.

Examples

>>> one_hot_encode("a", ["a", "b", "c"])
[1.0, 0.0, 0.0]
>>> one_hot_encode(2, [0, 1, 2])
[0.0, 0.0, 1.0]
>>> one_hot_encode(3, [0, 1, 2])
[0.0, 0.0, 0.0]
>>> one_hot_encode(3, [0, 1, 2], True)
[0.0, 0.0, 0.0, 1.0]

Parameters

val (int or str) – The value must be present in allowable_set.
allowable_set (List[int] or List[str]) – List of allowable quantities.
include_unknown_set (bool, default False) – If true, the index of all values not in allowable_set is len(allowable_set).

Returns

An one-hot vector of val. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

Raises

ValueError – If include_unknown_set is False and val is not in allowable_set.

get_atom_type_one_hot(atom: Any, allowable_set: List[str] = ['C', 'N', 'O', 'F', 'P', 'S', 'Cl', 'Br', 'I'], include_unknown_set: bool = True) → List[float][source]¶

Get an one-hot feature of an atom type.

Parameters

atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
allowable_set (List[str]) – The atom types to consider. The default set is [“C”, “N”, “O”, “F”, “P”, “S”, “Cl”, “Br”, “I”].
include_unknown_set (bool, default True) – If true, the index of all atom not in allowable_set is len(allowable_set).

Returns

An one-hot vector of atom types. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

construct_hydrogen_bonding_info(mol: Any) → List[Tuple[int, str]][source]¶

Construct hydrogen bonding infos about a molecule.

Parameters: mol (rdkit.Chem.rdchem.Mol) – RDKit mol object
Returns: A list of tuple (atom_index, hydrogen_bonding_type). The hydrogen_bonding_type value is “Acceptor” or “Donor”.
Return type: List[Tuple[int, str]]

get_atom_hydrogen_bonding_one_hot(atom: Any, hydrogen_bonding: List[Tuple[int, str]]) → List[float][source]¶

Get an one-hot feat about whether an atom accepts electrons or donates electrons.

Parameters

atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
hydrogen_bonding (List[Tuple[int, str]]) – The return value of construct_hydrogen_bonding_info. The value is a list of tuple (atom_index, hydrogen_bonding) like (1, “Acceptor”).

Returns

A one-hot vector of the ring size type. The first element indicates “Donor”, and the second element indicates “Acceptor”.

Return type

List[float]

get_atom_is_in_aromatic_one_hot(atom: Any) → List[float][source]¶

Get ans one-hot feature about whether an atom is in aromatic system or not.

Parameters: atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
Returns: A vector of whether an atom is in aromatic system or not.
Return type: List[float]

get_atom_hybridization_one_hot(atom: Any, allowable_set: List[str] = ['SP', 'SP2', 'SP3'], include_unknown_set: bool = False) → List[float][source]¶

Get an one-hot feature of hybridization type.

Parameters

atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
allowable_set (List[str]) – The hybridization types to consider. The default set is [“SP”, “SP2”, “SP3”]
include_unknown_set (bool, default False) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

An one-hot vector of the hybridization type. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

get_atom_total_num_Hs_one_hot(atom: Any, allowable_set: List[int] = [0, 1, 2, 3, 4], include_unknown_set: bool = True) → List[float][source]¶

Get an one-hot feature of the number of hydrogens which an atom has.

Parameters

atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
allowable_set (List[int]) – The number of hydrogens to consider. The default set is [0, 1, …, 4]
include_unknown_set (bool, default True) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

A one-hot vector of the number of hydrogens which an atom has. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

get_atom_chirality_one_hot(atom: Any) → List[float][source]¶

Get an one-hot feature about an atom chirality type.

Parameters: atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
Returns: A one-hot vector of the chirality type. The first element indicates “R”, and the second element indicates “S”.
Return type: List[float]

get_atom_formal_charge(atom: Any) → List[float][source]¶

Get a formal charge of an atom.

Parameters: atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
Returns: A vector of the formal charge.
Return type: List[float]

get_atom_partial_charge(atom: Any) → List[float][source]¶

Get a partial charge of an atom.

Parameters: atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
Returns: A vector of the parital charge.
Return type: List[float]

Notes

Before using this function, you must calculate GasteigerCharge like AllChem.ComputeGasteigerCharges(mol).

get_atom_total_degree_one_hot(atom: Any, allowable_set: List[int] = [0, 1, 2, 3, 4, 5], include_unknown_set: bool = True) → List[float][source]¶

Get an one-hot feature of the degree which an atom has.

Parameters

atom (rdkit.Chem.rdchem.Atom) – RDKit atom object
allowable_set (List[int]) – The degree to consider. The default set is [0, 1, …, 5]
include_unknown_set (bool, default True) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

A one-hot vector of the degree which an atom has. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

get_bond_type_one_hot(bond: Any, allowable_set: List[str] = ['SINGLE', 'DOUBLE', 'TRIPLE', 'AROMATIC'], include_unknown_set: bool = False) → List[float][source]¶

Get an one-hot feature of bond type.

Parameters

bond (rdkit.Chem.rdchem.Bond) – RDKit bond object
allowable_set (List[str]) – The bond types to consider. The default set is [“SINGLE”, “DOUBLE”, “TRIPLE”, “AROMATIC”].
include_unknown_set (bool, default False) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

A one-hot vector of the bond type. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

get_bond_is_in_same_ring_one_hot(bond: Any) → List[float][source]¶

Get an one-hot feature about whether atoms of a bond is in the same ring or not.

Parameters: bond (rdkit.Chem.rdchem.Bond) – RDKit bond object
Returns: A one-hot vector of whether a bond is in the same ring or not.
Return type: List[float]

get_bond_is_conjugated_one_hot(bond: Any) → List[float][source]¶

Get an one-hot feature about whether a bond is conjugated or not.

Parameters: bond (rdkit.Chem.rdchem.Bond) – RDKit bond object
Returns: A one-hot vector of whether a bond is conjugated or not.
Return type: List[float]

get_bond_stereo_one_hot(bond: Any, allowable_set: List[str] = ['STEREONONE', 'STEREOANY', 'STEREOZ', 'STEREOE'], include_unknown_set: bool = True) → List[float][source]¶

Get an one-hot feature of the stereo configuration of a bond.

Parameters

bond (rdkit.Chem.rdchem.Bond) – RDKit bond object
allowable_set (List[str]) – The stereo configuration types to consider. The default set is [“STEREONONE”, “STEREOANY”, “STEREOZ”, “STEREOE”].
include_unknown_set (bool, default True) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

A one-hot vector of the stereo configuration of a bond. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

get_bond_graph_distance_one_hot(bond: Any, graph_dist_matrix: numpy.ndarray, allowable_set: List[int] = [1, 2, 3, 4, 5, 6, 7], include_unknown_set: bool = True) → List[float][source]¶

Get an one-hot feature of graph distance.

Parameters

bond (rdkit.Chem.rdchem.Bond) – RDKit bond object
graph_dist_matrix (np.ndarray) – The return value of Chem.GetDistanceMatrix(mol). The shape is (num_atoms, num_atoms).
allowable_set (List[int]) – The graph distance types to consider. The default set is [1, 2, …, 7].
include_unknown_set (bool, default False) – If true, the index of all types not in allowable_set is len(allowable_set).

Returns

A one-hot vector of the graph distance. If include_unknown_set is False, the length is len(allowable_set). If include_unknown_set is True, the length is len(allowable_set) + 1.

Return type

List[float]

Grover Utilities¶

extract_grover_attributes(molgraph: deepchem.feat.graph_data.BatchGraphData)[source]¶

Utility to extract grover attributes for grover model

Parameters: molgraph (BatchGraphData) – A batched graph data representing a collection of molecules.
Returns: graph_attributes – A tuple containing atom features, bond features, atom to bond mapping, bond to atom mapping, bond to reverse bond mapping, atom to atom mapping, atom scope, bond scope, functional group labels and other additional features.
Return type: Tuple

Example

>>> import deepchem as dc
>>> from deepchem.feat.graph_data import BatchGraphData
>>> smiles = ['CC', 'CCC', 'CC(=O)C']
>>> featurizer = dc.feat.GroverFeaturizer(features_generator=dc.feat.CircularFingerprint())
>>> graphs = featurizer.featurize(smiles)
>>> molgraph = BatchGraphData(graphs)
>>> attributes = extract_grover_attributes(molgraph)

Debug Utilities¶

Docking Utilities¶

These utilities assist in file preparation and processing for molecular docking.

write_vina_conf(protein_filename: str, ligand_filename: str, centroid: numpy.ndarray, box_dims: numpy.ndarray, conf_filename: str, num_modes: int = 9, exhaustiveness: Optional[int] = None) → None[source]¶

Writes Vina configuration file to disk.

Autodock Vina accepts a configuration file which provides options under which Vina is invoked. This utility function writes a vina configuration file which directs Autodock vina to perform docking under the provided options.

Parameters

protein_filename (str) – Filename for protein
ligand_filename (str) – Filename for the ligand
centroid (np.ndarray) – A numpy array with shape (3,) holding centroid of system
box_dims (np.ndarray) – A numpy array of shape (3,) holding the size of the box to dock
conf_filename (str) – Filename to write Autodock Vina configuration to.
num_modes (int, optional (default 9)) – The number of binding modes Autodock Vina should find
exhaustiveness (int, optional) – The exhaustiveness of the search to be performed by Vina

write_gnina_conf(protein_filename: str, ligand_filename: str, conf_filename: str, num_modes: int = 9, exhaustiveness: Optional[int] = None, **kwargs) → None[source]¶

Writes GNINA configuration file to disk.

GNINA accepts a configuration file which provides options under which GNINA is invoked. This utility function writes a configuration file which directs GNINA to perform docking under the provided options.

Parameters

protein_filename (str) – Filename for protein
ligand_filename (str) – Filename for the ligand
conf_filename (str) – Filename to write Autodock Vina configuration to.
num_modes (int, optional (default 9)) – The number of binding modes GNINA should find
exhaustiveness (int, optional) – The exhaustiveness of the search to be performed by GNINA
kwargs – Args supported by GNINA documented here https://github.com/gnina/gnina#usage

load_docked_ligands(pdbqt_output: str) → Tuple[List[Any], List[float]][source]¶

This function loads ligands docked by autodock vina.

Autodock vina writes outputs to disk in a PDBQT file format. This PDBQT file can contain multiple docked “poses”. Recall that a pose is an energetically favorable 3D conformation of a molecule. This utility function reads and loads the structures for multiple poses from vina’s output file.

Parameters: pdbqt_output (str) – Should be the filename of a file generated by autodock vina’s docking software.
Returns: Tuple of molecules, scores. molecules is a list of rdkit molecules with 3D information. scores is the associated vina score.
Return type: Tuple[List[rdkit.Chem.rdchem.Mol], List[float]]

Notes

This function requires RDKit to be installed.

prepare_inputs(protein: str, ligand: str, replace_nonstandard_residues: bool = True, remove_heterogens: bool = True, remove_water: bool = True, add_hydrogens: bool = True, pH: float = 7.0, optimize_ligand: bool = True, pdb_name: Optional[str] = None) → Tuple[Any, Any][source]¶

This prepares protein-ligand complexes for docking.

Autodock Vina requires PDB files for proteins and ligands with sensible inputs. This function uses PDBFixer and RDKit to ensure that inputs are reasonable and ready for docking. Default values are given for convenience, but fixing PDB files is complicated and human judgement is required to produce protein structures suitable for docking. Always inspect the results carefully before trying to perform docking.

Parameters

protein (str) – Filename for protein PDB file or a PDBID.
ligand (str) – Either a filename for a ligand PDB file or a SMILES string.
replace_nonstandard_residues (bool (default True)) – Replace nonstandard residues with standard residues.
remove_heterogens (bool (default True)) – Removes residues that are not standard amino acids or nucleotides.
remove_water (bool (default True)) – Remove water molecules.
add_hydrogens (bool (default True)) – Add missing hydrogens at the protonation state given by pH.
pH (float (default 7.0)) – Most common form of each residue at given pH value is used.
optimize_ligand (bool (default True)) – If True, optimize ligand with RDKit. Required for SMILES inputs.
pdb_name (Optional[str]) – If given, write sanitized protein and ligand to files called “pdb_name.pdb” and “ligand_pdb_name.pdb”

Returns

Tuple of protein_molecule, ligand_molecule with 3D information.

Return type

Tuple[RDKitMol, RDKitMol]

Note

This function requires RDKit and OpenMM to be installed. Read more about PDBFixer here: https://github.com/openmm/pdbfixer.

Examples

>>> p, m = prepare_inputs('3cyx', 'CCC')

>> p.GetNumAtoms() >> m.GetNumAtoms()

>>> p, m = prepare_inputs('3cyx', 'CCC', remove_heterogens=False)

>> p.GetNumAtoms()

read_gnina_log(log_file: str) → numpy.ndarray[source]¶

Read GNINA logfile and get docking scores.

GNINA writes computed binding affinities to a logfile.

Parameters: log_file (str) – Filename of logfile generated by GNINA.
Returns: scores – Array of binding affinity (kcal/mol), CNN pose score, and CNN affinity for each binding mode.
Return type: np.array, dimension (num_modes, 3)

Print Threshold¶

The printing threshold controls how many dataset elements are printed when dc.data.Dataset objects are converted to strings or represnted in the IPython repl.

get_print_threshold() → int[source]¶

Return the printing threshold for datasets.

The print threshold is the number of elements from ids/tasks to print when printing representations of Dataset objects.

Returns: threshold – Number of elements that will be printed
Return type: int

set_print_threshold(threshold: int)[source]¶

Set print threshold

The print threshold is the number of elements from ids/tasks to print when printing representations of Dataset objects.

Parameters: threshold (int) – Number of elements to print.

get_max_print_size() → int[source]¶

Return the max print size for a dataset.

If a dataset is large, printing self.ids as part of a string representation can be very slow. This field controls the maximum size for a dataset before ids are no longer printed.

Returns: max_print_size – Maximum length of a dataset for ids to be printed in string representation.
Return type: int

set_max_print_size(max_print_size: int)[source]¶

Set max_print_size

If a dataset is large, printing self.ids as part of a string representation can be very slow. This field controls the maximum size for a dataset before ids are no longer printed.

Parameters: max_print_size (int) – Maximum length of a dataset for ids to be printed in string representation.

Fake Data Generator¶

The utilities here are used to generate random sample data which can be used for testing model architectures or other purposes.

class FakeGraphGenerator(min_nodes: int = 10, max_nodes: int = 10, n_node_features: int = 5, avg_degree: int = 4, n_edge_features: int = 3, n_classes: int = 2, task: str = 'graph', **kwargs)[source]¶

Generates a random graphs which can be used for testing or other purposes.

The generated graph supports both node-level and graph-level labels.

Example

>>> from deepchem.utils.fake_data_generator import FakeGraphGenerator
>>> fgg  = FakeGraphGenerator(min_nodes=8, max_nodes=10,  n_node_features=5, avg_degree=8, n_edge_features=3, n_classes=2, task='graph', z=5)
>>> graphs = fgg.sample(n_graphs=10)
>>> type(graphs)
<class 'deepchem.data.datasets.NumpyDataset'>
>>> type(graphs.X[0])
<class 'deepchem.feat.graph_data.GraphData'>
>>> len(graphs) == 10  # num_graphs
True

Note

The FakeGraphGenerator class is based on torch_geometric.dataset.FakeDataset class.

__init__(min_nodes: int = 10, max_nodes: int = 10, n_node_features: int = 5, avg_degree: int = 4, n_edge_features: int = 3, n_classes: int = 2, task: str = 'graph', **kwargs)[source]¶

Parameters

min_nodes (int, default 10) – Minimum number of permissible nodes in a graph
max_nodes (int, default 10) – Maximum number of permissible nodes in a graph
n_node_features (int, default 5) – Average number of node features in a graph
avg_degree (int, default 4) – Average degree of the graph (avg_degree should be a positive number greater than the min_nodes)
n_edge_features (int, default 3) – Average number of features in the edge
task (str, default 'graph') – Indicates node-level labels or graph-level labels
kwargs (optional) – Additional graph attributes and their shapes , e.g. global_features = 5

sample(n_graphs: int = 100) → deepchem.data.datasets.NumpyDataset[source]¶

Samples graphs

Parameters: n_graphs (int, default 100) – Number of graphs to generate
Returns: graphs – Generated Graphs
Return type: NumpyDataset

Electron Sampler¶

The utilities here are used to sample electrons in a given molecule and update it using monte carlo methods, which can be used for methods like Variational Monte Carlo, etc.

class ElectronSampler(central_value: numpy.ndarray, f: Callable[[numpy.ndarray], numpy.ndarray], batch_no: int = 10, x: numpy.ndarray = array([], dtype=float64), steps: int = 10, seed: Optional[int] = None, symmetric: bool = True, simultaneous: bool = True)[source]¶

This class enables to initialize electron’s position using gauss distribution around a nucleus and update using Markov Chain Monte-Carlo(MCMC) moves.

Using the probability obtained from the square of magnitude of wavefunction of a molecule/atom, MCMC steps can be performed to get the electron’s positions and further update the wavefunction. This method is primarily used in methods like Variational Monte Carlo to sample electrons around the nucleons. Sampling can be done in 2 ways: -Simultaneous: All the electrons’ positions are updated all at once.

-Single-electron: MCMC steps are performed only a particular electron, given their index value.

Further these moves can be done in 2 methods: -Symmetric: In this configuration, the standard deviation for all the steps are uniform.

-Asymmetric: In this configuration, the standard deviation are not uniform and typically the standard deviation is obtained a function like harmonic distances, etc.

Irrespective of these methods, the initialization is done uniformly around the respective nucleus and the number of electrons specified.

Example

>>> from deepchem.utils.electron_sampler import ElectronSampler
>>> test_f = lambda x: 2*np.log(np.random.uniform(low=0,high=1.0,size=np.shape(x)[0]))
>>> distribution=ElectronSampler(central_value=np.array([[1,1,3],[3,2,3]]),f=test_f,seed=0,batch_no=2,steps=1000,)
>>> distribution.gauss_initialize_position(np.array([[1],[2]]))

>> print(distribution.x) [[[[1.03528105 1.00800314 3.01957476]]

[[3.01900177 1.99697286 2.99793562]]

[[3.00821197 2.00288087 3.02908547]]]

[[[1.04481786 1.03735116 2.98045444]]

[[3.01522075 2.0024335 3.00887726]]

[[3.00667349 2.02988158 2.99589683]]]]

>>> distribution.move()
0.5115

>> print(distribution.x) [[[[-0.32441754 1.23330263 2.67927645]]

[[ 3.42250997 2.23617126 3.55806632]]

[[ 3.37491385 1.54374006 3.13575241]]]

[[[ 0.49067726 1.03987841 3.70277884]]

[[ 3.5631939 1.68703947 2.5685874 ]]

[[ 2.84560249 1.73998364 3.41274181]]]]

__init__(central_value: numpy.ndarray, f: Callable[[numpy.ndarray], numpy.ndarray], batch_no: int = 10, x: numpy.ndarray = array([], dtype=float64), steps: int = 10, seed: Optional[int] = None, symmetric: bool = True, simultaneous: bool = True)[source]¶

central_value: np.ndarray: Contains each nucleus’ coordinates in a 2D array. The shape of the array should be(number_of_nucleus,3).Ex: [[1,2,3],[3,4,5],..]
f:Callable[[np.ndarray],np.ndarray]: A function that should give the twice the log probability of wavefunction of the molecular system when called. Should taken in a 4D array of electron’s positions(x) as argument and return a numpy array containing the log probabilities of each batch.
batch_no: int, optional (default 10): Number of batches of the electron’s positions to be initialized.
x: np.ndarray, optional (default np.ndarray([])): Contains the electron’s coordinates in a 4D array. The shape of the array should be(batch_no,no_of_electrons,1,3). Can be a 1D empty array, when electron’s positions are yet to be initialized.
steps: int, optional (default 10): The number of MCMC steps to be performed when the moves are called.
seed: int, optional (default None): Random seed to use.
symmetric: bool, optional(default True): If true, symmetric moves will be used, else asymmetric moves will be followed.
simultaneous: bool, optional(default True): If true, MCMC steps will be performed on all the electrons, else only a single electron gets updated.

harmonic_mean(y: numpy.ndarray) → numpy.ndarray[source]¶

Calculates the harmonic mean of the value ‘y’ from the self.central value. The numpy array returned is typically scaled up to get the standard deviation matrix.

Parameters: y (np.ndarray) – Containing the data distribution. Shape of y should be (batch,no_of_electron,1,3)
Returns: Contains the harmonic mean of the data distribution of each batch. Shape of the array obtained (batch_no, no_of_electrons,1,1)
Return type: np.ndarray

log_prob_gaussian(y: numpy.ndarray, mu: numpy.ndarray, sigma: numpy.ndarray) → numpy.ndarray[source]¶

Calculates the log probability of a gaussian distribution, given the mean and standard deviation

Parameters

y (np.ndarray) – data for which the log normal distribution is to be found
mu (np.ndarray) – Means wrt which the log normal is calculated. Same shape as x or should be brodcastable to x
sigma (np.ndarray,) – The standard deviation of the log normal distribution. Same shape as x or should be brodcastable to x

Returns

Log probability of gaussian distribution, with the shape - (batch_no,).

Return type

np.ndarray

gauss_initialize_position(no_sample: numpy.ndarray, stddev: float = 0.02)[source]¶

Initializes the position around a central value as mean sampled from a gauss distribution and updates self.x. Parameters: ———- no_sample: np.ndarray,

Contains the number of samples to initialize under each mean. should be in the form [[3],[2]..], where here it means 3 samples and 2 samples around the first entry and second entry,respectively in self.central_value is taken.

stddev: float, optional (default 0.02): contains the stddev with which the electrons’ coordinates are initialized

move(stddev: float = 0.02, asymmetric_func: Optional[Callable[[numpy.ndarray], numpy.ndarray]] = None, index: Optional[int] = None) → float[source]¶

Performs Metropolis-Hasting move for self.x(electrons). The type of moves to be followed -(simultaneous or single-electron, symmetric or asymmetric) have been specified when calling the class. The self.x array is replaced with a new array at the end of each step containing the new electron’s positions.

asymmetric_func: Callable[[np.ndarray],np.ndarray], optional(default None): Should be specified for an asymmetric move.The function should take in only 1 argument- y: a numpy array wrt to which mean should be calculated. This function should return the mean for the asymmetric proposal. For ferminet, this function is the harmonic mean of the distance between the electron and the nucleus.
stddev: float, optional (default 0.02): Specifies the standard deviation in the case of symmetric moves and the scaling factor of the standard deviation matrix in the case of asymmetric moves.
index: int, optional (default None): Specifies the index of the electron to be updated in the case of a single electron move.

Returns: accepted move ratio of the MCMC steps.
Return type: float

Density Functional Theory Utilities¶

The utilites here are used to create an object that contains information about a system’s self-consistent iteration steps and other processes.

class KSCalc(qc: dqc.qccalc.base_qccalc.BaseQCCalc)[source]¶

Interface to DQC’s KS calculation.

Parameters: qc (BaseQCCalc) – object often acts as a wrapper around an engine class (from dqc.qccalc) that contains information about the self-consistent iterations.

References

Kasim, Muhammad F., and Sam M. Vinko. “Learning the exchange-correlation functional from nature with fully differentiable density functional theory.” Physical Review Letters 127.12 (2021): 126403. https://github.com/diffqc/dqc/blob/master/dqc/qccalc/ks.py

__init__(qc: dqc.qccalc.base_qccalc.BaseQCCalc)[source]¶

energy() → torch.Tensor[source]¶

Return type: The total energy of the Kohn-Sham calculation for a particular system.

aodmtot() → torch.Tensor[source]¶

Both interacting and non-interacting system’s total energy can be expressed in terms of the density matrix. The ground state properties of a system can be calculated by minimizing the energy w.r.t the density matrix.

Return type: The total density matrix in atomic orbital bases.

dens(rgrid: torch.Tensor) → torch.Tensor[source]¶

The ground state density n(r) of a system.

Parameters

rgrid (torch.Tensor) – Calculate integration grid using dqc.grid.

Returns

The total density profile in the given grid
Reference
———
https (//github.com/diffqc/dqc/blob/master/dqc/grid/base_grid.py)

force() → torch.Tensor[source]¶

The force on an atom is calculated as the gradient of energy with respect to the atomic position.

Return type: The force for each atom.

hashstr(s: str) → str[source]¶

Encodes the string into hashed format - hexadecimal digits.

Parameters: s (str) –