Arc State
A virtual cell model predicting perturbation responses
What's new
Arc's first virtual cell model: trained on 167M observational and 100M perturbed cells; predicts transcriptomic responses to genetic and chemical perturbations.
State is Arc Institute's first virtual cell model. It operates on sets of cells to capture both population- and cell-level perturbation effects.
Components
- State Embedding (SE) — representations learned from 167M observational cells.
- State Transition (ST) — perturbation effects learned from 100M+ perturbed cells across 70 contexts.
Released for non-commercial use, with code and checkpoints on GitHub and Hugging Face. It is one of the first models to consistently beat simple linear baselines on perturbation prediction.
Related tools
Single-Cell & Spatial
scvi-tools
Probabilistic, deep-learning models for single-cell omics
AI Models
Evo 2
A genome-scale DNA foundation model across all domains of life
Public Databases
CZ CELLxGENE Census
Programmatic access to tens of millions of single cells