C2S-Scale 27B (Cell2Sentence-Scale)

2025-10-15 · Publisher : Google · Domain : Artificial Intelligence

Google DeepMind and Yale release C2S-Scale 27B, a 27-billion-parameter foundation model for single-cell analysis, built on the open Gemma family, which proposed a therapeutic lead validated in the laboratory.

What changes

Foundation model for cell analysis at the single-cell scale, built on Gemma.
Task tested: finding a "conditional amplifier" drug for immunity; conditional reasoning capability described as emerging with scale.
Simulation of more than 4,000 drugs; prediction that silmitasertib (a CK2 kinase inhibitor) increases antigen presentation only in a favorable immune context.
Laboratory validation: about 50% increase in antigen presentation with silmitasertib and low-dose interferon.

### Specifications - Size: 27 billion parameters. - Availability: open model and resources for research (Hugging Face and GitHub), preprint on bioRxiv. ### Pricing Open model, with no usage pricing of its own. ### Benchmarks No standard benchmark: the post relies on experimental validation (about 50% increase in antigen presentation), a result published by the vendor. ### Key takeaways A scientific use case for the Gemma family, to be considered a preliminary research result and not a product.

C2S-Scale 27B (Cell2Sentence-Scale)

What changes

Links