Azimuth is a toolkit for reference-mapping, annotation, and interpretation of single-cell datasets. We are introducing Pan-Human Azimuth, a neural-network classifier for human single-cell and single-nucleus RNA-seq experiments that maps cells into a consistent hierarchical cell ontology across tissues and technologies.

Pan-Human Azimuth replaces the previous model of choosing separate Azimuth web applications for individual tissue or organ references. Instead, users can map data with a single pan-human, organ-spanning reference trained across 23 human tissues and 380 high-resolution cell types. The classifier returns hierarchical annotations, confidence scores, and Azimuth embeddings that can be used for downstream analysis and quality control.

Pan-Human Azimuth

Pan-Human Azimuth is currently available through a free command-line API/workflow and can be accessed from both R and Python:

• R users can run Pan-Human Azimuth directly on Seurat objects with the CloudAzimuth interface.
• Python users can run the panhumanpy package locally, either from the command line with annotate or interactively from Python on AnnData objects.
• For installation instructions, examples, and support links, see the Pan-Human Azimuth quick start guide.

Existing tissue-specific references

As part of this transition, we are no longer creating or updating tissue-specific Azimuth references, and we are no longer supporting the web applications for individual tissue-specific references.

Users can still map new datasets to the existing Azimuth references in R. The Seurat vignette on mapping and annotating query datasets describes this workflow.