---
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: image
    dtype: image
  - name: class_label
    dtype: string
  splits:
  - name: train
    num_bytes: 26836881491
    num_examples: 19976
  - name: test
    num_bytes: 6710563933
    num_examples: 4995
  download_size: 33548877512
  dataset_size: 33547445424
---
# CellRep Evaluation Data - Broad Institute Method of Action Classification

Approximately 25k fluorescence microscopy images from the Broad Institute to use for testing CellRep vision models.

## Description

These images are drawn from the [CDRP-BBBC047-Bray](https://academic.oup.com/gigascience/article/6/12/giw014/2865213?login=true) cell painting dataset from the Broad Institute and have been converted into 3-channel. The full dataset be downloaded from the Broad Institute's [Cell Painting Gallery](https://broadinstitute.github.io/cellpainting-gallery/publications.html). The images are five-channel microscopy images of cancer cells (the U2OS osteosarcoma cell line). The [cell painting assay](https://www.nature.com/articles/nprot.2016.105) used in these datasets captures distinct cellular components through the following channels:

1. DNA/nucleus (Hoechst 33342)
2. RNA/nucleoli and cytoplasmic RNA (SYTO 14)
3. ER/endoplasmic reticulum (concanavalin A)
4. AGP/actin, Golgi and plasma membrane (phalloidin and WGA)
5. Mito/mitochondria (MitoTracker Deep Red)

## Evaluation Task

This held-out subset of images from CDRP-BBBC047-Bray was created from cells treated with known bioactive compounds. Each compound in this dataset has been labelled with the compound's known Method of Action (MOA) and the task for any model is to predict the MoA based using the image as an input.

As multiple compounds can share the same MoA, this helps test models to see if they learn biologically meaningful features within the same assay rather than memorizing compound-specific artifacts or batch effects. To ensure statistical reliability, we restrict evaluation to the 23 most frequent MoA classes in CDRP-bio-BBBC036-Bray.

| Class Label | Count |
| ----------- | ----- |
| adrenergic-receptor-antagonist | 2,346 |
| dopamine-receptor-antagonist | 2,178 |
| cyclooxygenase-inhibitor | 1,752 |
| serotonin-receptor-antagonist | 1,722 |
| acetylcholine-receptor-antagonist | 1,613 |
| adrenergic-receptor-agonist | 1,560 |
| histamine-receptor-antagonist | 1,356 |
| serotonin-receptor-agonist | 1,356 |
| glutamate-receptor-antagonist | 1,242 |
| glucocorticoid-receptor-agonist | 1,104 |
| dopamine-receptor-agonist | 1,032 |
| phosphodiesterase-inhibitor | 1,014 |
| sodium-channel-blocker | 930 |
| calcium-channel-blocker | 876 |
| bacterial-cell-wall-synthesis-inhibitor | 738 |
| egfr-inhibitor | 660 |
| adenosine-receptor-antagonist | 660 |
| estrogen-receptor-agonist | 612 |
| cannabinoid-receptor-agonist | 510 |
| atpase-inhibitor | 504 |
| cytochrome-p450-inhibitor | 468 |
| cdk-inhibitor | 456 |
| tyrosine-kinase-inhibitor | 282 |

## Splits

The images were spit into test and train subsets using stratified sampling.

| Split | Number of Images |
| ----- | ---------------- |
| train | 19,976 |
| test | 4,995 |