azimuthpy 0.1.0

Python wrapper for Azimuth (Seurat-based single-cell mapping) using rpy2

azimuthpy

---

azimuthpy is a lightweight Python package that enables seamless integration with the Azimuth reference mapping framework from the Satija Lab, using rpy2.
It allows you to map single-cell RNA-seq datasets (in Python AnnData format) to Azimuth references without leaving Python.

• Supports direct AnnData ➔ Seurat conversion
• Maps queries using Azimuth::MapQuery()
• Pulls predictions and annotations back into AnnData.obs
• Fully in-memory, no saving intermediate files

---

Installation

You need to have R installed on your machine with the following R packages:

Install:

pip install azimuthpy

Or using Conda:

# takes 15+ min...
conda env create -f environment.yml
conda activate azimuthpy_conda

If r-azimuth is not available through Conda, install it manually inside R:

install.packages("remotes")
remotes::install_github("satijalab/azimuth")

How to Use

import scanpy as sc
from azimuthpy import annotate_adata_azimuth

Load example dataset

adata = sc.datasets.pbmc3k()
adata

Preprocess

sc.pp.highly_variable_genes(adata, n_top_genes=2000, subset=True, flavor="seurat_v3") # Optional

# Store raw counts in adata.layers["raw_counts"]
adata.layers["raw_counts"] = adata.X

Map to Azimuth reference

You can download references from https://azimuth.hubmapconsortium.org/

# e.g., from https://azimuth.hubmapconsortium.org/references/#Human%20-%20Lung%20v2%20%28HLCA%29
# -> Zenodo: https://zenodo.org/records/6342228
# wget -c https://zenodo.org/api/records/6342228/files-archive
# unzip ...

adata = annotate_adata_azimuth(adata, reference_path="path/to/folder/",
                                        use_layer="raw_counts",
                                        umap_name="ref.umap")

View predicted labels

adata.obs[['predicted.celltype.l1', 'prediction.score.celltype.l1']]

For a full tutorial, see tutorial/gettingStarted.ipynb.

For more info, take a look at the original azimuth R page.