opticlust

Single cell clustering and recommendations at a glance. Identify which clustering resolution(s) fit your data within minutes.

Opticlust currently offers:

• Automated clustering (leiden/louvain) at various resolutions
• Automatic selection of significant resolutions
• Clustering recommendations based on intra- and intercluster metrics
• Visualization of clusters per resolution and their relative compositions
• Easy to use, yet highly customizable Python API
• Cluster recoloring for opticlust and UMAP visualization (see below)

Installation

PyPi

pip install opticlust

Conda

conda install -c bioconda opticlust

GitHub

git clone https://github.com/siebrenf/opticlust.git
pip install opticlust

Develop

git clone https://github.com/siebrenf/opticlust.git
conda env create -n opticlust -f opticlust/requirements.yaml
conda activate opticlust
pip install --editable ./opticlust --no-deps --ignore-installed

Tutorial output

Output of clustering_plot() and score_resolutions():

Output of clustree():

Output of sc.pl.umap()

Output of sc.pl.rank_genes_groups_heatmap() and sc.pl.rank_genes_groups_dotplot():

top_low recommended resolution:

top_medium recommended resolution:

top_high recommended resolution:

Advantages of opticlust

The UMAPs and cluster tree plot can be compared immediately due to the automatic renaming and recoloring of the clusters. Without renaming and recoloring, figures would have looked like this:

Output of clustree(rename_cluster=False) and sc.pl.umap():

Note how cluster 2 becomes cluster 3 at resolution 0.43. This makes it difficult to track how changes in resolution impacted the clustering.

Acknowledgements

This tool was inspired by:

• The original Clustree R package.
• This BioStars post by firestar.

How to cite

When using this software package, please cite the accompanied DOI under “Citation” at https://doi.org/10.5281/zenodo.14513541