Researchers from the CeMM Research Center for Molecular Medicine of the Austrian Academy of Sciences have introduced a powerful new tool called CellWhisperer that changes how scientists can explore complex single-cell data.

Single-cell sequencing has transformed the life sciences by allowing researchers to study gene expression in individual cells rather than in bulk populations. This level of detail reveals the true diversity of cell types in tissues and helps uncover how cells behave during development, disease, and treatment response. However, analyzing this kind of data is no small task. Each experiment can produce information on thousands of cells and tens of thousands of genes, creating an enormous and intricate dataset that can be difficult to interpret.

The research team designed CellWhisperer as an artificial intelligence (AI) model and software tool that connects the power of language with biological data. By combining more than one million RNA sequencing profiles with descriptive text about those samples, the model learns to understand relationships between genes, cell types, and their biological context. This approach is known as multimodal learning, meaning it integrates multiple types of data, in this case, numerical gene expression values and text-based biological descriptions.

What makes CellWhisperer especially exciting is how scientists can use it. Instead of writing code or running complex analyses, users can simply chat with the system. They can ask questions such as “What genes define this cell type?” or “Which pathways are active in early embryonic cells?” The AI interprets the question and provides answers in natural language, making advanced analysis feel intuitive and interactive.

The research team demonstrated that CellWhisperer can predict cell types and biological features even without being specifically trained for those categories, a concept known as zero-shot prediction. This means the model can recognize patterns in new datasets based on what it has already learned, offering a way to accelerate discovery without additional training time.

Overview of the CellWhisperer multimodal AI for natural-language analysis of transcriptome data

a, Conceptual outline of CellWhisperer training dataset generation (left), model training and inference (center) and applications in scRNA-seq data analysis (right). b, UMAP visualization of CellWhisperer embeddings for human transcriptomes from the GEO repository. Clusters were computed using the Leiden algorithm and cluster labels were generated by CellWhisperer. The CellWhisperer-annotated dataset is available for interactive analysis on the project website (https://cellwhisperer.bocklab.org/geo). c, CellWhisperer scores for the free-text query term ‘infection’ projected on the UMAP of transcriptome embeddings from b. d, Retrieval of sample metadata (here: GEO submission date) for transcriptomes selected by CellWhisperer-generated cluster labels.

To make the tool even more practical, the developers integrated CellWhisperer into the CELLxGENE browser, a popular platform for visualizing single-cell datasets. This allows users to not only ask questions about gene expression but also see the answers reflected directly in the graphical display. For instance, when exploring a tissue sample, a user can highlight a cluster of cells and immediately ask the AI what defines that group or what makes it different from another cluster.

By linking massive community datasets with conversational AI, CellWhisperer makes complex data exploration feel natural and approachable. It provides a bridge between biology and computation, allowing scientists to spend more time interpreting results and less time managing data.

Perhaps most importantly, the tool shows how combining AI and biology can lead to faster, more intuitive discoveries. The researchers demonstrated CellWhisperer’s capabilities through a meta-analysis of human embryonic development, where it helped identify cell types and developmental patterns using only gene expression data and text-based insights. This approach could eventually extend to other areas such as cancer research, immunology, or regenerative medicine, where understanding how single cells behave is key to progress.

By bringing natural-language interaction into single-cell data analysis, CellWhisperer helps researchers uncover biological meaning in ways that were previously too time-consuming or technically complex. It represents a promising direction for the next generation of bioinformatics tools — ones that can truly “listen” to what scientists need and respond intelligently in return.

Availability – The source code underlying this project is available on GitHub (https://github.com/epigen/cellwhisperer). The CellWhisperer-annotated dataset is available for interactive analysis on the project website (https://cellwhisperer.bocklab.org/geo)