---
configs:
- config_name: processed_subset
  data_files:
  - split: train
    path: "processed_data_subset/train.csv"
  - split: validation
    path: "processed_data_subset/validation.csv"
  - split: test
    path: "processed_data_subset/test.csv"

license: unknown
task_categories:
- tabular-classification
task_ids:
- tabular-multi-class-classification

tags:
- healthcare
- breast-cancer
- gene-expression
- clinical-prediction
- uncertainty-quantification

pretty_name: Breast Cancer Gene Expression DMFS Dataset

size_categories:
- n<1K
---

# Breast Cancer Gene Expression DMFS Dataset

## Dataset Summary

This dataset is derived from public breast cancer gene-expression datasets distributed through Bioconductor and used in the `genefu` breast cancer analysis framework.

The original data and processing workflow are documented in the Bioconductor `genefu` vignette:
https://www.bioconductor.org/packages/devel/bioc/vignettes/genefu/inst/doc/genefu.html

The raw data are not distributed as standalone CSV files. Instead, they are programmatically accessed through Bioconductor experimental data packages and processed in R before being exported and further processed in Python.

This Hugging Face repository organizes the data into two levels:

1. **Intermediate Data**: CSV files exported from the initial R/Bioconductor processing step.
2. **Processed Data**: model-ready train, validation, and test splits generated in Python.
3. **Processed Data Subset**: subset of the processed rain, validation, and test splits generated in Python. These have a reduced feature space and are for display only.

The prediction task is binary classification of distant metastasis-free survival (DMFS) event status using gene-expression features.

---

## Source

This dataset is derived from publicly available breast cancer gene-expression cohorts distributed via Bioconductor and accessed using the `genefu` framework.

For full details on the original data sources and preprocessing pipeline, refer to the genefu website provided above.

The underlying cohorts include datasets such as:

- MAINZ
- TRANSBIG

These datasets are combined and processed using Bioconductor tools including `genefu` and `Biobase`.

Users are encouraged to consult the original source for detailed documentation of cohort composition, biological context, and preprocessing methodology.

---

## Initial R Processing (Bioconductor)

The intermediate data provided in this repository are generated using an R-based preprocessing pipeline.

Key steps include:

- Loading required libraries:
  - `genefu`
  - `Biobase`
  - Bioconductor breast cancer datasets
- Loading datasets:
  - `breastCancerMAINZ`
  - `breastCancerTRANSBIG`
- Removing phenotype columns in MAINZ that are entirely missing
- Combining datasets using `Biobase::combine`
- Extracting:
  - Phenotype data via `pData`
  - Gene-expression data via `exprs`
- Filtering samples with available DMFS information:
  - Non-missing `t.dmfs`
  - Non-missing `e.dmfs`
- Transposing the expression matrix so rows correspond to samples
- Constructing a cohort table with:
  - `sample_id`
  - `t_dmfs`
  - `e_dmfs`
  - `dmfs_label`

### Label Definition

The binary classification label is defined as:

- `dmfs_label = 1` if `e_dmfs == 1`
- `dmfs_label = 0` otherwise

---

## Repository Structure

```text
BC/
├── intermediate_data/
│   ├── cohort.csv
│   └── dataset.csv
│
├── processed_data/
│   ├── train.csv
│   ├── validation.csv
│   └── test.csv
│
├── processed_data_subset/
│   ├── train.csv
│   ├── validation.csv
│   └── test.csv
│
└── README.md