Hosted by
Sponsored by
Virtual Cell Challenge
Join us in shaping the future of cellular biology through predictive modeling.
Solve One of Biology’s Biggest Challenges with AI
Understanding, predicting, and ultimately programming how cells respond to internal cues and external stimuli is a fundamental challenge in biology.
Advances in single-cell RNA-seq technologies now enable large-scale measurements of cellular responses to genetic and chemical perturbations, fueling this exciting era of predictive cellular modeling. Virtual Cell Challenge is a recurring, open, community-driven challenge aimed at evaluating and improving computational models that predict cellular responses to genetic or chemical perturbations.
In 2025, the challenge will focus on context generalization: participants must predict the effects of perturbation in a held out cell type—the H1 human embryonic stem cell line. Using new experimental data we have generated for the Challenge, you will build a model that predicts these effects and submit the results to the Challenge leaderboard.
The top three models will win prizes valued at $100,000, $50,000, and $25,000.
Join us in shaping the future of cellular biology through predictive modeling.
Challenge in Cell
Challenge in Cell ↗
Why Participate?
World-class Benchmark Datasets
Significant Scientific Impact
Recognition and Awards
Global Community
Getting Started
Challenge Timeline
The Virtual Cell Challenge will be an annual challenge, with new data added each year to help improve model performance. The inaugural challenge will run from June to November 2025.
About the Data
Arc Institute has developed a dedicated, deeply sequenced, high cell coverage dataset of single-cell genetic perturbations for the Virtual Cell Challenge.
Learn more about the datasets ↗Arc’s Perturbation Data
Training data
Validation set
Final test set
Public Training Data
scBaseCount ↗
Tahoe-100M ↗
Public Perturb-seq Datasets
Evaluation
During the Challenge, submitted predictions will be compared to known results from the validation data set to calculate a performance score based on three metrics. Participants’ scores will be shown on the live ranking leaderboard in real time as they are submitted.
We have designed three metrics to evaluate model performance—
- Differential expression score to measure differential gene expression prediction accuracy.
- Perturbation discrimination score to rank predicted perturbations according to their similarity to the true perturbational effect.
- Mean Absolute Error to capture overall predictive accuracy and capture performance across the entire gene expression profile.
We use a combined score (overall score) that appropriately weighs each component relative to a baseline based on the cell-mean model of the training dataset.
In October, we will release the final test set for final submissions to the challenge. Scoring against the final test set will not be revealed publicly until winners are announced.
Prizes
The top three models submitted to the challenge will win a combination of cash and cloud GPU credits. Winners will be determined based on a score that takes into account the Evaluation criteria described above. Final results will be announced in December 2025 after submission and evaluation of final scoring.
The top three scores will win prizes as follows, each divided equally between cash and cloud credits—
Virtual Cell Challenge
