Skip to content

DATA_README.md

Latest commit

 

History

History
176 lines (122 loc) · 5.43 KB

DATA_README.md

File metadata and controls

176 lines (122 loc) · 5.43 KB
Raw

MLOmics Dataset — Download & Setup Instructions

⚠️ Disclaimer: This data is used for academic research only. Not for clinical use. This is an academic research prototype. It should not be used for any clinical decision-making.

1. Data Source

Property Detail
Dataset MLOmics: Cancer Multi-Omics Benchmark
Primary Source Figshare
Alternative Source HuggingFace
License CC-BY-4.0
Original TCGA Data Source data originates from The Cancer Genome Atlas (TCGA)

2. Citation

If you use this dataset, please cite:

Chen, Z. et al. "MLOmics: Cancer Multi-Omics Database for Machine Learning Research." Figshare. Dataset. https://doi.org/10.6084/m9.figshare.28729127

3. What to Download

This project uses Top (ANOVA-selected) features from the Classification datasets for two cancer types: GS-BRCA and GS-COAD.

GS-BRCA (Breast Cancer) — Top Features

File Description Size
BRCA_mRNA_top.csv mRNA expression (5,000 features × 671 samples) ~61 MB
BRCA_miRNA_top.csv miRNA expression (366 features × 671 samples) ~2 MB
BRCA_Methy_top.csv DNA Methylation (5,000 features × 671 samples) ~62 MB
BRCA_CNV_top.csv Copy Number Variation (5,000 features × 671 samples) ~61 MB
54814634_BRCA_label_num.csv Subtype labels (671 samples, 5 subtypes) ~1 KB

GS-COAD (Colon Cancer) — Top Features

File Description Size
COAD_mRNA_top.csv mRNA expression (5,000 features × 260 samples) ~24 MB
COAD_miRNA_top.csv miRNA expression (200 features × 260 samples) ~1 MB
COAD_Methy_top.csv DNA Methylation (5,000 features × 260 samples) ~24 MB
COAD_CNV_top.csv Copy Number Variation (5,000 features × 260 samples) ~24 MB
54814562_COAD_label_num.csv Subtype labels (260 samples, 4 subtypes) ~1 KB

Total download size: ~260 MB (10 files)

4. Where to Place Files

Place downloaded files in these directories (relative to project root):

data/raw/GS-BRCA/Top/
├── BRCA_mRNA_top.csv
├── BRCA_miRNA_top.csv
├── BRCA_Methy_top.csv
├── BRCA_CNV_top.csv
└── 54814634_BRCA_label_num.csv

data/raw/GS-COAD/Top/
├── COAD_mRNA_top.csv
├── COAD_miRNA_top.csv
├── COAD_Methy_top.csv
├── COAD_CNV_top.csv
└── 54814562_COAD_label_num.csv

5. Dataset Statistics (Verified from Actual Files)

Sample Counts

Cancer Samples mRNA Features miRNA Features Methy Features CNV Features Subtypes
GS-BRCA 671 5,000 366 5,000 5,000 5
GS-COAD 260 5,000 200 5,000 5,000 4

Subtype Label Distributions

GS-BRCA (5 subtypes):

Label Count Percentage
0 353 52.6%
1 42 6.3%
2 132 19.7%
3 31 4.6%
4 113 16.8%

GS-COAD (4 subtypes):

Label Count Percentage
0 174 66.9%
1 48 18.5%
2 34 13.1%
3 4 1.5%

Note: Both datasets exhibit class imbalance. GS-COAD label 3 has only 4 samples — this may impact model performance for that subtype.

6. Important Data Format Notes

CSV Layout (CRITICAL)

MLOmics CSV files are transposed relative to standard ML format:

  • Rows = features (genes / CpG sites / CNV regions)
  • Columns = samples (patients)
  • First column = feature identifiers
  • Header row = sample IDs

You MUST transpose the data (to samples × features) before using it for model input.

Label File Format

  • Single column named Label
  • Integer values starting from 0
  • Row order corresponds to sample order in modality files

Feature Scale: Top

Top features are pre-selected via global ANOVA statistical testing (p < 0.05), selecting the most significant features. This is a known limitation — ANOVA was applied globally rather than within each cross-validation fold. Document this in any reports.

7. Checksums

After downloading, verify file integrity using checksums. Generate with:

# Windows (PowerShell)
Get-FileHash data/raw/GS-BRCA/Top/* -Algorithm SHA256 | Format-Table Hash, Path > data/checksums_brca.txt
Get-FileHash data/raw/GS-COAD/Top/* -Algorithm SHA256 | Format-Table Hash, Path > data/checksums_coad.txt

Reference checksum files (if generated): data/checksums_brca.txt, data/checksums_coad.txt

8. Data Redistribution Policy

  • Do NOT redistribute the raw CSV files
  • Instead, share this README with download instructions and checksums
  • Original data is from TCGA and released under CC-BY-4.0 via MLOmics
  • Always cite the MLOmics dataset (see Section 2)

9. Related Resources

Resource URL
MLOmics GitHub (benchmark repo) https://github.com/chenzRG/Cancer-Multi-Omics-Benchmark
MLOmics Paper See Figshare page for associated publication
STRING Mapping File Cancer-Multi-Omics-Benchmark/Complementary Resources/Mapping File/
TCGA Data Portal https://portal.gdc.cancer.gov/

Last updated: February 2026 Project: MLOmics — Latent-Fusion Multi-Omics Classifier for Cancer Subtype Prediction