Datasets:

AIBIC
/

MLOmics

	---
	license: cc-by-4.0
	task_categories:
	- tabular-classification
	- tabular-regression
	- other
	language:
	- en
	tags:
	- biology
	- cancer
	- genomics
	- multi-omics
	- tcga
	- benchmark
	pretty_name: MLOmics
	size_categories:
	- 1B<n<10B
	---

	# MLOmics: Cancer Multi-Omics Database for Machine Learning

	MLOmics is an open benchmark database for developing and evaluating machine learning methods on cancer multi-omics data. It covers 8,314 patient samples across 32 cancer types with four omics modalities: mRNA expression, miRNA expression, DNA methylation, and copy number variation (CNV). Unlike raw TCGA releases, MLOmics provides standardized preprocessing, unified feature scales, and benchmark-ready task organization for reproducible machine learning studies.

	- Paper: [Scientific Data, 2025](https://www.nature.com/articles/s41597-025-05235-x)
	- GitHub: [chenzRG/Cancer-Multi-Omics-Benchmark](https://github.com/chenzRG/Cancer-Multi-Omics-Benchmark)
	- Project page: [chenzrg.github.io/project/mlomics](https://chenzrg.github.io/project/mlomics)

	---

	## Dataset Overview

	\| Datasets \| Cancer types \| Patients \| Omics types \|
	\|:--------:\|:------------:\|:--------:\|:-----------:\|
	\| 20 \| 32 \| 8,314 \| 4 (mRNA, miRNA, Methy, CNV) \|

	MLOmics is organized into three benchmark tasks covering supervised learning, unsupervised learning, and missing-value imputation.

	### Tasks

	\| Category \| Datasets \|
	\|----------\|----------\|
	\| Pan-cancer classification (1) \| Pan-cancer \|
	\| Golden-standard subtype classification (5) \| GS-BRCA, GS-COAD, GS-GBM, GS-LGG, GS-OV \|
	\| Cancer subtype clustering (9) \| ACC, KIRP, KIRC, LIHC, LUAD, LUSC, PRAD, THCA, THYM \|
	\| Omics data imputation (5) \| Imp-BRCA, Imp-COAD, Imp-GBM, Imp-LGG, Imp-OV \|

	---

	## Download

	Option 1 (recommended): Clone directly from Hugging Face (requires git-lfs):

	```bash
	git lfs install
	git clone https://huggingface.co/datasets/AIBIC/MLOmics
	```

	Option 2: Use the download script from the GitHub repository:

	```bash
	./download.sh
	```

	---

	## Data Structure

	```
	Main_Dataset/
	├── Classification_datasets/
	│ ├── Pan-cancer/Original/ # Pan-cancer (8314 samples, 32 types)
	│ ├── GS-BRCA/ {Original, Aligned, Top}/
	│ ├── GS-COAD/ {Original, Aligned, Top}/
	│ ├── GS-GBM/ {Original, Aligned, Top}/
	│ ├── GS-LGG/ {Original, Aligned, Top}/
	│ └── GS-OV/ {Original, Aligned, Top}/
	├── Clustering_datasets/
	│ └── {ACC,KIRP,KIRC,LIHC,LUAD,LUSC,PRAD,THCA,THYM}/
	│ └── {Original, Aligned, Top}/
	└── Imputation_datasets/
	└── {Imp-BRCA,Imp-COAD,Imp-GBM,Imp-LGG,Imp-OV}/Top/
	```

	Each dataset folder contains CSV files named `{DATASET}_{OMICS}.csv` (e.g., `BRCA_mRNA.csv`). Rows are features (genes/probes), columns are patient samples. Values are z-score normalized. All omics modalities of the same dataset share identical sample ordering.

	---

	## Feature Scales

	Three preprocessed versions are provided for classification and clustering datasets:

	\| Scale \| Description \|
	\|-------\|-------------\|
	\| Original \| Full feature set (no filtering) \|
	\| Top \| ANOVA-selected top features — mRNA: 5000, miRNA: 200, Methy: 5000, CNV: 5000 \|
	\| Aligned \| Intersection of features shared across all sub-datasets in the same task group \|

	Imputation datasets provide the Top scale only.

	---

	## Feature Dimensions

	\| Dataset \| Scale \| mRNA \| miRNA \| Methy \| CNV \|
	\|---------\|-------\|------\|-------\|-------\|-----\|
	\| ACC \| Original \| 18034 \| 368 \| 18711 \| 19519 \|
	\| KIRP \| Original \| 17254 \| 375 \| 18715 \| 19532 \|
	\| KIRC \| Original \| 18464 \| 352 \| 19045 \| 19523 \|
	\| LIHC \| Original \| 17945 \| 435 \| 18714 \| 19523 \|
	\| LUAD \| Original \| 18303 \| 427 \| 19034 \| 19532 \|
	\| LUSC \| Original \| 18577 \| 423 \| 19025 \| 19543 \|
	\| PRAD \| Original \| 17948 \| 447 \| 19028 \| 19528 \|
	\| THCA \| Original \| 17261 \| 345 \| 19024 \| 19532 \|
	\| THYM \| Original \| 18341 \| 535 \| 18716 \| 19532 \|
	\| Clustering (all) \| Aligned \| 10452 \| 254 \| 10347 \| 10154 \|
	\| Clustering (all) \| Top \| 5000 \| 200 \| 5000 \| 5000 \|
	\| GS-BRCA \| Original \| 18206 \| 345 \| 19049 \| 19533 \|
	\| GS-COAD \| Original \| 17261 \| 375 \| 19023 \| 19545 \|
	\| GS-GBM \| Original \| 17539 \| 308 \| 19031 \| 19539 \|
	\| GS-LGG \| Original \| 18339 \| 321 \| 19017 \| 19528 \|
	\| GS-OV \| Original \| 17344 \| 244 \| 19031 \| 19528 \|
	\| Classification (GS) \| Aligned \| 11343 \| 286 \| 11189 \| 11203 \|
	\| Classification (GS) \| Top \| 5000 \| 200 \| 5000 \| 5000 \|
	\| Pan-cancer \| Original \| 3217 \| 383 \| 3139 \| 3105 \|
	\| Imputation (all) \| Top \| 5000 \| 200 \| 5000 \| 5000 \|

	---

	## Citation

	```bibtex
	@article{2025mlomics,
	title={MLOmics: Cancer Multi-Omics Database for Machine Learning},
	author={Yang, Ziwei and Kotoge, Rikuto and Piao, Xihao and Chen, Zheng and Zhu, Lingwei and Gao, Peng and Matsubara, Yasuko and Sakurai, Yasushi and Sun, Jimeng},
	journal={Scientific Data},
	volume={12},
	number={1},
	pages={1--9},
	year={2025},
	publisher={Nature Publishing Group}
	}
	```