Google Unveils Cell2Sentence-Scale 27B AI Model for Cancer Drug Discovery

Google, in collaboration with Yale University, has introduced an advanced Artificial Intelligence model named Cell2Sentence-Scale 27B (C2S-Scale). The model achieved a rare milestone in AI research — proposing a potential drug combination for cancer detection that was previously unknown to human experts, and which proved effective in laboratory conditions.

About the C2S-Scale 27B Model

• C2S-Scale 27B is a 27-billion-parameter foundation model developed to understand the “language” of individual cells.
• It is built on Google’s Gemma framework, designed for large-scale biological and biomedical data analysis.
• The model represents a fusion of computational biology and generative AI, enabling it to analyze cellular-level interactions in unprecedented detail.

Objective and Functioning

• The model was tasked to find a drug that could boost immune signals only when low levels of interferon were present.
• Interferons are proteins that act as frontline defenders against infections and tumours.
• Low interferon levels are indicative of a situation where tumours evade immune detection, allowing them to grow undetected.

Data and Methodology

• The AI model was trained using two major datasets:
  1. Real-world patient samples showing tumour–immune interactions and low interferon signalling.
  2. Cell-line data without immune context, providing controlled biological comparisons.
• By learning across these datasets, the model identified complex cellular communication patterns invisible to traditional analysis.

Key Discovery: Silmitasertib

• The C2S-Scale 27B model pinpointed a chemical compound called silmitasertib, known for its anti-cancer potential.
• The AI revealed that silmitasertib enhanced immune response only when tumour activity was suspected, making it a context-sensitive immune booster.
• This insight marks a significant advancement in precision medicine, demonstrating AI’s potential in drug repurposing and novel drug design.

Significance and Future Implications

• This marks one of the first major instances of AI contributing directly to experimental drug discovery.
• It highlights the potential of large AI models in decoding biological systems, accelerating the development of personalized and targeted therapies.
• The study underscores how AI-driven biomedical research can complement human expertise to uncover hidden biological mechanisms.

(Source: TH)