Why in the News?

Google DeepMind and Google Research has unveiled Cell2Sentence-Scale 27B (C2S-Scale), an AI model based on the Gemma family, marking a major advance in scientific research.

About C2S-Scale:

• Overview: It is a large-language-model (LLM) foundation system created by Google Research, Google DeepMind, and Yale University, designed to interpret the language of cells by converting single-cell transcriptomic data into textual “cell sentences.”
• Foundation & Architecture: Built on the Gamma family of open models with 27 billion parameters, it is among the world’s largest LLMs for biological data analysis.
• Purpose: Bridges single-cell RNA sequencing (scRNA-seq) and natural-language reasoning, allowing biologists to query models conversationally and obtain mechanistic hypotheses instead of raw statistics.
• Experimental Validation: Predicted a CK2-inhibition (silmitasertib + interferon) pathway that increases MHC-I antigen presentation in “cold” tumours, subsequently validated in live-cell assays.

Key Features:

• Parameter Scale: ~27 B parameters showing clear scaling-law gains in biological task performance.
• Data Representation: Converts ranked gene-expression profiles into gene-name sequences, enabling LLMs to treat transcriptomes as text.
• Multimodal Training: Trained on 50 million + single-cell profiles (human + mouse) plus metadata and scientific literature, aligning molecular data with context.
• Functional Range: Performs cell-type identification, perturbation-response prediction, dataset summarisation, cluster captioning, and biological Q&A.
• Reasoning Capability: Generates new, testable hypotheses, extending AI use from pattern detection to biological inference.
• Open-Source Access: Model weights and code released via Hugging Face and partner labs for community replication and benchmarking.

Get an IAS/IPS ranker as your 1: 1 personal mentor for UPSC 2024

Attend Now