Xaira unveils X-Cell, a 4.9B-parameter "virtual cell" for drug discovery

On March 17, 2026, Xaira Therapeutics unveiled X-Cell, the company's first "virtual cell": a 4.9B-parameter model trained on X-Atlas/Pisces, the largest genome-wide cellular perturbation dataset ever assembled.

What is X-Cell

X-Cell is a model that predicts how gene expression changes when individual genes are knocked out one at a time. According to Xaira, it is the largest causal perturbation model to date. It scales according to a power law, in the same way that large language models improve as parameters and compute grow.

The training dataset was built from CRISPR experiments, with millions of tests run in parallel.

• X-Cell launch: March 17, 2026, the first virtual cell model from Xaira Therapeutics
• X-Atlas/Pisces dataset: 7 genome-wide Perturb-seq screens across 16 biological contexts, 25M cells after quality filtering
• Model size: 4.9B parameters, the largest causal perturbation model
• Scaling follows a power law, as with large language models
• On July 6, 2026, Bo Wang was promoted to Chief AI Scientist and Xi Chu to Chief Discovery Officer

Why causal models need causal data

Xaira argues that a model can only predict cause and effect if it is trained on causal data rather than descriptive genomics. Conventional datasets capture correlations, meaning which genes co-occur in a cell. Causal data shows what happens when you deliberately intervene on a specific gene.

"Causal models need causal data," is how

Bo Wang (Chief AI Scientist) and Xi Chu (Chief Discovery Officer) state the central thesis of their approach on the Latent Space podcast.

The company's practical takeaway: a model trained on causal data from cancer cells can be transferred to healthy donor cells, because the rules it has learned carry over.

How the model broke through the "data wall"

The first version of X-Cell hit a ceiling. At 3.1B parameters, test error stopped falling: the model saturated at roughly 1.5B parameters and did not improve beyond that. The cause was a shortage of informative data, not the architecture.

The X-Atlas/Pisces dataset delivered roughly a 30x increase in information density over the original approach. After that, the model began improving again with parameters and compute, and it beat a linear baseline that had previously outperformed earlier models. The foundation is scGPT, an earlier RNA model by Bo Wang published in Nature.

What this means

Xaira, an AI biotech with a billion dollars in starting capital, is betting on data rather than algorithms: its own factory of causal experiments matters more than someone else's off-the-shelf model. This reverses the usual AI logic of "internet data plus a big model". In biology, the data that matters is simply not available in the open, so it has to be generated from scratch.

Frequently asked questions

What is X-Cell in plain terms?

X-Cell is a "virtual cell", a 4.9B-parameter AI model that predicts how a cell responds when individual genes are knocked out, without running a physical experiment.

What is X-Cell trained on?

It is trained on the X-Atlas/Pisces dataset: 7 genome-wide Perturb-seq screens across 16 biological contexts, 25M cells after filtering. According to Xaira, it is the largest such dataset ever disclosed.

Who developed the model?

The work was led by Bo Wang (Chief AI Scientist and author of scGPT) and Xi Chu (Chief Discovery Officer). Both were promoted on July 6, 2026.