Arc Institute

Arc’s Virtual Cell Initiative aims to build ML models that predict how diverse cell types and states respond to drugs and genetic perturbations, generating large-scale perturbational single-cell datasets as the training foundation for predictive modeling.

S tate is Arc’s first-generation virtual cell model designed to predict how cells respond to chemical and genetic perturbations, producing predicted changes in RNA expression; Arc states it is trained on both observational and perturbational single-cell data and is released for noncommercial use.

Arc Virtual Cell Atlas is a portal for computation-ready single-cell measurements intended to support virtual cell modeling, launched with a dataset starting at over 300 million cells, and expanded via partnerships and additional data releases.

Arc describes Cell_Eval as a comprehensive evaluation suite for virtual cell modeling intended to go beyond conventional metrics and include biologically relevant, interpretable metrics related to differential expression prediction and perturbation strength estimation.

S tack is a foundation-model approach that Arc presents as simulating cellular conditions using prompt engineering without fine-tuning, applying a cell-as-prompt framing for predictions using observational data.

Evo 2 is Arc’s biological foundation model intended to model and design genetic sequences across domains of life; Arc describes it as developed with collaboration including NVIDIA and distributed collaborators across Stanford, UC Berkeley, and UCSF.

Bridge recombinases are described by Arc as a programmable, RNA-guided system beyond CRISPR for precise insertion, inversion, and excision of DNA segments, positioned as an enabling technology for genome engineering and biological design.

Altitude Therapy in a Pill is described by Arc as a small-molecule drug intended to offer benefits analogous to low-oxygen/high-altitude environment, with potential indications spanning cancer, mitochondrial diseases, and diabetes.