saiden89/xiang2021-spcas9

🧬 Xiang 2021 SpCas9 On-Target Efficiency

Merged dataset from Luo 2020 and Kim 2019 for SpCas9 on-target efficiency prediction. Published in Nature Communications. Contains 30mer sequences with measured CRISPR-Cas9 cutting efficiency.

Dataset Details

• Task: On Target Efficiency
• Nuclease: SpCas9 UniProt CasPEDIA
• Organism: Human
• Cell Line: HEK293T
• Paper: https://doi.org/10.1038/s41467-021-23576-0

Usage

from datasets import load_dataset

# Standard train/validation/test mapping (default config)
ds = load_dataset("saiden89/xiang2021-spcas9", name="default")
train, validation, test = ds["train"], ds["validation"], ds["test"]

from datasets import load_dataset

# Nested CV folds with held-out test
ds = load_dataset("saiden89/xiang2021-spcas9", name="folds")
folds = [ds[f"fold_{i}"] for i in range(1, 6)]
test = ds["test"]

Data Splits

Split	Samples
fold_1	3,984
fold_2	3,983
fold_3	3,984
fold_4	3,984
fold_5	3,984
test	3,983
total	23,902

Splits & Folds

• Folds: fold_1, fold_2, fold_3, fold_4, fold_5
• Test split: test
• Viewer note: This dataset provides two configurations: a standard train/validation/test mapping ("default") and fold-based splits ("folds") for nested CV. In the Hugging Face viewer, configurations are labeled as subsets.
• Note: Partition 6 is the held-out test set and is always exposed as test. Use folds 1-5 for nested CV; never use test for CV.

Data Quality

• Efficiency bounds: 0.0 to 100.0
• Out of bounds: 79 / 23,902 (0.3305%)
• Efficiency range: -4.411764621734619 to 95.10356903076172

Split	Count	Out of bounds	Out of bounds %
fold_1	3,984	13	0.3263%
fold_2	3,983	10	0.2511%
fold_3	3,984	16	0.4016%
fold_4	3,984	19	0.4769%
fold_5	3,984	10	0.251%
test	3,983	11	0.2762%

Citation

@article{xiang2021enhancing,
  title={Enhancing CRISPR-Cas9 gRNA efficiency prediction by fusing
         embeddings of sequence and structure information},
  author={Xiang, Xi and Corsi, Giulia I and Anthon, Christian and
          Qu, Kunli and Pan, Xiaochen and Liang, Xue and Han, Pei and
          Dong, Zhanying and Liu, Lijin and Zhong, Jiayan and others},
  journal={Nature Communications},
  volume={12},
  number={1},
  pages={3253},
  year={2021},
  publisher={Nature Publishing Group}
}

License

This dataset is released under the CC-BY-4.0 license.