gnn-datasets / README.pcqm4mv2-3d.md
alagesse's picture
Upload README.pcqm4mv2-3d.md with huggingface_hub
5c6a6e3 verified
metadata
language: []
pretty_name: PCQM4Mv2 3D
config_name: pcqm4mv2-3d
dataset_size: 562677714
size_categories: n>1M
license: mit
license_link: https://github.com/snap-stanford/ogb/blob/master/LICENSE
tags:
  - graphs
  - molecules
  - chemistry
  - parquet
  - torch-geometric
  - ogb
task_categories:
  - graph-regression
homepage: https://ogb.stanford.edu/docs/lsc/pcqm4mv2/

pcqm4mv2-3d

PCQM4Mv2 train split with 3D conformations aligned to OGB indices for HOMO-LUMO regression.

License: MIT License (OGB)

Splits

Split Rows File Size (MB)
train 3,195,733 train.parquet 562.68

Features

  • mol_id: int64 unique identifier per molecule
  • x: list[int64[9]], shape (num_nodes, 9), atom feature vector
  • edge_index: int64[2, num_edges], COO adjacency
  • edge_attr: list[int64[3]], shape (num_edges, 3), bond features
  • pos: list[float32[3]], shape (num_nodes, 3), 3D coordinates
  • num_nodes: int64 number of atoms
  • smiles: string canonical SMILES
  • target: float32 HOMO-LUMO gap

Citation

@article{hu2021ogblsc,
  title={OGB-LSC: A Large-Scale Challenge for Machine Learning on Graphs},
  author={Hu, Weihua and Fey, Matthias and Ren, Hongyu and Nakata, Maho and Dong, Yuxiao and Leskovec, Jure},
  journal={arXiv preprint arXiv:2103.09430},
  year={2021}
}