README.md

Virtual Cell Challenge

Overview

For Arc’s Virtual Cell Challenge, we generated a dedicated dataset measuring single-cell responses to perturbations in a human embryonic stem cell line (H1 hESC). This set of perturbations was carefully curated to span a broad range of phenotypic responses, and experimental parameters were optimized to maximize the reproducibility of observed effects. The Atlas includes the training, validation, and held-out test perturbation datasets leveraged throughout the Challenge.

• Format:
  • Count matrices: h5ad (AnnData)
  • Metadata: Parquet & CSV
• Data host:
  • Google Marketplace bucket: gs://arc-institute-virtual-cell-atlas/virtual-cell-challenge/
• Statistics
  • Cell count: ~300,000 cells
  • Target genes: 300

External Links

• Virtual Cell Challenge Datasets
• Behind the Data of the Virtual Cell Challenge (Arc Blog)

Data Structure

The data is organized by release year and split into training, validation, and test sets.

virtual-cell-challenge/
└── 2025/
    ├── test/
    ├── train/
    │   ├── adata_Training.h5ad
    │   └── pert_counts_Training.csv
    └── validation/

Metrics

The challenge evaluation framework centers around three metrics:

• Differential Expression Score (DES): Captures whether the model recovers the correct set of differentially expressed genes.
• Perturbation Discrimination Score (PDS): Measures whether the model assigns the correct effect to the correct perturbation.
• Mean Absolute Error (MAE): Assesses global expression accuracy across all genes.

Tutorials

• Python
• R

Contact

• GitHub Issues