roslu/BBBC021-Human-MCF7-Cells

---
license: cc-by-sa-3.0
category: biology
task_categories:
- image-classification
pretty_name: BBBC021 Human MCF7 Compound Cells
size_categories:
- 10K<n<100K
---

# BBBC021: Human MCF7 cells – compound-profiling experiment

[Link](https://bbbc.broadinstitute.org/BBBC021)

This dataset contains images of MCF-7 breast cancer cells treated with 113 small molecules across eight concentrations, labeled for DNA, F-actin, and B-tubulin.

## Citation and Copyright
As requested by the original authors:
> We used image set BBBC021v1 [Caie et al., Molecular Cancer Therapeutics, 2010], available from the Broad Bioimage Benchmark Collection [Ljosa et al., Nature Methods, 2012].

**Copyright:** The images and ground truth are copyright AstraZeneca Pharmaceuticals.

## Dataset Summary
- **Accession Number:** BBBC021 v1
- **Images:** 39,600 TIFF files (13,200 fields of view in 3 channels)
- **Format:** 55 ZIP archives (~750 MB each)
- **Biological Application:** Phenotypic profiling to predict the mechanisms of action (MoA) of various compounds.

## Metadata
The dataset also includes these three CSV files:
1. `BBBC021_v1_image.csv`: Image paths and treatment metadata.
2. `BBBC021_v1_compound.csv`: SMILES structures for most compounds.
3. `BBBC021_v1_moa.csv`: Mechanism of Action (MoA) ground truth for 103 compound-concentrations.

## Important Note on Evaluation
When evaluating MoA classification accuracy, it is critical to use **leave-one-compound-out** cross-validation. Ensure that no replicates or concentrations of the test compound are present in the training set.