Embeddings from language models are good learners for single-cell data analysis.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Foundation models (FMs) show promise for single-cell data analysis, but their application varies.
• Analyzing complex single-cell data requires advanced computational methods.

Purpose of the Study:

• Introduce scELMo, a novel method utilizing large language models (LLMs) for single-cell data analysis.
• Demonstrate scELMo's capability in various single-cell data analysis tasks without requiring new model training.

Main Methods:

• scELMo integrates embeddings from LLM-generated metadata descriptions with raw single-cell data.
• Employs a zero-shot learning framework for initial analysis and a fine-tuning framework for advanced tasks.
• Leverages LLMs to generate metadata descriptions and their corresponding embeddings.

Main Results:

• scELMo effectively performs cell clustering, batch effect correction, and cell-type annotation.
• The fine-tuning framework enables analysis of complex tasks like in silico treatment and perturbation modeling.
• Achieves these results without the need for training new models.

Conclusions:

• scELMo provides a versatile and efficient tool for single-cell data analysis by utilizing LLMs.
• The method offers a lighter structure with lower resource requirements, presenting a promising direction for future research.
• Facilitates advanced analyses such as perturbation modeling and treatment effect prediction.