AmbiDose

AmbiDose removes ambient RNA from droplet scRNA-seq. Empty droplets estimate the soup profile χ. Each cell receives an operational dose d_c = ρ_c · n_c. Type-aware subtraction writes non-negative integer counts. The method does not correct batch effects or assign cell types.

Docs: Read the Docs.

Install

pip install ambidose
# or: conda install -c conda-forge -c bioconda ambidose

Python 3.10–3.13. A GPU is not required. Details: installation.

First run

import scanpy as sc
import ambidose as amdose

adata = sc.read_10x_mtx("filtered_feature_bc_matrix/")
adata = amdose.denoise(adata, raw="raw_feature_bc_matrix.h5", sample_key=None)
# adata.X is denoised integer UMI; original counts in layers["raw_counts"]

ambidose denoise --input /path/to/outs --output cleaned.h5ad

Treat rho as an operational dose. Check obs["ambidose_rho_trust"] and the QC report before using it as a contamination rate.

Status

0.5.13 (Alpha). Import as import ambidose as amdose. The entry point is denoise(). See the API reference.

Next steps

1. Installation
2. Quickstart
3. Tutorials
4. FAQ
5. From R

Citation

If you use AmbiDose in published work, please cite the preprint:

Zhao Li, Aaron W. James, Shengxuan Li.
AmbiDose: per-cell removal budgets for ambient RNA contamination.
bioRxiv (2026). https://doi.org/10.64898/2026.09.27.754849

@article{li2026ambidose,
  title   = {AmbiDose: per-cell removal budgets for ambient RNA contamination},
  author  = {Li, Zhao and James, Aaron W. and Li, Shengxuan},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.64898/2026.09.27.754849},
  url     = {https://www.biorxiv.org/content/10.64898/2026.09.27.754849v1},
}

For the software record, cite the Zenodo DOI. Pin the package version used in the analysis, for example ambidose==0.5.13. See CITATION.cff and the citation page.

License

Software: Apache License 2.0.

Author

Zhao Li (李钊) Email: leelieber@gmail.com