# Illumina says its AI model won Arc's virtual cell challenge

Published 2026-11-20 · AI.info Team · AI.info

> Illumina says its CellularAI predictions placed first among approximately 1,500 competitors in Arc Institute's 2026 Virtual Cell Challenge. The company also released an open-source version of the algorithm.

## A test on unseen cells

Illumina's CellularAI model placed first in Arc Institute's 2026 Virtual Cell Challenge by predicting how unfamiliar cell lines would respond to genetic changes, the company said on November 19. Illumina said the model's predictions came closest to experimental results among approximately 1,500 competitors.

Source [prnewswire.com](https://www.prnewswire.com/news-releases/illumina-wins-virtual-cell-challenge-powered-by-billion-cell-atlas-302946270.html)

The company's BioInsight AI Lab used data from its Billion Cell Atlas alongside the new algorithm. An open-source version of CellularAI was released the same day, giving researchers a view of the method behind the result.

The challenge asked teams to predict responses to gene knockdowns in cell lines they had not seen, using information about the cells before perturbation. Illumina's results were scored against hidden experimental data generated by Arc.

## How Arc built the test

Arc's challenge covered six cell lines chosen to reflect biological differences including tissue of origin, cell lineage and disease status. In the previous year's competition, entrants predicted the transcriptional effects of genetic perturbations in a single cellular context after receiving some perturbation profiles.

For this year's test, participants received transcriptional profiles from unperturbed cells across multiple cell lines. Arc used separate perturbation data as the reference for scoring predictions.

To create its test data, Arc used Perturb-seq, which pairs a targeted genetic perturbation with a measurement of the resulting transcriptional profile. In this case, the method used CRISPR interference to knock down a single gene; single-cell RNA sequencing captured both the perturbation's identity and the cell's response.

Arc said its experimental teams handled over 1 billion cells to produce deeply sequenced transcriptional profiles from over 30 million cells across the six cell lines. From the larger dataset, the institute selected perturbations with at least 80% median on-target gene expression reduction and subsampled to 400 cells per perturbation for the challenge.

## The atlas behind CellularAI

Illumina's Billion Cell Atlas captures how cells respond to genetic changes made using CRISPR across more than 200 disease-relevant cell lines. The company is generating the data with pharmaceutical partners.

In January, Illumina introduced the atlas as the first tranche of a program to build a 5 billion-cell atlas over three years. The alliance's founding participants were AstraZeneca, Merck and Eli Lilly and Company; Formation Bio later joined along with two other AI-driven drug discovery partners.

Illumina said in July that more than 350 million cells had been sequenced, generating upwards of six petabytes of genomic data. The company describes the atlas as the world's largest genome-wide genetic perturbation dataset.

## Research uses

Illumina says researchers will be able to use CellularAI to predict responses in cell types that have not yet been experimentally tested and then test hypotheses using CRISPR capture capabilities in its Single Cell Preparation workflow. The company identifies potential applications including validating targets from human genetics, modeling disease-relevant cell types and studying how drugs work.

The BioInsight AI Lab is run by Kyle Farh, Illumina's vice president of artificial intelligence. The winning team included Kishore Jaganathan, Eyal Ben-David, Atreyo Pal and Aniketh Janardhan Reddy.

## Sources

1. Original source · Press release [PR Newswire](https://www.prnewswire.com/news-releases/illumina-wins-virtual-cell-challenge-powered-by-billion-cell-atlas-302946270.html) prnewswire.com

Other pages read

2. [Arc Institute](https://arcinstitute.org/news/behind-the-data-virtual-cell-challenge-2026) arcinstitute.org
3. [Illumina · Press release details](https://www.illumina.com/company/news-center/press-releases/press-release-details.html?newsid=7401) illumina.com
4. [Illumina · page 2](https://www.illumina.com/company/news-center/press-releases/2026/27191.html) illumina.com

## Related stories

- [Google details AlphaGenome research uses after cell-specific model release](https://ai.info/news/google-deepmind-alphagenome-disease-research-tools)
- [Google DeepMind releases AlphaGenome 2 for cell-specific research](https://ai.info/news/google-deepmind-alphagenome-2-cell-specific-disease-research)
- [AWS adds model leaderboard to SageMaker JumpStart](https://ai.info/news/aws-sagemaker-jumpstart-model-leaderboard)
- [Microsoft plans licensed MAI model weights through Frontier Company deals](https://ai.info/news/microsoft-mai-licensed-weights-azure)
- [JetBrains test finds AI models repeatedly criticize Python](https://ai.info/news/jetbrains-ai-models-programming-language-bias-test)

## Explore

[More articles](https://ai.info/news)

---

Canonical: https://ai.info/news/illumina-cellularai-virtual-cell-challenge
