Source code for

import sys

from .csv_dataset import MoleculeCSVDataset
from ..utils import smiles_to_bigraph
from ...utils import get_download_dir, download, _get_dgl_url
from .... import backend as F
from ....base import dgl_warning
from ....contrib.deprecation import deprecated

    import pandas as pd
except ImportError:

[docs]class Tox21(MoleculeCSVDataset): """Tox21 dataset. The Toxicology in the 21st Century ( initiative created a public database measuring toxicity of compounds, which has been used in the 2014 Tox21 Data Challenge. The dataset contains qualitative toxicity measurements for 8014 compounds on 12 different targets, including nuclear receptors and stress response pathways. Each target results in a binary label. A common issue for multi-task prediction is that some datapoints are not labeled for all tasks. This is also the case for Tox21. In data pre-processing, we set non-existing labels to be 0 so that they can be placed in tensors and used for masking in loss computation. See examples below for more details. All molecules are converted into DGLGraphs. After the first-time construction, the DGLGraphs will be saved for reloading so that we do not need to reconstruct them everytime. Parameters ---------- smiles_to_graph: callable, str -> DGLGraph A function turning smiles into a DGLGraph. Default to :func:``. node_featurizer : callable, rdkit.Chem.rdchem.Mol -> dict Featurization for nodes like atoms in a molecule, which can be used to update ndata for a DGLGraph. Default to None. edge_featurizer : callable, rdkit.Chem.rdchem.Mol -> dict Featurization for edges like bonds in a molecule, which can be used to update edata for a DGLGraph. Default to None. load : bool Whether to load the previously pre-processed dataset or pre-process from scratch. ``load`` should be False when we want to try different graph construction and featurization methods and need to preprocess from scratch. Default to True. """ @deprecated('Import Tox21 from instead.', 'class') def __init__(self, smiles_to_graph=smiles_to_bigraph, node_featurizer=None, edge_featurizer=None, load=True): if 'pandas' not in sys.modules: dgl_warning("Please install pandas") self._url = 'dataset/tox21.csv.gz' data_path = get_download_dir() + '/tox21.csv.gz' download(_get_dgl_url(self._url), path=data_path) df = pd.read_csv(data_path) = df['mol_id'] df = df.drop(columns=['mol_id']) super(Tox21, self).__init__(df, smiles_to_graph, node_featurizer, edge_featurizer, "smiles", "tox21_dglgraph.bin", load=load) self._weight_balancing() def _weight_balancing(self): """Perform re-balancing for each task. It's quite common that the number of positive samples and the number of negative samples are significantly different. To compensate for the class imbalance issue, we can weight each datapoint in loss computation. In particular, for each task we will set the weight of negative samples to be 1 and the weight of positive samples to be the number of negative samples divided by the number of positive samples. If weight balancing is performed, one attribute will be affected: * self._task_pos_weights is set, which is a list of positive sample weights for each task. """ num_pos = F.sum(self.labels, dim=0) num_indices = F.sum(self.mask, dim=0) self._task_pos_weights = (num_indices - num_pos) / num_pos @property def task_pos_weights(self): """Get weights for positive samples on each task Returns ------- numpy.ndarray numpy array gives the weight of positive samples on all tasks """ return self._task_pos_weights