Multimodal learning enables chat-based exploration of single-cell data.

CellWhisperer enables chat-based exploration of single-cell RNA-sequencing data with zero-shot prediction accuracy demonstrated in human embryonic development.

• Why it matters: Interpreting large-scale single-cell sequencing data remains difficult, limiting biological discovery and understanding of complex cellular processes.
• What they did: The team developed CellWhisperer, an AI model that creates a multimodal embedding of transcriptomes and textual annotations from 1 million RNA profiles, using contrastive learning, and integrates it with a large language model for natural-language queries.
• The result: This tool allows users to interactively explore gene expression and predict cell types without prior training, facilitating biological discovery and enabling an interactive graphical and chat interface within the CELLxGENE browser.