Datasets:

Xaira-Therapeutics
/

X-Atlas-Orion

license: cc-by-nc-sa-4.0
configs:
  - config_name: default
    data_files:
      - split: HCT116
        path: data/HCT116*.parquet
      - split: HEK293T
        path: data/HEK293T*.parquet
  - config_name: gene_metadata
    data_files: metadata/gene_metadata.parquet

X-Atlas/Orion

X-Atlas: Orion edition (X-Atlas/Orion) is a Perturb-seq atlas containing two genome-wide Fix-Cryopreserve-ScRNAseq (FiCS) Perturb-seq screens that target all human protein-coding genes (n = 18,903 genes). The dataset is comprised of eight million HCT116 and HEK293T cells, each deeply sequenced to a median of 16,000 unique molecular identifiers (UMIs) per cell. The median on-target knockdown efficiency is 75.4% in HCT116 cells and 51.5% in HEK293T cells, with a median of at least 140 cells per perturbation. Through the release of X-Atlas/Orion, we highlight the potential of FiCS Perturb-seq to address current scalability and variability challenges in data generation, advance foundation model development that incorporates gene-dosage effects, and accelerate biological discoveries.

Preprint: X-Atlas/Orion: Genome-wide Perturb-seq Datasets via a Scalable Fix-Cryopreserve Platform for Training Dose-Dependent Biological Foundation Models
Processed h5ads and other metadata: https://doi.org/10.25452/figshare.plus.29190726

Tutorial

from datasets import load_dataset

# load the entire dataset in streaming mode
ds = load_dataset("Xaira-Therapeutics/X-Atlas-Orion", streaming=True)
# load only hct116
hct116_ds = load_dataset("Xaira-Therapeutics/X-Atlas-Orion", streaming=True, split="HCT116")
# load only hek293t
hek293t_ds = load_dataset("Xaira-Therapeutics/X-Atlas-Orion", streaming=True, split="HEK293T")

Dataset

The dataset contains the following information:

name	description
`gene_token_id`	gene identifiers corresponding to genes with non-zero expression in each cell. to be used with `gene_expression`. `metadata/gene_metadata.parquet` contains the mapping from `gene_token_id` to Ensembl ID and official gene symbol
`gene_expression`	raw counts for genes with non-zero expression. to be used with `gene_token_id`
`cell_barcode`	10X-generated cell barcode. the suffix `-1` is replaced with `-<SAMPLE>`
`sample`	GEM batch
`num_features`	number of guides
`guide_target`	guide identity
`gene_target`	gene targeted by guide
`n_genes_by_counts`	number of genes with non-zero counts
`total_counts`	total UMIs
`total_counts_mt`	total UMIs from MT genes
`pct_counts_mt`	% UMIs from MT genes
`pass_guide_filter`	boolean if cells contains two guides from the same guide pair

Gene metadata

All samples were aligned to the 10x Genomics GRCh38 2024-A pre-built reference genome (human reference (GRCh38) - 2024-A). Official gene symbols and ensembl IDs were extracted from the genes.gtf file.

# load metadata containing mappings to gene tokens and names
gene_metadata = load_dataset("Xaira-Therapeutics/X-Atlas-Orion","gene_metadata")

name	description
`ensembl_id`	Ensembl ID
`gene_name`	official gene symbol
`gene_token_id`	gene identifiers corresponding to genes with non-zero expression in each cell. to be used with `gene_token_id` in the dataset

Citation

@article{huang2025xatlasorion,
  title={X-Atlas/Orion: Genome-wide Perturb-seq Datasets via a Scalable Fix-Cryopreserve Platform for Training Dose-Dependent Biological Foundation Models},
  author={Huang, Ann C and Hsieh, Tsung-Han S and Zhu, Jiang and Michuda, Jackson and Teng, Ashton and Kim, Soohong and Rumsey, Elizabeth M and Lam, Sharon K and Anigbogu, Ikenna and Wright, Philip and Ameen, Mohamed and You, Kwontae and Graves, Christopher J and Kim, Hyunsung John and Litterman, Adam J and Sit, Rene V  and Blocker, Alex and Chu, Ci},
  journal={bioRxiv},
  year={2025},
  url={https://www.biorxiv.org/content/10.1101/2025.06.11.659105v1}
}