DBCata: Adsorption Structures and Model Checkpoints for Catalyst Screening

4.3 GB of cleaned adsorption structures from CatHub data, used for training the DBCata model. The dataset includes model checkpoints, fine-tuning scripts, and results for out-of-distribution testing. It was authored by Songze Huo and last updated on May 25, 2026.

Use Cases

• Training machine learning models for catalyst property prediction based on equilibrium adsorption structures.
• Fine-tuning and testing the UMA model using provided checkpoints and datasets.
• Conducting out-of-distribution (OOD) tests for model generalization using the included results.
• Benchmarking new catalyst screening algorithms against the provided cleaned CatHub structures.

Strengths

• Dataset size is 4.3 GB, indicating a substantial collection of structures and model artifacts.
• Includes multiple file formats (MD, PY, CKPT, etc.) supporting reproducibility and different analysis workflows.
• Contains cleaned data from CatHub, likely improving data quality for model training.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment for specific tasks.

Provenance

Source

figshare

Collection Method

Cleaned from CatHub data.

Freshness

Last updated 2026-05-25 06:15:00; freshness should be verified.