Learn PLN

resources to learn methodological aspects as well as how to use PLN in practice

This post presents some resources to learn more about Poisson Log Normal models and its implementations.

Publications

The three seminal articles are the following ones :

• PLN for dimension reduction and visualisation in probabilistic Poisson PCA ,
• PLN with sparsity constraint on thelatent precision matrix for network inference from count data,
• Description of the generic framework with PLN equivalents for discriminant analysis, mixture model, on top of the PCA and network variants.

See also Aitchison et Ho (1989) for previous work on the Multivariate Poisson Log Normal distribution.

Vignettes and notebooks

The book “Statistical Approaches for Hidden Variables in Ecology” contains a chapter entitled “The Poisson Log-Normal Model: A Generic Framework for Analyzing Joint Abundance Distributions” . The book is available [in English](https://onlinelibrary.wiley.com/doi/book/10.1002/9781119902799) and in French and all the chapters have a corresponding notebook reproducing the analysis. The french version of the vignette dedicated to PLN model is [here](https://oliviergimenez.github.io/code_livre_variables_cachees/chiquet.html).

R {PLNmodels} package

The R package {PLNmodels} website proposes different vignettes from data importation to the use of different models.

1. Description of the Trichoptera data set
   The Trichoptera data set is included in the R package and used as example in different vignettes.

2. Data importation in {PLNmodels}
3. Analyzing multivariate count data with the Poisson log-normal model
4. Dimension reduction of multivariate count data with PLN-PCA
5. Sparse structure estimation for multivariate count data with PLN-network
6. Supervized classification of multivariate count table with the Poisson discriminant Analysis
7. Clustering of multivariate count data with PLN-mixture

How to cite ?

From R, you can see the references using the command r citation("PLNmodels").

Python {pyPLNmodels} package

If you prefer Python, use the {pyPLNmodels} package. This package implements efficient algorithms for PLN or ZIPLN models as well as PLN-PCA. It has been built to scale on large datasets even though it has memory limitations.
A notebook to get started is available here.

Slides

Slideshow

References