Oriel — AI & technology worth noticing, every day.

nvidia/RFD4-Proteina

Dataset · https://huggingface.co/datasets/nvidia/RFD4-Proteina

The structures behind biomolecular design.Indexed molecular records, sampling metadata, and traceable source inventories. Project page · Model · Inventory · Training guide RFD4-Proteína training datasets The structure datasets used to train RFD4-Proteína. This repository holds every training dataset used by the current release recipes: the flow model (hero_pretrain, 2026-10-19) and the unified biomolecular autoencoder (2026-10-15). It covers experimental PDB structures and… See the full description on the dataset page: https://huggingface.co/datasets/nvidia/RFD4-Proteina.

First seen by Oriel on .

AI summary

The repository publishes the training data and provenance for RFD4-Proteína’s flow and autoencoder recipes, spanning 17 component directories and 2.13 TB of molecular structures and metadata.

  • Its 19 configurations cover experimental structures and interfaces, protein–ligand systems, glycans, predicted complexes, RNA, cyclic peptides, and small molecules.
  • Parquet tables hold sampling and quality metadata; indexed, independently compressed mmCIF records are stored in binary shards that can be downloaded by component and read with AtomWorks.
  • The repository mirrors the training store layout for local use, but Hugging Face ranged reads are not a drop-in backend; keep indices and shards from the same Hub revision.
  • Predicted structures are not experimental measurements, and the dataset does not audit overlap with evaluation targets or guarantee independent, diverse samples.
  • An OMol audit found that 126 of 128 path values did not match CIF blocks retrieved by lmdb_idx; identifiers and related metadata remain unreconciled.

#protein design #biomolecular datasets #mmcif #reproducibility

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Datasets: 30, latest observation ; +15 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Datasets: 77,539, latest observation ; +59,598 over observed period since

Aliases

SourceExternal IDURL
Hugging Face Datasetsnvidia/RFD4-Proteinahttps://huggingface.co/datasets/nvidia/RFD4-Proteina Read discussion