Bases: BaseDataset
QM1B is a dataset containing 1 billion conformations for 1.09M small molecules generated using a custom
PySCF library that incorporates hardware acceleration via IPUs. The molecules contain 9-11 heavy atoms and are
subsampled from the Generated Data Bank (GDB). For each molecule, 1000 geometries are generated using RDKit.
Electronic properties for each conformation are then calculated using the density functional B3LYP
and the basis set STO-3G.
Usage:
from openqdc.datasets import QM1B
dataset = QM1B()
References
https://arxiv.org/pdf/2311.01135
https://github.com/graphcore-research/qm1b-dataset/
Source code in openqdc/datasets/potential/qm1b.py
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142 | class QM1B(BaseDataset):
"""
QM1B is a dataset containing 1 billion conformations for 1.09M small molecules generated using a custom
PySCF library that incorporates hardware acceleration via IPUs. The molecules contain 9-11 heavy atoms and are
subsampled from the Generated Data Bank (GDB). For each molecule, 1000 geometries are generated using RDKit.
Electronic properties for each conformation are then calculated using the density functional B3LYP
and the basis set STO-3G.
Usage:
```python
from openqdc.datasets import QM1B
dataset = QM1B()
```
References:
https://arxiv.org/pdf/2311.01135\n
https://github.com/graphcore-research/qm1b-dataset/
"""
__name__ = "qm1b"
__energy_methods__ = [PotentialMethod.B3LYP_STO3G]
__force_methods__ = []
energy_target_names = ["b3lyp/sto-3g"]
force_target_names = []
__energy_unit__ = "ev"
__distance_unit__ = "bohr"
__forces_unit__ = "ev/bohr"
__links__ = {
"qm1b_validation.parquet": "https://ndownloader.figshare.com/files/43005175",
**{f"part_{i:03d}.parquet": f"https://ndownloader.figshare.com/files/{FILE_NUM[i]}" for i in range(0, 256)},
}
@property
def root(self):
return p_join(get_local_cache(), "qm1b")
@property
def preprocess_path(self):
path = p_join(self.root, "preprocessed", self.__name__)
os.makedirs(path, exist_ok=True)
return path
def read_raw_entries(self):
filenames = list(map(lambda x: p_join(self.root, f"part_{x:03d}.parquet"), list(range(0, 256)))) + [
p_join(self.root, "qm1b_validation.parquet")
]
def read_entries_parallel(filename):
df = pd.read_parquet(filename)
def extract_parallel(df, i):
return extract_from_row(df.iloc[i])
fn = partial(extract_parallel, df)
list_of_idxs = list(range(len(df)))
results = dm.utils.parallelized(fn, list_of_idxs, scheduler="threads", progress=False)
return results
list_of_list = dm.utils.parallelized(read_entries_parallel, filenames, scheduler="processes", progress=True)
return [x for xs in list_of_list for x in xs]
|