State is an open-source machine learning model from the Arc Institute that predicts how cells respond to perturbations across diverse biological contexts. It is part of the push toward virtual cell models that can forecast experimental results before a single lab test is run. The repository has about 731 stars on GitHub and was updated in 2026.
Drug discovery depends on knowing how a cell’s gene expression changes when it is exposed to a compound, a cytokine or a genetic edit. Measuring every combination in the lab is impossibly slow, so Arc Institute trained State on large-scale single-cell perturbation data to predict those responses in silico. Given a cell’s baseline state and a perturbation, the model forecasts the resulting expression profile, and it is designed to generalize across cell types and contexts it has not seen for that exact perturbation. The code and model are released openly so the research community can benchmark, extend and build on them.
The open release is aimed at reproducibility: other groups can evaluate State on their own perturbation datasets, compare it with in-house approaches, and use it as a baseline while the virtual cell field develops shared benchmarks.
State is free and open source. Researchers can download the model and code from GitHub and run them on their own compute; there is no hosted paid tier, license fee or usage charge from the Arc Institute.
State is for computational biologists, bioinformaticians and drug discovery teams who run single-cell perturbation experiments. It helps them prioritize which compounds or edits to test in the wet lab, and it gives ML researchers a strong open baseline for virtual cell modeling.