Xaira Combines Anthropic Operon with X-Cell for Virtual Cell Discovery

7/ The loop: Operon builds the reward → X-Cell simulates and proposes → agent filters for druggability → wet lab validates → data flows back to improve X-Cell.

Less bespoke model-building, more orchestration of biology and chemistry we already have. And it compounds as Anthropic's models get sharper and Operon grows its exploratory space.

2/ The obvious move: scale X-Cell as more data fold in. That matters, but it's expensive and slow. With Operon, we can already deploy X-Cell for focused target discovery today, pointing Anthropic's life sciences agentic harness at a mountain of existing biomedical data and letting it build the connective tissue between X-Cell and a specific disease indication.

3/ Step one: Operon mines existing raw biological data for a disease of interest and builds a "disease program" : a scoring function that reads a cell's expression state and rates how disease-like it is. It holds out orthogonal data and validates on an independent cohort. A data-grounded reward model, not a hand-curated gene set or pathway.

6/ Operon also adds the constraint that actually matters for a therapeutic: can you design a specific, developable binder against this target, with the right cell-type selectivity? It filters the ranked hits down to what you can realistically build : the adjacent possible candidates.

4/ That reward model pulls double duty. First: it guides X-Cell, keeping perturbation predictions anchored to the biology that actually defines the disease , not just what's transcriptionally plausible.

5/ Second: it selects. X-Cell takes a candidate target, simulates the perturbation, and the disease program scores whether the simulated cell moves back toward a healthy state. Rank thousands of in silico interventions by how much they rescue the disease signature.

8/ We will share more soon on Operon-guided virtual-cell discovery with X-Cell. Preprint: http://biorxiv.org/content/10.64898/2026.03.18.712807v1

@BoWang87 Yes🙌