metadata

dataset_info:
  features:
    - name: input_ids
      sequence: int32
    - name: coords
      sequence:
        sequence: float64
    - name: labels
      dtype: float64
    - name: token_type_ids
      sequence: int8
  splits:
    - name: train
      num_bytes: 788049618
      num_examples: 3507
    - name: val
      num_bytes: 90114995
      num_examples: 466
    - name: test
      num_bytes: 96803651
      num_examples: 490
  download_size: 472884474
  dataset_size: 974968264
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: val
        path: data/val-*
      - split: test
        path: data/test-*

LBA: Ligand Binding Affinity

Overview

In this task, we predict the binding affinity of ligands to their corresponding proteins based on the co-crystallized structure of the protein-ligand complex. We predict experimentally measured binding affinity as pK, defined as -log(Ki) or -log(Kd), depending on which measurement is available.

We derive crystal structures and ligand binding data from PDBBind (Wang et al., 2004), a widely-used curated database of protein-ligand complexes with experimental affinities derived from literature. We use the 2019 update of the so-called "refined set", a subset of complexes selected based on the quality of the structures and the affinity data. After filtering ligands which could not be read by RDKit due to invalid bonding data, our final dataset consists of 4,463 complexes.

Datasets

splits:
- split-by-sequence-identity-30: No proteins with seq. id. of more than 30% are in the same set.

Format

Each entry in the dataset contains the following keys:

['input_ids'] The set of atomic numbers for the protein, pocket and ligand atoms concatenated together.
['coords'] The 3D coordinates for the protein, pocket and ligand atoms concatenated together.
['label'] experimentally measured binding affinity as pK.
['token_type_ids'] A mask that corresponds to which inputs_ids/coords belong to the protein, pocket or ligand atoms (0 for protein , 1 for pocket, 2 for ligand)

Citation Information

@article{townshend2020atom3d,
  title={Atom3d: Tasks on molecules in three dimensions},
  author={Townshend, Raphael JL and V{\"o}gele, Martin and Suriana, Patricia and Derry, Alexander and Powers, Alexander and Laloudakis, Yianni and Balachandar, Sidhika and Jing, Bowen and Anderson, Brandon and Eismann, Stephan and others},
  journal={arXiv preprint arXiv:2012.04035},
  year={2020}
}

@article{wang2004pdbbind,
  title={The PDBbind database: Collection of binding affinities for protein- ligand complexes with known three-dimensional structures},
  author={Wang, Renxiao and Fang, Xueliang and Lu, Yipin and Wang, Shaomeng},
  journal={Journal of medicinal chemistry},
  volume={47},
  number={12},
  pages={2977--2980},
  year={2004},
  publisher={ACS Publications}
}

@article{liu2015pdb,
  title={PDB-wide collection of binding data: current status of the PDBbind database},
  author={Liu, Zhihai and Li, Yan and Han, Li and Li, Jie and Liu, Jie and Zhao, Zhixiong and Nie, Wei and Liu, Yuchen and Wang, Renxiao},
  journal={Bioinformatics},
  volume={31},
  number={3},
  pages={405--412},
  year={2015},
  publisher={Oxford University Press}
}