azimuth-annotation

Celltype annotation workflow

Overview

The annotation workflow utilizes Azimuth to transfer celltype labels from an annotated reference to a query dataset. This CWL workflow wraps Azimuth and runs within a docker container containing all dependencies.

Requirements

Running the pipeline requires a CWL workflow execution engine and container runtime; we recommend Docker and the cwltool reference implementation. cwltool is written in Python and can be installed into a sufficiently recent Python environment with pip install cwltool. Afterward, clone this repository, check out a tag, and invoke the pipeline as::

cwltool pipeline.cwl --matrix EXPR_H5AD --reference REFERENCE --secondary-analysis-matrix SECONDARY_ANALYSIS_H5AD --assay ASSAY

The supported values for --reference are RK, LK, RL, LL, HT. These two character codes indicate the side the organ was derived from (if applicable) and the organ type (kidney, lung, or heart). If the value for --reference doesn't match one of the five options, the workflow will run without performing annotation. The supported values for --assay are included here.

Azimuth to Cell Ontology Mapping

The mapping from Azimuth cell types to Cell Ontology IDs is described in csv format in the data directory. Each Azimuth label for each organ maps to a corresponding Azimuth ID, CL ID, and standardized label. There is also associated metadata for each organ in all_metadata.json in the data directory. The JSON contains a key per organ. Within each organ's metadata there are the same three keys: versions, reviewers, and disclaimer. versions contains the three keys described below:

Key	Description
azimuth_reference	version of Azimuth. organ used. annotation_level indicates which Azimuth annotations were used for the mapping since there are usually multiple per reference. doi of the Azimuth reference.
CL_version	The version of Cell Ontology used in mapping.
mapping_version	The version of table itself, as it could hypothetically change independent of Azimuth and ASCTB versions.

Mapping tables are pulled into this workflow's Docker image, so for updates to be propogated the new Docker image will need to be pushed to Docker Hub.