Skip to main content

Robust Aitchison Tensor Decomposition for sparse count data

Project description

Build Status


Gemelli is a tool box for running both Robust Aitchison PCA (RPCA) and Compositional Tensor Factorization (CTF) on sparse compositional omics datasets.

RPCA can be used on cross-sectional datasets where each subject is sampled only once. CTF can be used on repeated-measure data where each subject is sampled multiple times (e.g. longitudinal sampling). Both methods are unsupervised and aim to describe sample/subject variation and the biological features that separate them.

The preprocessing transform for both RPCA and CTF is the robust centered log-ratio transform (rlcr) which accounts for sparse data (i.e. many missing/zero values). Details on the rclr can be found here and a interactive introduction into the transformation can be found here. In short, the rclr log transforms the observed (nonzero) values before centering. RPCA and CTF then perform a matrix or tensor factorization on only the observed values after rclr transformation, similar to Aitchison PCA performed on dense data.


To install the most up to date version of gemelli, run the following command

# pip (only supported for QIIME2 >= 2018.8)
pip install gemelli

Note: that gemelli is not compatible with python 2, and is compatible with Python 3.4 or later.


Gemelli can be run standalone or through QIIME2 and as a python API or CLI.

Cross-sectional study (i.e. one sample per subject) with RPCA

If you have a cross-sectional study design with only one sample per subject then RPCA is the appropriate method to use in gemelli. There are two commands within RPCA. The first is rpca and the second is auto-rpca. The only difference is that auto-rpca automatically estimates the underlying-rank of the matrix and requires no input for the n_components parameter. In the rpca command the n_components must be set explicitly. For examples of using RPCA we provide tutorials below exploring the microbiome between body sites.


Tutorials with QIIME2

Standalone tutorial outside of QIIME2

Repeated measures study (i.e. one sample per subject) with CTF


If you have a repeated measures study design with multiple samples per subject over time or space then CTF is the appropriate method to use in gemelli. For optimal results CTF requires samples for each subject in each time or space measurement. In some cases, this can require binning time my larger windows (e.g. instead of days use months). For examples of using CTF we provide a microbiome time series IBD study in the tutorials below.

Tutorials with QIIME2

Standalone tutorial outside of QIIME2


If you found this tool useful please cite the method(s) you used:

Citation for CTF

Martino, C. and Shenhav, L. et al. Context-aware dimensionality reduction deconvolutes gut microbial community dynamics. Nat. Biotechnol. (2020) doi:10.1038/s41587-020-0660-7
@article {Martino2020,
	author = {Martino, Cameron and Shenhav, Liat and Marotz, Clarisse A and Armstrong, George and McDonald, Daniel and V{\'a}zquez-Baeza, Yoshiki and Morton, James T and Jiang, Lingjing and Dominguez-Bello, Maria Gloria and Swafford, Austin D and Halperin, Eran and Knight, Rob},
	title = {Context-aware dimensionality reduction deconvolutes gut microbial community dynamics},
	year = {2020},
	journal = {Nature biotechnology},

Citation for RPCA

Martino, C. et al. A Novel Sparse Compositional Technique Reveals Microbial Perturbations. mSystems 4, (2019)
@article {Martino2019,
	author = {Martino, Cameron and Morton, James T. and Marotz, Clarisse A. and Thompson, Luke R. and Tripathi, Anupriya and Knight, Rob and Zengler, Karsten},
	editor = {Neufeld, Josh D.},
	title = {A Novel Sparse Compositional Technique Reveals Microbial Perturbations},
	volume = {4},
	number = {1},
	elocation-id = {e00016-19},
	year = {2019},
	doi = {10.1128/mSystems.00016-19},
	publisher = {American Society for Microbiology Journals},
	URL = {},
	eprint = {},
	journal = {mSystems}

Other Resources

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for gemelli, version 0.0.6
Filename, size File type Python version Upload date Hashes
Filename, size gemelli-0.0.6-py3-none-any.whl (55.9 kB) File type Wheel Python version py3 Upload date Hashes View
Filename, size gemelli-0.0.6.tar.gz (13.2 MB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring DigiCert DigiCert EV certificate Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page