Apo2Mol: 24,601 Paired Apo-Holo Protein Structures for Ligand Generation

Apo2Mol Dataset is a structure-based drug design resource containing 24,601 paired apo and holo protein structures, along with their associated ligands and binding pockets. It was created by AIDD-LiLab and last updated on Hugging Face in December 2025. The dataset is designed for training and evaluating pocket-aware 3D molecule generation models.

Use Cases

• Training dynamic pocket-aware diffusion models for 3D ligand generation based on paired protein structures.
• Benchmarking structure-based drug design algorithms based on the predefined train/validation/test splits.
• Studying protein structural dynamics and ligand binding based on the apo-holo structure pairs.

Strengths

• Contains 24,601 paired apo-holo protein structures, a substantial collection for model training.
• Includes associated ligands and corresponding binding pockets, providing a complete context for generation tasks.
• Provides a predefined train/validation/test split, facilitating standardized benchmarking.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but file formats, data size, and license information are unknown.
• Data may reflect bias inherent to the source platforms and methods used for its creation.

Provenance

Source

AIDD-LiLab via Hugging Face.

Collection Method

Developed for the Apo2Mol research project on 3D molecule generation via dynamic pocket-aware diffusion models.

Time Range

null

Freshness

Last updated 2025-12-22 15:56:03; freshness should be verified.

Geography

null