C2S-Scale: Cancer Treatment by Google AI Model

Context

Google has developed a new AI model, C2S-Scale (Cell2Sentence-Scale 27B), which has identified a potential new path for cancer treatment by understanding sensitive biological signals at the cellular level.

Cancer Treatment by Google AI Model

• Google has developed the C2S-Scale 27B model, based on its Gemma model series, specifically designed to understand single-cell data as a 'language'.
• This AI model formulated a new biological hypothesis about how a specific drug combination could make tumors more visible to the immune system by increasing the cells' antigen count.
• This hypothesis was tested in vitro in human neuroendocrine cell models and found an increase of approximately 50%, indicating that the model is not limited to data analysis but is also capable of generating new biological insights.
• Google and Yale University are working to advance this research into clinical trials.

About the C2S-Scale Model

• C2S-Scale is a 27 billion parameter (27B) foundation model based on the Gemma model family.
• This model converts single-cell RNA sequencing (scRNA-seq) data into text-based sequences called 'cell sentences'; This allows AI models to understand a cell's gene expression profile by converting it into a string of words.
• This allows cellular information and scientific literature to be understood by combining it with natural language modeling techniques.
• The C2S-Scale model is available in small and large versions so researchers can customize it according to their resources and needs.

Uses

• Cancer immunotherapy: Making "cold" tumors (hidden from the immune system) more visible to the immune system
• Novel drug combination discovery: Testing AI-suggested combinations in the laboratory
• Personalized medicine: Designing treatment strategies based on a patient's cell profile
• Uses in other diseases: Predicting cell responses in autoimmune and neurodegenerative diseases
• Increasing drug development speed: Reducing the number of laboratory tests by using AI to screen potential candidates first