MolecularDiffusion.utils.geom_analyzer¶

Attributes¶

`EDGE_THRESHOLD`
`SCALE_FACTOR`
`allowed_bonds`
`bond_dict`
`bonds1`
`bonds2`
`bonds3`
`num2symbol`
`stdv`
`symbol2num`

Classes¶

`BasicMolecularMetrics`	Valid amongst all generated molecules
`Histogram_cont`
`Histogram_discrete`

Functions¶

`analyze_node_distribution`(mol_list)
`analyze_stability_for_molecules`(→ Tuple[Dict[str, ...)	Analyze the stability of a list of molecules.
`build_molecule`(positions, atom_types, atom_decoder)
`build_xae_molecule`(positions, atom_types, atom_decoder)	Returns a triplet (X, A, E): atom_types, adjacency matrix, edge_types
`check_connected`(am[, tol])
`check_consistency_bond_dictionaries`()
`check_quality`(cartesian_coordinates_tensor, ...)
`check_stability`(positions, atom_type)	Check the stability of a molecule based on atom positions and types.
`check_stability`(positions, atom_type)	Check the stability of a molecule based on atom positions and types.
`check_symmetric`(am[, tol])
`correct_edges`(data[, scale_factor])	Corrects the edges in a molecular grapSCALE_FACTORh based on covalent radii.
`create_pyg_graph`(cartesian_coordinates_tensor, ...[, ...])	Creates a PyTorch Geometric graph from given cartesian coordinates and atomic numbers.
`geom_predictor`(p, l[, margin1, limit_bonds_to_one])	p: atom pair (couple of str)
`get_bond_order`(atom1, atom2, distance[, check_exists])
`get_cutoffs`(z[, radii, mult])
`is_fully_connected`(edge_index, num_nodes)	Determines if the graph is fully connected.
`mol2smiles`(mol)
`normalize_histogram`(hist)
`save_xyz_tmp`(path, atom_type, position)
`single_bond_only`(threshold, length[, margin1])

Module Contents¶

class MolecularDiffusion.utils.geom_analyzer.BasicMolecularMetrics(ratio, dataset_smiles_list)¶

Bases: object

Valid amongst all generated molecules Uniqueness amongst valid molecules Novelty amongst unique molecules

compute_novelty(unique)¶

compute_uniqueness(valid)¶: valid: list of SMILES strings.

compute_validity(generated)¶: generated smiles

evaluate(generated)¶: generated: list of pairs (positions: n x 3, atom_types: n [int]) the positions and atom types should already be masked.

dataset_smiles_list¶

ratio¶

class MolecularDiffusion.utils.geom_analyzer.Histogram_cont(num_bins=100, range=(0.0, 13.0), name='histogram', ignore_zeros=False)¶

add(elements)¶

plot(save_path=None)¶

plot_both(hist_b, save_path=None, wandb_obj=None)¶

bins = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,...¶

ignore_zeros = False¶

name = 'histogram'¶

range = (0.0, 13.0)¶

class MolecularDiffusion.utils.geom_analyzer.Histogram_discrete(name='histogram')¶

add(elements)¶

normalize()¶

plot(save_path=None)¶

bins¶

name = 'histogram'¶

MolecularDiffusion.utils.geom_analyzer.analyze_node_distribution(mol_list)¶

MolecularDiffusion.utils.geom_analyzer.analyze_stability_for_molecules(molecule_list: Dict[str, torch.Tensor], atom_decoder: List[str], dataset_smiles_list: List[str], use_rdkit: bool = True, debug: bool = True) → Tuple[Dict[str, float], List[float] | None]¶

Analyze the stability of a list of molecules. If xyz cannot be converted to mol and smiles, the molecule is considered unstable. Then for unstable molecules, the number of bonds is checked against the allowed bonds. Only “stable” smiles are then used for the RDKit metrics.

Parameters: - molecule_list (Dict[str, torch.Tensor]): Dictionary containing ‘one_hot’, ‘x’, and ‘node_mask’ tensors. - atom_decoder (List[str]): List to decode atomic types to element symbols. - dataset_smiles_list (List[str]): List of reference SMILES strings from the dataset. - use_rdkit (bool): Whether to use RDKit for additional metrics. Default is True.

Returns: Tuple[Dict[str, float], Optional[List[float]]]: - Dictionary with molecule and atomic stability. - List of RDKit metrics if use_rdkit is True, otherwise None.

MolecularDiffusion.utils.geom_analyzer.build_molecule(positions, atom_types, atom_decoder)¶

MolecularDiffusion.utils.geom_analyzer.build_xae_molecule(positions, atom_types, atom_decoder)¶: Returns a triplet (X, A, E): atom_types, adjacency matrix, edge_types args: positions: N x 3 (already masked to keep final number nodes) atom_types: N returns: X: N (int) A: N x N (bool) (binary adjacency matrix) E: N x N (int) (bond type, 0 if no bond) such that A = E.bool()

MolecularDiffusion.utils.geom_analyzer.check_connected(am, tol=1e-08)¶

MolecularDiffusion.utils.geom_analyzer.check_consistency_bond_dictionaries()¶

MolecularDiffusion.utils.geom_analyzer.check_quality(cartesian_coordinates_tensor, atomic_numbers_tensor)¶

MolecularDiffusion.utils.geom_analyzer.check_stability(positions, atom_type)¶

Check the stability of a molecule based on atom positions and types.

Parameters: - positions (np.ndarray): An array of shape (N, 3) containing the 3D coordinates of the atoms. - atom_type (List[int]): A list of atom types corresponding to the positions. - atom_decoder (Dict[int, str]): A dictionary mapping atom type indices to atom symbols.

Returns: Tuple[bool, int, int]: A tuple containing a boolean indicating if the molecule is stable,

the number of stable bonds, and the total number of atoms.

MolecularDiffusion.utils.geom_analyzer.check_stability(positions, atom_type)¶

Check the stability of a molecule based on atom positions and types.

Returns: Tuple[bool, int, int]: A tuple containing a boolean indicating if the molecule is stable,

the number of stable bonds, and the total number of atoms.

MolecularDiffusion.utils.geom_analyzer.check_symmetric(am, tol=1e-08)¶

MolecularDiffusion.utils.geom_analyzer.correct_edges(data, scale_factor=1.3)¶

Corrects the edges in a molecular grapSCALE_FACTORh based on covalent radii. This function iterates over the nodes and their adjacent nodes in the given molecular graph data. It calculates the bond length between each pair of nodes and checks if it is within the allowed bond length threshold (sum of covalent radii plus relaxation factor). If the bond length is valid, the edge is kept; otherwise, it is removed.

Parameters: data (torch_geometric.data.Data): The input molecular graph data containing node features,

edge indices, and positions.

scale_factor (float): The scaling factor to apply to the covalent radii. Default is 1.3.

Returns: torch_geometric.data.Data: The corrected molecular graph data with updated edge indices.

MolecularDiffusion.utils.geom_analyzer.create_pyg_graph(cartesian_coordinates_tensor, atomic_numbers_tensor, xyz_filename=None, r=5.0)¶

Creates a PyTorch Geometric graph from given cartesian coordinates and atomic numbers. :param cartesian_coordinates_tensor: A tensor containing the cartesian coordinates of the atoms. :type cartesian_coordinates_tensor: torch.Tensor :param atomic_numbers_tensor: A tensor containing the atomic numbers of the atoms. :type atomic_numbers_tensor: torch.Tensor :param xyz_filename: The filename of the XYZ file. :type xyz_filename: str :param r: The radius within which to consider edges between nodes. Default is 5.0. :type r: float, optional

Returns:: A PyTorch Geometric Data object containing the graph representation of the molecule.
Return type:: torch_geometric.data.Data

MolecularDiffusion.utils.geom_analyzer.geom_predictor(p, l, margin1=5, limit_bonds_to_one=False)¶: p: atom pair (couple of str) l: bond length (float)

MolecularDiffusion.utils.geom_analyzer.get_bond_order(atom1, atom2, distance, check_exists=False)¶

MolecularDiffusion.utils.geom_analyzer.get_cutoffs(z, radii=ase.data.covalent_radii, mult=1)¶

MolecularDiffusion.utils.geom_analyzer.is_fully_connected(edge_index, num_nodes)¶

Determines if the graph is fully connected. :param edge_index: The edge indices of the graph. :type edge_index: torch.Tensor :param num_nodes: The number of nodes in the graph. :type num_nodes: int

Returns:: True if the graph is fully connected, False otherwise. int: The number of connected components in the graph.
Return type:: bool

MolecularDiffusion.utils.geom_analyzer.mol2smiles(mol)¶

MolecularDiffusion.utils.geom_analyzer.normalize_histogram(hist)¶

MolecularDiffusion.utils.geom_analyzer.save_xyz_tmp(path, atom_type, position)¶

MolecularDiffusion.utils.geom_analyzer.single_bond_only(threshold, length, margin1=5)¶

MolecularDiffusion.utils.geom_analyzer.EDGE_THRESHOLD = 2¶

MolecularDiffusion.utils.geom_analyzer.SCALE_FACTOR = 1.3¶

MolecularDiffusion.utils.geom_analyzer.allowed_bonds¶

MolecularDiffusion.utils.geom_analyzer.bond_dict¶

MolecularDiffusion.utils.geom_analyzer.bonds1¶

MolecularDiffusion.utils.geom_analyzer.bonds2¶

MolecularDiffusion.utils.geom_analyzer.bonds3¶

MolecularDiffusion.utils.geom_analyzer.num2symbol¶

MolecularDiffusion.utils.geom_analyzer.stdv¶

MolecularDiffusion.utils.geom_analyzer.symbol2num¶