# -*- coding: utf-8 -*-
#
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: Apache-2.0
#
# PCBA from MoleculeNet for the prediction of biological activities
import pandas as pd
from dgl.data.utils import get_download_dir, download, _get_dgl_url, extract_archive
from .csv_dataset import MoleculeCSVDataset
__all__ = ['PCBA']
[docs]class PCBA(MoleculeCSVDataset):
r"""PCBA from MoleculeNet for the prediction of biological activities
Quoting [1], "PubChem BioAssay (PCBA) is a database consisting of biological activities of
small molecules generated by high-throughput screening. We use a subset of PCBA, containing
128 bioassays measured over 400 thousand compounds, used by previous work to benchmark machine
learning methods."
References:
* [1] MoleculeNet: A Benchmark for Molecular Machine Learning.
* [2] Massively Multitask Networks for Drug Discovery.
* [3] DeepChem
Parameters
----------
smiles_to_graph: callable, str -> DGLGraph
A function turning a SMILES string into a DGLGraph. If None, it uses
:func:`dgllife.utils.SMILESToBigraph` by default.
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 False.
log_every : bool
Print a message every time ``log_every`` molecules are processed. Default to 1000.
cache_file_path : str
Path to the cached DGLGraphs, default to 'pcba_dglgraph.bin'.
n_jobs : int
The maximum number of concurrently running jobs for graph construction and featurization,
using joblib backend. Default to 1.
Examples
--------
>>> import torch
>>> from dgllife.data import PCBA
>>> from dgllife.utils import SMILESToBigraph, CanonicalAtomFeaturizer
>>> smiles_to_g = SMILESToBigraph(node_featurizer=CanonicalAtomFeaturizer())
>>> dataset = PCBA(smiles_to_g)
>>> # Get size of the dataset
>>> len(dataset)
437929
>>> # Get the 0th datapoint, consisting of SMILES, DGLGraph, labels, and masks
>>> dataset[0]
('CC(=O)N1CCC2(CC1)NC(=O)N(c1ccccc1)N2',
DGLGraph(num_nodes=20, num_edges=44,
ndata_schemes={'h': Scheme(shape=(74,), dtype=torch.float32)}
edata_schemes={}),
tensor([0., ..., 0.]),
tensor([1., ..., 0.]))
The dataset instance also contains information about molecule ids.
>>> dataset.ids[i]
We can also get the id along with SMILES, DGLGraph, labels, and masks at once.
>>> dataset.load_full = True
>>> dataset[0]
('CC(=O)N1CCC2(CC1)NC(=O)N(c1ccccc1)N2',
DGLGraph(num_nodes=20, num_edges=44,
ndata_schemes={'h': Scheme(shape=(74,), dtype=torch.float32)}
edata_schemes={}),
tensor([0., ..., 0.]),
tensor([1., ..., 0.]),
'CID1511280')
To address the imbalance between positive and negative samples, we can re-weight
positive samples for each task based on the training datapoints.
>>> train_ids = torch.arange(1000)
>>> dataset.task_pos_weights(train_ids)
tensor([7.3400, 489.0000, ..., 1.0000])
"""
def __init__(self,
smiles_to_graph=None,
node_featurizer=None,
edge_featurizer=None,
load=False,
log_every=1000,
cache_file_path='./pcba_dglgraph.bin',
n_jobs=1):
self._url = 'dataset/pcba.zip'
data_path = get_download_dir() + '/pcba.zip'
dir_path = get_download_dir() + '/pcba'
download(_get_dgl_url(self._url), path=data_path, overwrite=False)
extract_archive(data_path, dir_path)
df = pd.read_csv(dir_path + '/pcba.csv')
self.ids = df['mol_id'].tolist()
self.load_full = False
df = df.drop(columns=['mol_id'])
super(PCBA, self).__init__(df=df,
smiles_to_graph=smiles_to_graph,
node_featurizer=node_featurizer,
edge_featurizer=edge_featurizer,
smiles_column='smiles',
cache_file_path=cache_file_path,
load=load,
log_every=log_every,
init_mask=True,
n_jobs=n_jobs)
self.ids = [self.ids[i] for i in self.valid_ids]
[docs] def __getitem__(self, item):
"""Get datapoint with index
Parameters
----------
item : int
Datapoint index
Returns
-------
str
SMILES for the ith datapoint
DGLGraph
DGLGraph for the ith datapoint
Tensor of dtype float32 and shape (T)
Labels of the ith datapoint for all tasks. T for the number of tasks.
Tensor of dtype float32 and shape (T)
Binary masks of the ith datapoint indicating the existence of labels for all tasks.
str, optional
Id for the ith datapoint, returned only when ``self.load_full`` is True.
"""
if self.load_full:
return self.smiles[item], self.graphs[item], self.labels[item], \
self.mask[item], self.ids[item]
else:
return self.smiles[item], self.graphs[item], self.labels[item], self.mask[item]