Nanofold Public: 10,000 Protein Chains for Folding Model Training

NanoFold Public is the public train/validation portion of the nanoFold protein-folding benchmark. It packages a compact, fixed, auditable subset of OpenProteinSet/OpenFold-derived protein structure training data for fast iteration on data-efficient folding models. The dataset contains 10,000 training chains and 1,000 public validation chains, with each row representing one protein chain.

Use Cases

• Training protein structure prediction models based on the derived protein structure training data.
• Benchmarking data-efficient folding models based on the fixed train/validation split.
• Iterating on model architectures quickly based on the compact, auditable data subset.

Strengths

• Contains 10,000 training chains and 1,000 validation chains, providing a substantial base for model development.
• Described as a compact, fixed, and auditable subset, which may aid in reproducibility.

Limitations

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

Provenance

Source

Derived from OpenProteinSet/OpenFold.

Collection Method

Processed .npz tensors unrolled into Hugging Face Dataset columns.

Freshness

Last updated 2026-05-01 04:11:19; freshness should be verified.