Datasets:

AIDD-LiLab
/

Apo2Mol_Dataset

ArXiv:

License:

Dataset card Files Files and versions

xet

Community

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Please provide the following information so we can review your access request.

Apo2Mol Dataset

This repository hosts the dataset for Apo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models [Paper, Github].

Overview

Apo2Mol Dataset is a structure-based drug design (SBDD) resource developed for pocket-based 3D ligand generation under protein structural dynamics. It contains 24,601 paired apo-holo protein structures, along with their associated ligands and corresponding binding pockets, and includes a predefined train/validation/test split.

The dataset is organized using PLINDER Protein–Ligand Interaction (PLI) IDs, where each entry contains a consistent and well-aligned set of structural components required for apo-to-holo modeling.

Files in This Repository

This repository contains two files:

├── data_folder.tar.gz
├── split_druglike_dict.pkl

1. `data_folder.tar.gz`

This archive contains all structural data files.

After extraction:

tar -xzf data_folder.tar.gz

you will obtain a directory:

data_folder/
├── <PLINDER_PLI_ID_1>/
│   ├── receptor_holo_pocket10.pdb
│   ├── receptor_apo_pocket10.pdb
│   ├── ligand.sdf
│   ├── receptor_holo_prot.pdb
│   └── receptor_apo_prot.pdb
├── <PLINDER_PLI_ID_2>/
│   └── ...

Each subfolder is named by a PLINDER PLI ID and contains exactly five files:

File name	Description
`receptor_holo_prot.pdb`	Full holo protein structure
`receptor_holo_pocket10.pdb`	Binding pocket* extracted from the holo protein structure
`receptor_apo_prot.pdb`	Full apo protein structure
`receptor_apo_pocket10.pdb`	Binding pocket* extracted from the apo protein structure
`ligand.sdf`	Bound ligand structure

*For each complex, the holo pocket is defined as all residues within 10Å of any ligand atom, and the corresponding apo pocket is obtained by extracting the same residues from the unbound structure.

2. `split_druglike_dict.pkl`

This file defines the data split used in Apo2Mol.

It is a Python dict with three keys:

{
  "train": [...],
  "valid": [...],
  "test":  [...]
}

Each entry is a tuple with the following structure:

(
holo_pocket_path, apo_pocket_path, ligand_path,
holo_protein_path, apo_protein_path,
date, holo_apo_pocket_rmsd
)

Field Description

Index	Field	Description
0	`holo_pocket_path`	Path to holo pocket structure (`receptor_holo_pocket10.pdb`)
1	`apo_pocket_path`	Path to apo pocket structure (`receptor_apo_pocket10.pdb`)
2	`ligand_path`	Path to ligand file (`ligand.sdf`)
3	`holo_protein_path`	Path to full holo protein structure
4	`apo_protein_path`	Path to full apo protein structure
5	`date`	Deposition or release date
6	`holo_apo_pocket_rmsd`	RMSD between apo and holo pockets

Example entry:

(
 '3txj__1__1.A__1.K/receptor_holo_pocket10.pdb',
 '3txj__1__1.A__1.K/receptor_apo_pocket10.pdb',
 '3txj__1__1.A__1.K/1.K.sdf',
 '3txj__1__1.A__1.K/receptor_holo_prot.pdb',
 '3txj__1__1.A__1.K/receptor_apo_prot.pdb',
 '2011-09-23',
 1.3904783
)

All paths are relative to data_folder/.

Loading Example

import pickle

with open("split_druglike_dict.pkl", "rb") as f:
    splits = pickle.load(f)

train_set = splits["train"]

Citation

If you feel this dataset is useful in your research, please cite the following works:

Apo2Mol

@article{zheng2025apo2mol,
  title={Apo2Mol: 3D Molecule Generation via Dynamic Pocket-Aware Diffusion Models},
  author={Zheng, Xinzhe and Jiang, Shiyu and Seabra, Gustavo and Li, Chenglong and Li, Yanjun},
  journal={arXiv preprint arXiv:2511.14559},
  year={2025}
}

Acknowledgment

This dataset is built upon the PLINDER (Protein-Ligand Interactions Dataset and Evaluation Resource). We thank the PLINDER team for developing and providing the comprehensive protein-ligand interaction data and infrastructure. Users of this dataset should also acknowledge the original PLINDER work:

@inproceedings{durairaj24plinder,
  title={PLINDER: The protein-ligand interactions dataset and evaluation resource},
  author={Durairaj, Janani and Adeshina, Yusuf and Cao, Zhonglin and Zhang, Xuejin and Oleinikovas, Vladas and Duignan, Thomas and McClure, Zachary and Robin, Xavier and Rossi, Emanuele and Zhou, Guoqing and others},
  booktitle={ICML'24 Workshop ML for Life and Material Science: From Theory to Industry Applications},
  year={2024}
}

Downloads last month: 5